diff --git a/lib/crewai/tests/agents/test_agent_executor.py b/lib/crewai/tests/agents/test_agent_executor.py index 4995564bf..d0243350a 100644 --- a/lib/crewai/tests/agents/test_agent_executor.py +++ b/lib/crewai/tests/agents/test_agent_executor.py @@ -721,3 +721,116 @@ class TestAgentExecutorPlanning: # Plan should exist assert captured_plan[0] is not None + + +class TestResponseFormatWithKickoff: + """Test that Agent.kickoff(response_format=MyModel) returns structured output. + + Real LLM calls via VCR cassettes. Tests both with and without planning, + using real tools for the planning case to exercise the full Plan-and-Execute + path including synthesis with response_model. + """ + + @pytest.mark.vcr() + def test_kickoff_response_format_without_planning(self): + """Test that kickoff(response_format) returns structured output without planning.""" + from pydantic import BaseModel, Field + from crewai import Agent + from crewai.llm import LLM + + class MathResult(BaseModel): + answer: int = Field(description="The numeric answer") + explanation: str = Field(description="Brief explanation of the solution") + + llm = LLM("gpt-4o-mini") + + agent = Agent( + role="Math Assistant", + goal="Solve math problems and return structured results", + backstory="A precise math assistant that always returns structured data", + llm=llm, + verbose=False, + ) + + result = agent.kickoff("What is 15 + 27?", response_format=MathResult) + + assert result is not None + assert result.pydantic is not None + assert isinstance(result.pydantic, MathResult) + assert result.pydantic.answer == 42 + assert len(result.pydantic.explanation) > 0 + + @pytest.mark.vcr() + def test_kickoff_response_format_with_planning_and_tools(self): + """Test response_format with planning + tools (multi-step research). + + This is the key test for _synthesize_final_answer_from_todos: + 1. Planning generates steps that use the EXA search tool + 2. StepExecutor runs each step in isolation with tool calls + 3. The synthesis step produces a structured BaseModel output + + The response_format should be respected by the synthesis LLM call, + NOT by intermediate step executions. + """ + from pydantic import BaseModel, Field + from crewai import Agent, PlanningConfig + from crewai.llm import LLM + from crewai_tools import EXASearchTool + + class ResearchSummary(BaseModel): + topic: str = Field(description="The research topic") + key_findings: list[str] = Field(description="List of 3-5 key findings") + conclusion: str = Field(description="A brief conclusion paragraph") + + llm = LLM("gpt-4o-mini") + exa = EXASearchTool() + + agent = Agent( + role="Research Analyst", + goal="Research topics using search tools and produce structured summaries", + backstory=( + "You are a research analyst who searches the web for information, " + "identifies key findings, and produces structured research summaries." + ), + llm=llm, + planning_config=PlanningConfig(max_attempts=1, max_steps=5), + tools=[exa], + verbose=False, + ) + + result = agent.kickoff( + "Research the current state of autonomous AI agents in 2025. " + "Search for recent developments, then summarize the key findings.", + response_format=ResearchSummary, + ) + + assert result is not None + # The synthesis step should have produced structured output + assert result.pydantic is not None + assert isinstance(result.pydantic, ResearchSummary) + # Verify the structured fields are populated + assert len(result.pydantic.topic) > 0 + assert len(result.pydantic.key_findings) >= 1 + assert len(result.pydantic.conclusion) > 0 + + @pytest.mark.vcr() + def test_kickoff_no_response_format_returns_raw_text(self): + """Test that kickoff without response_format returns plain text.""" + from crewai import Agent + from crewai.llm import LLM + + llm = LLM("gpt-4o-mini") + + agent = Agent( + role="Math Assistant", + goal="Solve math problems", + backstory="A helpful math assistant", + llm=llm, + verbose=False, + ) + + result = agent.kickoff("What is 10 + 10?") + + assert result is not None + assert result.pydantic is None + assert "20" in str(result) diff --git a/lib/crewai/tests/cassettes/agents/TestResponseFormatWithKickoff.test_kickoff_no_response_format_returns_raw_text.yaml b/lib/crewai/tests/cassettes/agents/TestResponseFormatWithKickoff.test_kickoff_no_response_format_returns_raw_text.yaml new file mode 100644 index 000000000..e8834d7ac --- /dev/null +++ b/lib/crewai/tests/cassettes/agents/TestResponseFormatWithKickoff.test_kickoff_no_response_format_returns_raw_text.yaml @@ -0,0 +1,214 @@ +interactions: +- request: + body: '{"messages":[{"role":"system","content":"You are Math Assistant. A helpful + math assistant\nYour personal goal is: Solve math problems"},{"role":"user","content":"\nCurrent + Task: What is 10 + 10?\n\nProvide your complete response:"}],"model":"gpt-4o-mini"}' + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '255' + content-type: + - application/json + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NfywWXSxBmPTEjnpyaIWqF80Bz8\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413350,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"10 + 10 equals 20.\",\n \"refusal\": + null,\n \"annotations\": []\n },\n \"logprobs\": null,\n + \ \"finish_reason\": \"stop\"\n }\n ],\n \"usage\": {\n \"prompt_tokens\": + 46,\n \"completion_tokens\": 8,\n \"total_tokens\": 54,\n \"prompt_tokens_details\": + {\n \"cached_tokens\": 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:29:10 GMT + Server: + - cloudflare + Set-Cookie: + - SET-COOKIE-XXX + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '288' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: '{"messages":[{"role":"system","content":"You are Math Assistant. A helpful + math assistant\nYour personal goal is: Solve math problems"},{"role":"user","content":"\nCurrent + Task: What is 10 + 10?\n\nProvide your complete response:"}],"model":"gpt-4o-mini"}' + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '255' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NfyZWuh1ZpzpgkqSDcyZKEhjYnm\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413350,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"10 + 10 equals 20.\",\n \"refusal\": + null,\n \"annotations\": []\n },\n \"logprobs\": null,\n + \ \"finish_reason\": \"stop\"\n }\n ],\n \"usage\": {\n \"prompt_tokens\": + 46,\n \"completion_tokens\": 8,\n \"total_tokens\": 54,\n \"prompt_tokens_details\": + {\n \"cached_tokens\": 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:29:11 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '323' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +version: 1 diff --git a/lib/crewai/tests/cassettes/agents/TestResponseFormatWithKickoff.test_kickoff_response_format_with_planning_and_tools.yaml b/lib/crewai/tests/cassettes/agents/TestResponseFormatWithKickoff.test_kickoff_response_format_with_planning_and_tools.yaml new file mode 100644 index 000000000..910311097 --- /dev/null +++ b/lib/crewai/tests/cassettes/agents/TestResponseFormatWithKickoff.test_kickoff_response_format_with_planning_and_tools.yaml @@ -0,0 +1,8339 @@ +interactions: +- request: + body: '{"trace_id": "d9fbeecd-afdc-48fa-817c-fe3612c4df17", "execution_type": + "crew", "user_identifier": null, "execution_context": {"crew_fingerprint": null, + "crew_name": "Unknown Crew", "flow_name": null, "crewai_version": "1.9.3", "privacy_level": + "standard"}, "execution_metadata": {"expected_duration_estimate": 300, "agent_count": + 0, "task_count": 0, "flow_method_count": 0, "execution_started_at": "2026-02-06T21:27:29.538489+00:00"}}' + headers: + Accept: + - '*/*' + Connection: + - keep-alive + Content-Length: + - '434' + Content-Type: + - application/json + User-Agent: + - X-USER-AGENT-XXX + X-Crewai-Organization-Id: + - 3433f0ee-8a94-4aa4-822b-2ac71aa38b18 + X-Crewai-Version: + - 1.9.3 + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + method: POST + uri: https://app.crewai.com/crewai_plus/api/v1/tracing/batches + response: + body: + string: '{"id":"666ec39b-46f2-4680-a271-d72357641b3f","trace_id":"d9fbeecd-afdc-48fa-817c-fe3612c4df17","execution_type":"crew","crew_name":"Unknown + Crew","flow_name":null,"status":"running","duration_ms":null,"crewai_version":"1.9.3","privacy_level":"standard","total_events":0,"execution_context":{"crew_fingerprint":null,"crew_name":"Unknown + Crew","flow_name":null,"crewai_version":"1.9.3","privacy_level":"standard"},"created_at":"2026-02-06T21:27:30.086Z","updated_at":"2026-02-06T21:27:30.086Z"}' + headers: + Connection: + - keep-alive + Content-Length: + - '492' + Content-Type: + - application/json; charset=utf-8 + Date: + - Fri, 06 Feb 2026 21:27:30 GMT + cache-control: + - no-store + content-security-policy: + - CSP-FILTERED + etag: + - ETAG-XXX + expires: + - '0' + permissions-policy: + - PERMISSIONS-POLICY-XXX + pragma: + - no-cache + referrer-policy: + - REFERRER-POLICY-XXX + strict-transport-security: + - STS-XXX + vary: + - Accept + x-content-type-options: + - X-CONTENT-TYPE-XXX + x-frame-options: + - X-FRAME-OPTIONS-XXX + x-permitted-cross-domain-policies: + - X-PERMITTED-XXX + x-request-id: + - X-REQUEST-ID-XXX + x-runtime: + - X-RUNTIME-XXX + x-xss-protection: + - X-XSS-PROTECTION-XXX + status: + code: 201 + message: Created +- request: + body: '{"messages":[{"role":"system","content":"You are a strategic planning assistant. + Create minimal, effective execution plans. Prefer fewer steps over more."},{"role":"user","content":"Create + a focused execution plan for the following task:\n\n## Task\nResearch the current + state of autonomous AI agents in 2025. Search for recent developments, then + summarize the key findings.\n\n## Expected Output\nComplete the task successfully\n\n## + Available Tools\nexa_search_tool\n\n## Planning Principles\nFocus on WHAT needs + to be accomplished, not HOW. Group related actions into logical units. Fewer + steps = better. Most tasks need 3-6 steps. Hard limit: 5 steps.\n\n## Step Types + (only these are valid):\n1. **Tool Step**: Uses a tool to gather information + or take action\n2. **Output Step**: Synthesizes prior results into the final + deliverable (usually the last step)\n\n## Rules:\n- Each step must either USE + A TOOL or PRODUCE THE FINAL OUTPUT\n- Combine related tool calls: \"Research + A, B, and C\" = ONE step, not three\n- Combine all synthesis into ONE final + output step\n- NO standalone \"thinking\" steps (review, verify, confirm, refine, + analyze) - these happen naturally between steps\n\nFor each step: State the + action, specify the tool (if any), and note dependencies.\n\nAfter your plan, + state READY or NOT READY."}],"model":"gpt-4o-mini","tool_choice":"auto","tools":[{"type":"function","function":{"name":"create_reasoning_plan","description":"Create + or refine a reasoning plan for a task with structured steps","strict":true,"parameters":{"type":"object","properties":{"plan":{"type":"string","description":"A + brief summary of the overall plan."},"steps":{"type":"array","description":"List + of discrete steps to execute the plan","items":{"type":"object","properties":{"step_number":{"type":"integer","description":"Step + number (1-based)"},"description":{"type":"string","description":"What to do + in this step"},"tool_to_use":{"type":["string","null"],"description":"Tool to + use for this step, or null if no tool needed"},"depends_on":{"type":"array","items":{"type":"integer"},"description":"Step + numbers this step depends on (empty array if none)"}},"required":["step_number","description","tool_to_use","depends_on"],"additionalProperties":false}},"ready":{"type":"boolean","description":"Whether + the agent is ready to execute the task."}},"required":["plan","steps","ready"],"additionalProperties":false}}}]}' + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '2421' + content-type: + - application/json + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NeMrAGAvmdblHHByFjeRUw89VGU\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413250,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": null,\n \"tool_calls\": [\n {\n + \ \"id\": \"call_HMhihofgWY1ITLg0xbQk5j04\",\n \"type\": + \"function\",\n \"function\": {\n \"name\": \"create_reasoning_plan\",\n + \ \"arguments\": \"{\\\"plan\\\":\\\"Research the current state + of autonomous AI agents in 2025 and summarize key findings.\\\",\\\"steps\\\":[{\\\"step_number\\\":1,\\\"description\\\":\\\"Conduct + research on recent developments in autonomous AI agents in 2025.\\\",\\\"tool_to_use\\\":\\\"exa_search_tool\\\",\\\"depends_on\\\":[]},{\\\"step_number\\\":2,\\\"description\\\":\\\"Summarize + the key findings from the research conducted on autonomous AI agents.\\\",\\\"tool_to_use\\\":null,\\\"depends_on\\\":[1]}],\\\"ready\\\":true}\"\n + \ }\n }\n ],\n \"refusal\": null,\n \"annotations\": + []\n },\n \"logprobs\": null,\n \"finish_reason\": \"tool_calls\"\n + \ }\n ],\n \"usage\": {\n \"prompt_tokens\": 456,\n \"completion_tokens\": + 105,\n \"total_tokens\": 561,\n \"prompt_tokens_details\": {\n \"cached_tokens\": + 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:27:32 GMT + Server: + - cloudflare + Set-Cookie: + - SET-COOKIE-XXX + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '2008' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: '{"messages":[{"role":"system","content":"You are Research Analyst. You + are a research analyst who searches the web for information, identifies key + findings, and produces structured research summaries.\n\nYour goal: Research + topics using search tools and produce structured summaries\n\nYou are executing + a specific step in a multi-step plan. Focus ONLY on completing\nthe current + step. Do not plan ahead or worry about future steps.\n\nBefore acting, briefly + reason about what you need to do and which approach\nor tool would be most helpful + for this specific step."},{"role":"user","content":"## Current Step\nConduct + research on recent developments in autonomous AI agents in 2025.\n\nSuggested + tool: exa_search_tool\n\nComplete this step and provide your result."}],"model":"gpt-4o-mini","tool_choice":"auto","tools":[{"type":"function","function":{"name":"exa_search_tool","description":"Search + the internet using Exa","strict":true,"parameters":{"properties":{"search_query":{"description":"Mandatory + search query you want to use to search the internet","title":"Search Query","type":"string"},"start_published_date":{"anyOf":[{"type":"string"},{"type":"null"}],"default":null,"description":"Start + date for the search","title":"Start Published Date"},"end_published_date":{"anyOf":[{"type":"string"},{"type":"null"}],"default":null,"description":"End + date for the search","title":"End Published Date"},"include_domains":{"anyOf":[{"items":{"type":"string"},"type":"array"},{"type":"null"}],"default":null,"description":"List + of domains to include in the search","title":"Include Domains"}},"required":["search_query","start_published_date","end_published_date","include_domains"],"type":"object","additionalProperties":false}}}]}' + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '1734' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NeOunS1e4G3PiTUi5abM5jlDxPW\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413252,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": null,\n \"tool_calls\": [\n {\n + \ \"id\": \"call_NpXIbiQ1fcI2sBA84tVHMIRu\",\n \"type\": + \"function\",\n \"function\": {\n \"name\": \"exa_search_tool\",\n + \ \"arguments\": \"{\\\"search_query\\\":\\\"autonomous AI agents + developments 2025\\\",\\\"start_published_date\\\":\\\"2025-01-01\\\",\\\"end_published_date\\\":\\\"2025-12-31\\\",\\\"include_domains\\\":null}\"\n + \ }\n }\n ],\n \"refusal\": null,\n \"annotations\": + []\n },\n \"logprobs\": null,\n \"finish_reason\": \"tool_calls\"\n + \ }\n ],\n \"usage\": {\n \"prompt_tokens\": 273,\n \"completion_tokens\": + 53,\n \"total_tokens\": 326,\n \"prompt_tokens_details\": {\n \"cached_tokens\": + 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:27:36 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '3226' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: '{"query": "autonomous AI agents developments 2025", "startPublishedDate": + "2025-01-01", "endPublishedDate": "2025-12-31", "type": "auto", "contents": + {"text": {"maxCharacters": 10000}}}' + headers: + Accept: + - '*/*' + Connection: + - keep-alive + Content-Length: + - '185' + Content-Type: + - application/json + User-Agent: + - X-USER-AGENT-XXX + accept-encoding: + - ACCEPT-ENCODING-XXX + x-api-key: + - X-API-KEY-XXX + method: POST + uri: https://api.exa.ai/search + response: + body: + string: "{\"requestId\":\"f2db8fbf2c04de1744531f35fde7a72b\",\"resolvedSearchType\":\"neural\",\"results\":[{\"id\":\"https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality\",\"title\":\"AI + Agents in 2025: Expectations vs. Reality\",\"url\":\"https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality\",\"publishedDate\":\"2025-02-26T00:00:00.000Z\",\"author\":null,\"text\":\"AI + Agents in 2025: Expectations vs. Reality | IBM\\nTags\\n[Artificial Intelligence] + \\n# AI agents in 2025: Expectations vs. reality\\n![the same person doing + multiple jobs] \\n## Authors\\n[Ivan Belcic] \\nStaff writer\\n[Cole Stryker] + \\nStaff Editor, AI Models\\nIBM Think\\n## AI agents in 2025: Expectations + vs. reality\\nIt\u2019s impossible to take two steps across the tech media + landscape without stumbling over an article hailing 2025 as the year of the[AI + agent]. Agents, we\u2019re told, will transform the way work is done, impacting + every facet of our lives, personal and professional.\\nWe\u2019d barely surfaced + from a landslide of NFT and crypto hype that characterized the early 2020s, + and the metaverse bubble that followed, before media voices began singing + the praises of[generative AI] (gen AI) in the wake of releases such as OpenAI\u2019s[GPT] + model family, Anthropic\u2019s[Claude] and Microsoft\u2019s Copilot.\\nWhile + the chorus hasn\u2019t moved on entirely, the focus in 2025 has shifted from[large + language models (LLMs)] to advancements in the ostensibly autonomous[artificial + intelligence (AI)] agents ushering in the future of work.\\nDespite a momentary + surge in gen AI interest around[Deepseek] \u2019s R1, which promised significant + performance improvements over ChatGPT, the dominant innovation narrative in + 2025 is the AI agent.\\nMedia coverage highlights the promises of innovation,[automation] + and efficiency agents will bring, but how much of the conversation is click-hungry + hype?\\nThe ad-supported media world thrives on clicks, and it\u2019s reasonable + to expect sensational, attention-grabbing headlines crafted to garner yours. + But what can we realistically expect from[agentic AI] in 2025, and how will + it affect our lives?\\nWe spoke with several IBM experts to cut through the + hype, with the goal of holding a more reasonable conversation about AI agents + and what they\u2019re going to do. Our team of informed insiders includes:\\n* + [Maryam Ashoori, PhD]: Director of Product Management,[IBM\xAE watsonx.ai\u2122] + \\n* [Marina Danilevsky]: Senior Research Scientist, Language Technologies\\n* + [Vyoma Gajjar]: AI Technical Solutions Architect\\n* [Chris Hay]: Distinguished + Engineer\\nIndustry newsletter\\n### The latest AI trends, brought to you + by experts\\nGet curated insights on the most important\u2014and intriguing\u2014AI + news. Subscribe to our weekly Think newsletter. See the[IBM Privacy Statement].\\n### + Thank you! You are subscribed.\\nYour subscription will be delivered in English. + You will find an unsubscribe link in every newsletter. You can manage your + subscriptions or unsubscribe[here]. Refer to our[IBM Privacy Statement] for + more information.\\n## What are AI agents?\\nAn AI agent is a software program + capable of acting autonomously to understand, plan and execute tasks. AI agents + are powered by LLMs and can interface with tools, other models and other aspects + of a system or network as needed to fulfill user goals.\\nWe\u2019re going + beyond asking a chatbot to suggest a dinner recipe based on the available + ingredients in the fridge. Agents are more than automated[customer experience] + emails that inform you it\u2019ll be a few days until a real-world human can + get to your inquiry.\\n[AI agents differ from traditional AI assistants] that + need a prompt each time they generate a response. In theory, a user gives + an agent a high-level task, and the agent figures out how to complete it.\\nCurrent + offerings are still in the early stages of approaching this idea. \u201CWhat\u2019s + commonly referred to as \u2018agents\u2019 in the market is the addition of + rudimentary planning and tool-calling (sometimes called function calling) + capabilities to LLMs,\u201D says Ashoori. \u201CThese enable the LLM to break + down complex tasks into smaller steps that the LLM can perform.\u201D\\nHay + is optimistic that more robust agents are on the way: \u201CYou wouldn\u2019t + need any further progression in models today to build[future AI agents],\u201D + he says.\\nWith that out of the way, what\u2019s the conversation about agents + over the coming year, and how much of it can we take seriously?\\nAI agents\\n### + What are AI agents?\\nFrom monolithic models to compound AI systems, discover + how AI agents integrate with databases and external tools to enhance problem-solving + capabilities and adaptability.\\n[\\nLearn more\\n] \\n## Narrative 1: 2025 + is the year of the AI agent\\n\u201CMore and better agents\u201D are on the + way, predicts Time.[1] \u201CAutonomous \u2018agents\u2019 and profitability + are likely to dominate the artificial intelligence agenda,\u201D reports Reuters.[2] + \u201CThe age of agentic AI has arrived,\u201D promises Forbes, in response + to a claim from Nvidia\u2019s Jensen Huang.[3] \\nTech media is awash with + assurances that our lives are on the verge of a total transformation. Autonomous + agents are poised to streamline and alter our jobs, drive optimization and + accompany us in our daily lives, handling our mundanities in real time and + freeing us up for creative pursuits and other higher-level tasks.\\n### 2025 + as the year of agentic exploration\\n\u201CIBM and Morning Consult did a survey + of 1,000 developers who are building AI applications for enterprise, and 99% + of them said they are exploring or developing AI agents,\u201D\_explains Ashoori. + \u201CSo yes, the answer is that 2025 is going to be the year of the agent.\u201D + However, that declaration is not without nuance.\\nAfter establishing the + current market conception of agents as LLMs with function calling, Ashoori + draws a distinction between that idea and truly autonomous agents. \u201CThe + true definition [of an AI agent] is an intelligent entity with reasoning and + planning capabilities that can autonomously take action. Those reasoning and + planning capabilities are up for discussion. It depends on how you define + that.\u201D\\n\u201CI definitely see AI agents heading in this direction, + but we\u2019re not fully there yet,\u201D says Gajjar. \u201CRight now, we\u2019re + seeing early glimpses\u2014AI agents can already analyze data, predict trends + and automate workflows to some extent. But building AI agents that can autonomously + handle complex decision-making will take more than just better algorithms. + We\u2019ll need big leaps in contextual reasoning and testing for edge cases,\u201D + she adds.\\nDanilevsky isn\u2019t convinced that this is anything new. \u201CI'm + still struggling to truly believe that this is all that different from just + orchestration,\u201D she says. \u201CYou've renamed orchestration, but + now it's called agents, because that's the cool word. But orchestration + is something that we've been doing in programming forever.\u201D\\nWith + regard to 2025 being the year of the agent, Danilevsky is skeptical. \u201CIt + depends on what you say an agent is, what you think an agent is going to accomplish + and what kind of value you think it will bring,\u201D she says. \u201CIt's + quite a statement to make when we haven't even yet figured out ROI (return + on investment) on LLM technology more generally.\u201D\\nAnd it\u2019s not + just the business side that has her hedging her bets. \u201CThere's the + hype of imagining if this thing could think for you and make all these decisions + and take actions on your computer. Realistically, that's terrifying.\u201D\\nDanilevsky + frames the disconnect as one of miscommunication. \u201C[Agents] tend to be + very ineffective because humans are very bad communicators. We still can't + get chat agents to interpret what you want correctly all the time.\u201D\\nStill, + the forthcoming year holds a lot of promise as an era of experimentation. + \u201CI'm a big believer in [2025 as the year of the agent],\u201D says + Hay excitedly.\\nEvery large tech company and hundreds of startups are now + experimenting with agents. Salesforce, for example, has released their Agentforce + platform, which enables users to create agents that are easily integrated + within the Salesforce app ecosystem.\\n\u201CThe wave is coming and we're + going to have a lot of agents. It's still a very nascent ecosystem, so + I think a lot of people are going to build agents, and they're going to + have a lot of fun.\u201D\\n## Narrative 2: Agents can handle highly complex + tasks on their own\\nThis narrative assumes that today\u2019s agents meet + the theoretical definition outlined in the introduction to this piece. 2025\u2019s + agents will be fully autonomous AI programs that can scope out a project and + complete it with all the necessary tools they need and with no help from human + partners. But what\u2019s missing from this narrative is nuance.\\n### Today\u2019s + models are more than enough\\nHay believes that the groundwork has already + been laid for such developments. \u201CThe big thing about agents is that + they have the ability to plan,\u201D he outlines. \u201CThey have the ability + to reason, to use tools and perform tasks, and they need to do it at speed + and scale.\u201D\\nHe cites 4 developments that, compared to the best models + of 12 to 18 months ago, mean that the models of early 2025 can power the agents + envisioned by the proponents of this narrative:\\n* Better, faster, smaller + models\\n* Chain-of-thought (COT) training\\n* Increased context windows\\n* + Function calling\\n\u201CNow, most of these things are in play,\u201D Hay + continues. \u201CYou can have the AI call tools. It can plan. It can reason + and come back with good answers. It can use inference-time compute. You\u2019ll + have better chains of thought and more memory to work with. It's going + to run fast. It\u2019s going to be cheap. That leads you to a structure where + I think you can have agents. The models are improving and they're getting + better, so that's only going to accelerate.\u201D\\n### Realistic expectations + are a must\\nAshoori is careful to differentiate between what agents will + be able to do later, and what they can do now. \u201CThere is the promise, + and there is what the agent's capable of doing today,\u201D she says. + \u201CI would say the answer depends on the use case. For simple use cases, + the agents are capable of [choosing the correct tool], but for more sophisticated + use cases, the technology\",\"image\":\"https://www.ibm.com/content/dam/connectedassets-adobe-cms/worldwide-content/gbs/cf/ul/g/5a/88/336.jpg/_jcr_content/renditions/cq5dam.thumbnail.1280.1280.png\",\"favicon\":\"https://www.ibm.com/content/dam/adobe-cms/default-images/icon-32x32.png\"},{\"id\":\"https://www.youtube.com/watch?v=RhRJsjr-I3I\",\"title\":\"AWS + re:Invent 2025 - Autonomous AI agents you can truly ...\",\"url\":\"https://www.youtube.com/watch?v=RhRJsjr-I3I\",\"publishedDate\":\"2025-12-06T21:27:38.021Z\",\"author\":\"AWS + Events\",\"text\":\"Learn how to power accurate and trusted agents at scale + with AI-ready data and governance. This presentation is brought to you by + IBM, an AWS Partner.\\\\n\\\\nLearn more:\\\\nMore AWS events: https://go.aws/3kss9CP + \\\\n\\\\nSubscribe:\\\\nMore AWS videos: http://bit.ly/2O3zS75\\\\nMore AWS + events videos: http://bit.ly/316g9t4\\\\n\\\\nABOUT AWS:\\\\nAmazon Web Services + (AWS) hosts events, both online and in-person, bringing the cloud computing + community together to connect, collaborate, and learn from AWS experts.\\\\n + AWS is the world's most comprehensive and broadly adopted cloud platform, + offering over 200 fully featured services from data centers globally. Millions + of customers\u2014including the fastest-growing startups, largest enterprises, + and leading government agencies\u2014are using AWS to lower costs, become + more agile, and innovate faster.\\\\n\\\\n#AWSreInvent #AWSreInvent2025 #AWS\\n| + view_count: 128 views | short_view_count: 128 views | num_likes: 1 like | + num_subscribers: 170 thousand | duration: 19 minutes 10 seconds\",\"image\":\"https://i.ytimg.com/vi/RhRJsjr-I3I/hqdefault.jpg\"},{\"id\":\"https://kpmg.com/sk/en/insights/2025/06/autonomous-ai-agents-reshape-business-landscape.html\",\"title\":\"Autonomous + AI Agents Are Reshaping the Business ...\",\"url\":\"https://kpmg.com/sk/en/insights/2025/06/autonomous-ai-agents-reshape-business-landscape.html\",\"publishedDate\":\"2025-06-25T00:00:00.000Z\",\"author\":null,\"text\":\"Autonomous + AI Agents Are Reshaping the Business Landscape\\n[Skip to main content] \\n[contact\\\\_mail\\nContact + us\\n] \\n#### ###### [\\n] \\n[tips\\\\_and\\\\_updates\\nStay informed with + our newsletters\\n] \\nlanguage\\nSlovakia (EN)\\nexpand\\\\_more\\nclose\\nSearchcancel\\nmenu\\n[![KPMG + logo]] \\n[contact\\\\_mail\\nContact us\\n] \\n#### ###### [\\n] \\n[tips\\\\_and\\\\_updates\\nStay + informed with our newsletters\\n] \\nlanguage\\nSlovakia (EN)\\nexpand\\\\_more\\nsearch\\nSearch\\nsearch\\nYour + search term was too short.\\nSearch terms must include 3 or more characters.\\nPrevious + search terms\\nclose\\nCancel\\n# Autonomous AI Agents Are Reshaping the Business + Landscape\\nArtificial intelligence that can think and act independently is + no longer just a vision of the future \u2014it\u2019s becoming a competitive + advantage\\nevent25-06-2025\\nShare\\n* [] \\n* [] \\n![3D Rendering futuristic + robot technology development, artificial intelligence AI, and machine learning + concept. Global robotic bionic science research for future of human life.] + \\n* [From GenAI to AgAI] \\n* [Technical Foundations of AgAI] \\n* [AI Agents + as Digital Workforce] \\nAgentic artificial intelligence\u2014autonomous systems + capable of perceiving, reasoning, and acting independently\u2014is no longer + just a vision of the future. This rapidly evolving technology is already transforming + how businesses operate, innovate, and compete to stay ahead.\\nIn our two-part + blog series,***The Age of Agentic AI***, we explore the rise of agentic AI + (AgAI) from both a business and a technological perspective. We\u2019ll trace + its development, examine its technical foundations, highlight real-world applications, + and assess its impact across industries\u2014along with the risks and opportunities + it presents. Through expert insights and case studies, we\u2019ll show that + agentic AI isn\u2019t just another tool\u2014it\u2019s a new kind of digital + workforce, reshaping the future of work, decision-making, and enterprise strategy.\\n### + Author\\n##### **\\nAlexander Zagnetko\\n***Manager,*Process Organization + and Improvement\\n[Contact us for more information] \\n#### From Generative + to Agentic AI\\nOver the past five years, we\u2019ve witnessed a dramatic + leap in AI capabilities. From the launch of GPT-3 in 2020 to the widespread + adoption of models like GPT-4, Claude, and Gemini between 2023 and 2024, generative + AI has already revolutionized numerous industries. But as impressive as these + developments are, they represent just the beginning of a much broader transformation.\\nThe + next major milestone is the rise of autonomous agents\u2014systems that can + now carry out complex workflows and collaborate with humans or other agents. + These systems set their own goals, plan their actions, use tools, and adapt + to changing conditions. Unlike traditional AI agents designed for narrow, + specific tasks, agentic AI systems operate with much greater independence. + They are built to:\\n* Perceive their environment through both structured + and unstructured data,\\n* Plan actions using logic and decision-making frameworks,\\n* + Execute tasks using tools, APIs, or physical devices,\\n* Learn from feedback + and improve over time,\\n* Collaborate with humans and other agents.\\nThe + shift from generative to agentic AI marks a fundamental leap in artificial + intelligence. It stems from the integration of multiple capabilities: large + language models (LLMs), memory systems, planning algorithms, tool integration, + and multi-agent coordination. Together, these elements enable AI systems to + function as digital workers\u2014capable of performing tasks that once required + human judgment and coordination. Agentic AI is opening up entirely new possibilities + and will have a profound impact on business strategy, organizational design, + and competitive advantage.\\n#### Technical Foundations of Agentic AI\\nAgAI + systems are built on a layered architecture that combines several advanced + technologies. Let\u2019s break down the core components. At the heart of most + agentic systems is a**large language model (LLM)**such as GPT-4, Claude, or + Gemini. These models are trained on massive datasets and can understand and + generate human-like language. They provide the linguistic and reasoning capabilities + that underpin agent behaviour. Unlike traditional chatbots, agentic systems + need to**remember past interactions**, decisions, and outcomes. Memory enables + agents to learn from experience, personalize interactions, and maintain continuity + over time.\\nFor agents to act autonomously, they must be able to plan several + steps ahead and consider the consequences of their decisions. This is why + they combine reasoning with tool use\u2014a method known as ReAct (Reasoning + + Acting)\u2014and explore multiple logical pathways before choosing a + course of action, a technique referred to as Tree-of-Thoughts. Another well-known + approach to evaluating potential actions is Monte Carlo Tree Search, originally + developed for game-playing AI. These techniques enable agents to break down + complex goals into manageable tasks and respond flexibly to feedback.\\nAs + agents become more autonomous, ensuring their responsible and ethical behavior + becomes increasingly important. This is essential for building trust and deploying + agentic AI in a safe and accountable way. Key mechanisms include:\\n* **Human-in-the-loop**\u2013 + giving humans the ability to approve or override an agent\u2019s decisions.\\n* + **Audit trails**\u2013 tracking the agent\u2019s actions to ensure accountability.\\n* + **Ethical constraints**\u2013 rules that prevent harmful or unethical behavior.\\n* + **Value alignment**\u2013 ensuring agents act in accordance with human values.\\n#### + AI Agents as the New Digital Workforce\\nAgentic AI is opening up new possibilities + for businesses in terms of efficiency, scalability, and innovation. Whether + it\u2019s automating routine tasks or orchestrating complex workflows, agentic + AI is finding broad applications across sectors\u2014from financial services + and healthcare to education, retail, and software development.\\naccount\\\\_balance\\\\_wallet## + Financial Services\\nThe financial sector has long been a frontrunner in AI + adoption, and agentic AI takes automation to the next level. Autonomous agents + can monitor market conditions in real time, assess risks, and execute trades + with minimal human oversight. They continuously scan transactions for suspicious + activity, track regulatory updates, and ensure compliance. In customer service + and advisory roles, agents analyze client data and recommend tailored financial + products, effectively acting as digital financial advisors.\\nmedical\\\\_services## + Healthcare\\nHealthcare is one of the most complex and data-intensive sectors, + making it a natural fit for agentic AI. Agents can analyze medical records, + test results, and imaging data, flag potential drug interactions, suggest + additional diagnostics, and assist with documentation. They also coordinate + care by scheduling appointments, managing referrals, and reducing redundancies. + On the administrative side, agents streamline tasks like billing, claims processing, + and regulatory compliance, allowing medical professionals to focus more on + patient care.\\ngavel## Legal and Compliance\\nThe legal sector is undergoing + a major transformation thanks to agentic AI. Autonomous agents can read, interpret, + and generate legal documents, analyze contracts, highlight key clauses, flag + risks, suggest edits, and draft standard agreements based on templates and + client requirements. They also monitor regulatory changes across jurisdictions, + evaluate their impact, and recommend compliance actions. In litigation, they + assist with legal research, summarizing precedents, and preparing arguments.\\ncast\\\\_for\\\\_education## + Education and Training\\nDigital transformation has reached education as well. + AI agents serve as interactive tutors, answering questions in real time, providing + feedback, and adapting to individual learning styles. They can even detect + emotional cues to offer support when motivation is low. In assessments, agents + automatically grade assignments, generate tests, and provide detailed feedback, + while identifying performance trends and recommending timely interventions.\\nstorefront## + Retail and E-commerce\\nIn the retail sector, companies use agentic AI to + enhance customer experience, optimize operations, and boost sales. In logistics, + agents forecast demand, manage inventory, coordinate with suppliers, and respond + to market changes in real time\u2014for example, by rerouting deliveries or + adjusting pricing. In marketing, they design targeted campaigns, segment audiences, + and optimize content, improving product placement and promotional strategies + based on customer behavior.\\nwebhook## Software Development\\nAgentic AI + is also revolutionizing the entire software development lifecycle\u2014from + design to deployment. Agents can generate code from natural language, detect + and fix bugs, optimize performance, and create documentation and tests. In + product management, they gather user feedback, prioritize features, draft + specifications, and coordinate team collaboration, accelerating innovation + and improving product quality.\\n#### Speed of Implementation Is the Competitive + Advantage\\nAgentic AI is not just about boosting efficiency\u2014it\u2019s + a fundamental shift in how companies operate, make decisions, and create value. + By combining large language models, memory systems, planning algorithms, and + the ability to interact with external tools, agentic AI becomes an active + participant in business processes. Its impact goes beyond any single department\u2014it\u2019s + transforming entire industries, from finance and healthcare to software development.\\nOrganizations + that understand and implement this technology early will gain a competitive + edge\u2014not only through speed and precision but also through their ability + to respond to market changes, personalize services, and innovate faster than + ever before.\\n## Contact our experts\\nShould you wish more information on + how we can help your business or to arrange a meeting for personal presentation\",\"image\":\"/content/dam/kpmgsites/sk/images/2025/Agentic_AI_cover.jpg\",\"favicon\":\"https://kpmg.com/etc.clientlibs/kpmg/clientlibs/clientlib-site/resources/images/favicons/favicon-32x32.svg\"},{\"id\":\"https://cobusgreyling.medium.com/the-state-of-ai-agents-3e13fa123ab8\",\"title\":\"The + State of AI Agents - Cobus Greyling\",\"url\":\"https://cobusgreyling.medium.com/the-state-of-ai-agents-3e13fa123ab8\",\"author\":\"Cobus + Greyling\",\"text\":\"The State of AI Agents. AI Agents burst on the scene + not too\u2026 | by Cobus Greyling | Medium\\n[Sitemap] \\n[Open in app] \\nSign + up\\n[Sign in] \\n[Medium Logo] \\n[\\nWrite\\n] \\n[\\nSearch\\n] \\nSign + up\\n[Sign in] \\n![] \\nPress enter or click to view image in full size\\n![] + \\n# The State of AI Agents\\n## AI Agents burst on the scene not too long + ago\u2026the potential of AI Agents is real\u2026but the hype has been met + with real-world impediments when it comes to implementation.\\n[\\n![Cobus + Greyling] \\n] \\n[Cobus Greyling] \\n6 min read\\n\xB7Nov 3, 2025\\n[\\n] + \\n--\\n2\\n[] \\nListen\\nShare\\nThere has been a number of[***sobering + studies***] when it comes to the realities of implementing AI Agents\u2026\\nThe + existence of the[***Pareto Frontier***] ***(cost vs accuracy)***has been well + examined and documented.\\n[NVIDIA] has been taking the lead in moving away + from the idea of an AI Agent being underpinned by a single LLM.\\nThey are + advocating the use of multiple Small Language Models (SLMs) and orchestrating + the SLMs in an[Agentic workflow].\\n> Latency and cost have also been impediments + to AI agent use.\\nCost is being addressed by using and orchestrating multiple[***open-sourced + SLMs***] which is***fine-tuned***for specific tasks within an AI Agent.\\n***Latency***is + addressed with[parallel running task] s within an AI Agent.\\nOr having multiple + AI Agents within a Agentic workflow, and executing those AI Agents in parallel + for complex tasks.\\n[Lighter AI Agents] \u2026NVIDIA has also been backing + this growing trend of**agentic workflows**where the orchestrating AI Agent + (the \u201Cconductors\u201D that plan, route tasks, and synthesise outputs) + are indeed***leaning toward being more***[***lightweight***] ***and less code-intensive + (AI Agents)***to implement.\\n***Agentic Workflows***are easy to build in + general, especially when presented via a GUI builder, that is no-code and + handles tasks like versioning, collaboration and more.\\nDrift and continuous + improvement are addressed with what[NVIDIA refers to as a***data-flywhee***] + ***l***.\\nWhere a***continuous feedback loop***is established for training + data to improve the different models orchestrated in the AI Agent/Agentic + Workflow.\\n> Most of these challenges I detailed in an [**> earlier post\\n**] + > .\\nAlso, for a long time there has been this notion of different[**levels + of AI Agents**].\\nTypically there are[**5 or 6 levels defined**, in a number + of studies].\\n[**HuggingFace**also published a study] where they detail the + shifting relationship between humans and AI Agents\u2026\\nThis is why I found + this recent study interesting (*the image below*) where they extended the + levels from what was typically**5 levels**, to**6 levels**.\\nThe one extreme + is Level 0, here the human dominates and AI plays no role. And on the other + extreme, Level 5, where the human has no role and the AI takes full control.\\nAnd + in between are these steps of going from one level to another.\\n> This approach + really resonates with me as it so sober in dissecting the gradual approach + in task automation transition.\\n***Where are we now, based on this chart?***\\nPress + enter or click to view image in full size\\n![] \\nWe have been introducing + assistance from the days of Chatbots and NLU models\u2026\\n\u2026currently + we are in the process of transferring tasks from human dominance to AI dominance. + This is happening on a***task-by-task basis.***\\nHence this***movement***must + not be seen as a single wave hitting enterprises and our way-of-work.\\nBut + rather a detailed approach of identifying tasks suited (for now) to move along + this trajectory.\\nConsidering the image below, it breaks down the***traditional***four + pillars of AI Agents.\\n*In this article I will stay with this for the sake + of simplicity. But to some degree this approach has been upended by Agentic + Workflows and multiple smaller language models.*\\n> A recent study also introduced + a data component\u2026data lakes.\\nSo not only is there focus on context + / memory and[tools]. But also giving the AI Agent access to vast amounts of + data to help the AI Agent perform user requests.\\nPress enter or click to + view image in full size\\n![] \\nThe table below (from the study) shows the + contrast between**traditional AI Agents**and what the study terms**Data Agents**.\\nPress + enter or click to view image in full size\\n![] \\n***The future?***\\n[For + AI Agents to be truly autonomous and able to address virtually any task], + it needs to be able to write code.\\n> On this front [> OpenAI\\n] > and [> + xAI\\n] > introduced functions / tools which are able to write code.\\nHence + the AI Agent gets an instruction, there is no pre-existing tool or function + to perform the request, but the AI Agent has the ability to write code to + achieve a goal.\\n***There exist limitations\u2026***\\nSandbox Restrictions + for Security\u2026AI models operate within tightly controlled sandboxes to + mitigate risks, prohibiting arbitrary software installs or direct internet + access.\\nThey\u2019re confined to pre-loaded libraries, forcing clever workarounds + like built-in proxies.\\nWhile this setup enables basic functionality, it + blocks the creation of truly innovative tools that demand unrestricted resources + \u2014limiting adaptability to predefined, bounded tasks.\\nDebugging remains + a weak spot, models often hallucinate subtle bugs in intricate logic and finite + token limits***(coupled with cost constraints)***make exhaustive testing impractical.\\nThey + excel at 70\u201380% of straightforward objectives but stumble on edge cases, + frequently demanding human intervention.\\nIn my experience, AI Agents generate + code that***looks***flawless, only for execution in notebooks or live environments + to reveal flaws \u2014necessitating rounds of re-prompting.\\nThe bigger the + code block, the steeper the challenge. Autocomplete and highly contextual + use in IDE\u2019s work quite well.\\nAmbiguous instructions derail progress, + with agents chasing***tangential sub-goals instead of the core intent***.\\nThey\u2019re + far from self-directed; precision is essential to keep them aligned.\\nThis + is why detailed, step-by-step breakdowns are non-negotiable.\\nTools like + Gemini thrive in Colab notebooks precisely because they leverage rich contextual + anchors \u2014iterating on proven, working code cell by cell, rather than + from scratch.\\n***If you are still reading, thank you\u2026I hope it was + helpful, any feedback will be welcome\u2026***\\nBelow are a number of resources + that might be of help on this topic.\\nPress enter or click to view image + in full size\\n![] \\n[***Chief Evangelist***] ***@***[*Kore.ai*] *| I\u2019m + passionate about exploring the intersection of AI and language. Language Models, + AI Agents, Agentic Apps, Dev Frameworks & Data-Driven Tools shaping tomorrow.*\\nPress + enter or click to view image in full size\\n![] \\n[\\n## A Survey of Data + Agents: Emerging Paradigm or Overstated Hype?\\n### The rapid advancement + of large language models (LLMs) has spurred the emergence of data agents--autonomous + systems\u2026\\narxiv.org\\n] \\n[\\n## COBUS GREYLING\\n### Where AI Meets + Language | Language Models, AI Agents, Agentic Applications, Development Frameworks + & Data-Centric\u2026\\nwww.cobusgreyling.com\\n] \\n[\\n## Five Levels + Of AI Agents\\n### AI Agents are defined as artificial entities which can + perceive their environment, make decisions and take actions\u2026\\ncobusgreyling.medium.com\\n] + \\n[\\n## 5 Levels Of AI Agents (Updated)\\n### \U0001D5D4\U0001D602\U0001D601\U0001D5FC\U0001D5FB\U0001D5FC\U0001D5FA\U0001D5FC\U0001D602\U0001D600\U0001D5D4\U0001D5DC\U0001D5D4\U0001D5F4\U0001D5F2\U0001D5FB\U0001D601\U0001D600are + AI systems capable of performing a series of complex tasks independently to\u2026\\ncobusgreyling.medium.com\\n] + \\n[\\n## Fully Autonomous AI Agents Should Not Be Developed\\n### Research + from HuggingFace\u2026and they coined a new definition for what an AI Agent + is.\\ncobusgreyling.medium.com\\n] \\n[\\n## The Evolution of AI Agents\\n### + \u2026it is helpful to consider how far we have come in few years\u2019 time\u2026\\ncobusgreyling.medium.com\\n] + \\n[\\n## GitHub - HKUSTDial/awesome-data-agents: Continuously updated paper + list on advancements in Data\u2026\\n### Continuously updated paper list on + advancements in Data Agents. Companion repo to our paper "A Survey of + Data Agents\u2026\\ngithub.com\\n] \\n[\\nAI\\n] \\n[\\nArtificial Intelligence\\n] + \\n[\\nMachine Learning\\n] \\n[\\nSoftware Development\\n] \\n[\\nSoftware + Engineering\\n] \\n[\\n![Cobus Greyling] \\n] \\n[\\n![Cobus Greyling] \\n] + \\n[## Written byCobus Greyling\\n] \\n[29K followers] \\n\xB7[0 following] + \\nat the intersection of AI & language.[www.cobusgreyling.com] \\n## + Responses (2)\\n[] \\nSee all responses\\n[\\nHelp\\n] \\n[\\nStatus\\n] \\n[\\nAbout\\n] + \\n[\\nCareers\\n] \\n[\\nPress\\n] \\n[\\nBlog\\n] \\n[\\nPrivacy\\n] \\n[\\nRules\\n] + \\n[\\nTerms\\n] \\n[\\nText to speech\\n]\",\"image\":\"https://miro.medium.com/v2/resize:fit:1200/1*1FdlyvCKYt0nc3rGwqZ6dA.png\",\"favicon\":\"https://miro.medium.com/v2/5d8de952517e8160e40ef9841c781cdc14a5db313057fa3c3de41c6f5b494b19\"},{\"id\":\"https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325\",\"title\":\"AI + agents arrived in 2025 \u2013 here's what happened and ...\",\"url\":\"https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325\",\"publishedDate\":\"2025-12-29T00:00:00.000Z\",\"author\":\"Thomas + \u015Eerban von Davier\",\"text\":\"AI agents arrived in 2025 \u2013here\u2019s + what happened and the challenges ahead in 2026\\n[] [] \\n[![The Conversation]] + \\nAcademic rigour, journalistic flair\\n![a couple dozen robot face emojis + floating between two human hands] \\nAI agents have emerged from the lab, + bringing promise and peril.[tadamichi/iStock via Getty Images] \\n# **AI agents + arrived in 2025 \u2013here\u2019s what happened and the challenges ahead in2026**\\nPublished: + December 29, 2025 4.35pm CET\\n[****Thomas \u015Eerban von Davier,*Carnegie + Mellon University*] \\n### Author\\n1. [![] Thomas \u015Eerban von Davier] + \\nAffiliated Faculty Member, Carnegie Mellon Institute for Strategy and Technology, + Carnegie Mellon University\\n### Disclosure statement\\nThomas \u015Eerban + von Davier does not work for, consult, own shares in or receive funding from + any company or organisation that would benefit from this article, and has + disclosed no relevant affiliations beyond their academic appointment.\\n### + Partners\\n[] \\n[Carnegie Mellon University] provides funding as a member + of The Conversation US.\\n[View all partners] \\n### DOI\\n[https://doi.org/10.64628/AAI.maxh7d4en] + \\nhttps://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325\\nhttps://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325\\nLink + copied\\nShare article\\nShare article\\nCopy link[Email] \\n[Bluesky] [Facebook] + [WhatsApp] [Messenger] [LinkedIn] [X (Twitter)] \\nPrint article\\nIn artificial + intelligence, 2025 marked a decisive shift. Systems once confined to research + labs and prototypes began to appear as everyday tools. At the center of this + transition was the rise of AI agents \u2013AI systems that can use other software + tools and act on their own.\\nWhile researchers have studied AI for more than + 60 years, and the term \u201Cagent\u201D has long been part of the field\u2019s + vocabulary, 2025 was the year the concept became concrete for developers and + consumers alike.\\nAI agents moved from theory to infrastructure, reshaping + how people interact with large language models, the systems that power chatbots + like ChatGPT.\\nIn 2025, the definition of AI agent shifted from the[academic + framing] of systems that perceive, reason and act to AI company[Anthropic\u2019s + description] of large language models that are capable of using software tools + and taking autonomous action. While large language models have long excelled + at text-based responses, the recent change is their expanding capacity to + act, using tools, calling[APIs], coordinating with other systems and completing + tasks independently.\\nThis shift did not happen overnight. A key inflection + point came in late 2024, when Anthropic released the[Model Context Protocol]. + The protocol allowed developers to connect large language models to external + tools in a standardized way, effectively giving models the ability to act + beyond generating text. With that, the stage was set for 2025 to become the + year of AI agents.\\n[![Embedded YouTube video]] \\nAI agents are a whole + new ballgame compared with generative AI.## The milestones that defined 2025\\nThe + momentum accelerated quickly. In January, the release of Chinese model[DeepSeek-R1] + as an[open-weight] model disrupted assumptions about who could build high-performing + large language models, briefly rattling markets and intensifying global competition. + An open-weight model is an AI model whose training, reflected in values called + weights, is publicly available. Throughout 2025, major U.S. labs such as[OpenAI],[Anthropic],[Google] + and[xAI] released larger, high-performance models, while Chinese tech companies + including[Alibaba],[Tencent], and[DeepSeek] expanded the open-model ecosystem + to the point where the Chinese models have been[downloaded more than American + models].\\n##### Another turning point came in April, when Google introduced + its[Agent2Agent protocol]. While Anthropic\u2019s Model Context Protocol focused + on how agents use tools, Agent2Agent addressed how agents communicate with + each other. Crucially, the two protocols were designed to work together. Later + in the year, both[Anthropic] and[Google] donated their protocols to the open-source + software nonprofit Linux Foundation, cementing them as open standards rather + than proprietary experiments.\\nThese developments quickly found their way + into consumer products. By mid-2025, \u201Cagentic browsers\u201D began to + appear. Tools such as[Perplexity\u2019s Comet],[Browser Company\u2019s Dia],[OpenAI\u2019s + GPT Atlas],[Copilot in Microsoft\u2019s Edge],[ASI X Inc.\u2019s Fellou],[MainFunc.ai\u2019s + Genspark],[Opera\u2019s Opera Neon] and others reframed the browser as an + active participant rather than a passive interface. For example, rather than + helping you search for vacation details, it plays a part in booking the vacation.\\nAt + the same time, workflow builders like[n8n] and[Google\u2019s Antigravity] + lowered the technical barrier for creating custom agent systems beyond what + has already happened with coding agents like[Cursor] and[GitHub Copilot].\\n## + New power, new risks\\nAs agents became more capable, their risks became harder + to ignore. In November, Anthropic disclosed how its Claude Code agent[had + been misused] to automate parts of a cyberattack. The incident illustrated + a broader concern: By automating repetitive, technical work, AI agents can + also lower the barrier for malicious activity.\\nThis tension defined much + of 2025. AI agents expanded what individuals and organizations could do, but + they also[amplified existing vulnerabilities]. Systems that were once isolated + text generators became interconnected, tool-using actors operating with little + human oversight.\\n[![Embedded YouTube video]] \\nThe business community is + gearing up for multiagent systems.## What to watch for in 2026\\nLooking ahead, + several open questions are likely to shape the next phase of AI agents.\\nOne + is benchmarks. Traditional benchmarks, which are like a structured exam with + a series of questions and standardized scoring, work well for single models, + but[agents are composite systems] made up of models, tools, memory and decision + logic. Researchers increasingly want to evaluate[not just outcomes, but processes]. + This would be like asking students to show their work, not just provide an + answer.\\nProgress here will be critical for improving reliability and trust, + and ensuring that an AI agent will perform the task at hand. One method is + establishing clear definitions around[AI agents and AI workflows]. Organizations + will need to map out exactly where AI will[integrate into workflows or introduce + new ones].\\nAnother development to watch is governance. In late 2025, the + Linux Foundation announced the creation of the[Agentic AI Foundation], signaling + an effort to establish shared standards and best practices. If successful, + it could play a role like the[World Wide Web Consortium] in shaping an open, + interoperable agent ecosystem.\\nThere is also a growing debate over model + size. While large, general-purpose models dominate headlines, smaller and + more specialized models are often[better suited to specific tasks]. As agents + become configurable consumer and business tools, whether through browsers + or workflow management software, the power to choose the right model increasingly + shifts to users rather than labs or corporations.\\n## The challenges ahead\\nDespite + the optimism, significant socio-technical challenges remain. Expanding data + center infrastructure[strains energy grids] and affects local communities. + In workplaces, agents raise concerns about automation,[job displacement] and + surveillance.\\nFrom a security perspective, connecting models to tools and + stacking agents together[multiplies risks] that are already unresolved in + standalone large language models. Specifically, AI practitioners are addressing + the dangers of[indirect prompt injections], where prompts are hidden in open + web spaces that are readable by AI agents and result in harmful or unintended + actions.\\nRegulation is another unresolved issue. Compared with[Europe] and[China], + the United States has relatively limited oversight of algorithmic systems. + As AI agents become embedded across digital life, questions about access, + accountability and limits remain largely unanswered.\\nMeeting these challenges + will require more than technical breakthroughs. It demands[rigorous engineering + practices], careful design and clear documentation of how systems work and + fail. Only by treating AI agents as socio-technical systems rather than mere + software components, I believe, can we build an AI ecosystem that is both + innovative and safe.\\n**\\n* [Artificial intelligence (AI)] \\n* [Google] + \\n* [Technology] \\n* [OpenAI] \\n* [Anthropic] \\n* [AI safety] \\n* [AI + agents] \\n### Events\\n[More events] \\n### Jobs\\n* ##### [Deputy Editor] + \\n* ##### [Director of Professional Development] \\n* ##### [Research Fellow + in Cybersecurity for Cyber-physical systems] \\n* ##### [University Librarian] + \\n* ##### [Video Commissioning Editor] \\n[More jobs]\",\"image\":\"https://images.theconversation.com/files/709953/original/file-20251219-66-te6uyi.jpg?ixlib=rb-4.1.0&rect=0%2C250%2C8000%2C4000&q=45&auto=format&w=1356&h=668&fit=crop\",\"favicon\":\"https://cdn.theconversation.com/static/tc/logos/web-app-logo-192x192-2d05bdd6de6328146de80245d4685946.png\"},{\"id\":\"https://www.sculptsoft.com/agentic-ai-in-action-how-autonomous-ai-agents-are-changing-software-development-in-2025/\",\"title\":\"How + Autonomous AI Agents Are Changing Software ...\",\"url\":\"https://www.sculptsoft.com/agentic-ai-in-action-how-autonomous-ai-agents-are-changing-software-development-in-2025/\",\"publishedDate\":\"2025-10-15T00:00:00.000Z\",\"author\":\"Hiloni + Mehta\",\"text\":\"Agentic AI in Action: How Autonomous AI Agents Are Changing + Software Development in 2025 - SculptSoft\\n[![logo]![logo]] \\n**\\n**\\n**\\n[\\nBusiness + Inquiry\\n] \\n[\\nHire Our Experts\\n] \\n# Agentic AI in Action: How Autonomous + AI Agents Are Changing Software Development in 2025\\nPublished By:**[Hiloni + Mehta] **\\nPublished On:**[October 15, 2025] **\\nPublished In:**[AI] **\\n[Introduction] + \\n[What Is Agentic AI and How Does It Work?] \\n[How Autonomous AI Agents + Are Transforming Software Development in 2025] \\n[Key Benefits of Agentic + AI in Software Development] \\n[Use Cases of Agentic AI in Software Development] + \\n[The Future of Software Development: From Teams to AI Ecosystems] \\n[How + SculptSoft Builds Agentic AI Solutions for Modern Businesses] \\n[Conclusion] + \\n[FAQs] \\n#### Introduction\\nSoftware development is entering a new era + where artificial intelligence doesn\u2019t just assist developers, but actively + builds, tests, and improves applications on its own.\\nIn 2025, this evolution + is being led by a new class of intelligent systems known as Agentic AI – + autonomous AI agents capable of perceiving, reasoning, and taking actions + independently. Unlike traditional AI tools that require constant human input, + Agentic AI can analyze goals, make decisions, and execute tasks end-to-end + transforming how businesses create and manage software.\\nFor years, companies + have struggled with familiar challenges:\\n* Slow development cycles that + delay releases.\\n* Rising engineering costs with limited resources.\\n* Complex + integrations across legacy systems.\\n* Dependency on manual processes that + limit scalability.\\nAgentic AI changes this dynamic. These intelligent agents + can automate entire parts of the development lifecycle from writing and reviewing + code to running tests and deploying updates, freeing teams to focus on innovation + rather than repetitive work.\\nAs a result, organizations adopting Agentic + AI aren\u2019t just accelerating development speed; they\u2019re reimagining + what software teams can achieve with autonomous systems that think, act, and + learn continuously.\\nIn this blog, we\u2019ll explore how Agentic AI is transforming + software development in 2025, the real-world impact it\u2019s already making, + and how businesses can strategically implement it to gain a long-term competitive + edge.\\n#### What Is Agentic AI and How Does It Work?\\nBefore diving into + how Agentic AI is transforming software development, it\u2019s important to + understand what it actually is and why it\u2019s considered the next major + leap in[artificial intelligence].\\nAgentic AI refers to a new generation + of autonomous AI systems that can perceive information, reason through objectives, + and take actions independently all without requiring continuous human instructions. + Unlike traditional AI tools that respond to commands or prompts, Agentic AI + Agents operate based on goals. They can interpret context, plan tasks, make + decisions, and execute actions on their own much like how a skilled developer + might handle an assignment from start to finish.\\n##### How Agentic AI Works\\nAgentic + AI works through a continuous cognitive loop that mirrors human-like thinking. + Each agent follows a process of perception, reasoning, action, and reflection, + enabling it to learn and improve over time.\\nHere\u2019s how this plays out + in practice:\\n* **Perception:**The AI Agent gathers context such as project + requirements, past code commits, or bug reports from connected systems.\\n* + **Reasoning:**It analyzes this information using its internal logic, memory, + and large language models to determine the best next steps.\\n* **Action:**The + agent performs the required tasks like generating code, testing modules, or + deploying builds through integrated tools or APIs.\\n* **Reflection:**It evaluates + the outcome, identifies what worked or failed, and refines its next decisions + accordingly.\\n##### How Agentic AI Differs from Traditional AI\\nTraditional + AI models are reactive, they provide answers when prompted. In contrast, Agentic + AI is proactive. It acts on defined goals, manages tasks across multiple systems, + and even collaborates with other AI agents to complete larger objectives.\\nFor + instance, while a conventional AI coding assistant can help generate snippets + of code, an Agentic AI system can oversee the entire process from analyzing + requirements to writing, testing, and deploying code autonomously.\\nThis + shift marks the beginning of AI-driven software ecosystems where intelligent + agents work alongside human teams to optimize speed, quality, and scalability.\\nTo + understand**what Agentic AI is, how it functions, and the complete lifecycle + of an AI agent**, read our detailed guide here:[**What Is Agentic AI? A Complete + Guide to Its Lifecycle and Functions**].\\n#### How Autonomous AI Agents Are + Transforming Software Development in 2025\\nSoftware development has always + been about collaboration of designers, developers, testers, and project managers + working together to bring an idea to life. But in 2025, a new kind of collaborator + has joined the team:**Autonomous AI Agents**.\\nThese intelligent, self-operating + AI systems go beyond simple automation. They understand objectives, make independent + decisions, and execute end-to-end workflows within the[software development + lifecycle] (SDLC) transforming how modern teams plan, build, test, and deploy + software.\\nLet\u2019s explore how Agentic AI is reshaping every stage of + development:\\n1. **Planning and Requirement Analysis**\\nIn traditional software + planning, teams spend extensive time gathering requirements, analyzing dependencies, + and setting milestones. Agentic AI accelerates this process through intelligent + automation.\\nAI Agents can read documentation, project briefs, and performance + data to extract accurate requirements automatically. They detect dependencies, + forecast development timelines, and assess risks using real-time data insights.\\nBy + leveraging AI-driven project analysis, businesses gain faster planning cycles, + better decision-making, and improved resource utilization – all key + to delivering successful, high-quality products.\\n2. **Code Generation and + Review**\\nThe coding stage is where autonomous AI Agents truly revolutionize + AI software development. Unlike basic AI coding assistants, these agents can + generate entire modules, optimize logic, and ensure adherence to industry + standards.\\nThey interpret requirements, create efficient algorithms, and + perform automated code reviews to identify syntax errors, security gaps, or + performance bottlenecks. This not only enhances code quality but also reduces + development time and human dependency.\\nBy integrating AI-powered code generation + and validation, development teams achieve faster release cycles, improved + maintainability, and greater consistency across projects.\\n3. **Testing and + Quality Assurance**\\nAI in software testing has evolved dramatically with + the rise of Agentic AI. Instead of relying on manual QA teams, AI Agents autonomously + execute testing workflows from creating test cases to identifying and fixing + bugs.\\nThese agents continuously learn from past results, improving their + accuracy and predicting potential failures before they occur. They validate + builds, monitor regression, and ensure that every release meets high performance + and reliability standards.\\nThe outcome: shorter testing cycles, higher accuracy, + and more stable software deployments.\\n4. **Deployment and CI/CD Automation**\\nDeployment + is one of the most complex and time-sensitive stages of software delivery. + Autonomous AI agents simplify this through end-to-end[CI/CD automation.] \\nThey + integrate seamlessly with DevOps pipelines, handling build management, environment + setup, and deployment tasks with precision. During live deployments, AI Agents + monitor performance, detect anomalies, and automatically roll back faulty + builds ensuring zero downtime and continuous delivery stability.\\nThis creates + a self-managing DevOps ecosystem that enhances release reliability and operational + efficiency.\\n5. **Continuous Improvement and Maintenance**\\nWith Agentic + AI, software maintenance transforms from a reactive task into a proactive, + self-optimizing process.\\nAI agents constantly monitor application performance, + analyze user behavior, and identify inefficiencies in real time. They recommend + performance optimizations, manage dependency updates, and even handle minor + issue resolutions autonomously.\\nThis continuous learning loop ensures that + your systems evolve intelligently improving scalability, stability, and efficiency + over time without manual intervention.\\nBy adopting Agentic AI solutions, + organizations move beyond static automation into a new era of AI-driven software + development where every stage of the lifecycle is optimized, intelligent, + and self-improving.\\n[SculptSoft], a trusted custom AI software development + company, helps enterprises implement these autonomous AI systems to accelerate + innovation, reduce time-to-market, and achieve higher-quality outcomes in + 2025 and beyond.\\n#### Key Benefits of Agentic AI in Software Development\\nAgentic + AI is transforming software development from a manual, time-consuming process + into a dynamic, intelligent workflow driven by autonomous AI Agents. These + systems don\u2019t just assist developers, they think, plan, and act on their + own, making development faster, smarter, and more efficient.\\nLet\u2019s + look at the key benefits of Agentic AI in software development that are reshaping + how modern businesses build and scale their digital products in 2025.\\n1. + **Accelerated Development Cycles**\\nTraditional development teams spend weeks + (sometimes months) moving from planning to deployment. With autonomous AI + Agents, that process is drastically reduced.\\nAgentic AI automates repetitive + tasks like code generation, testing, and bug fixing so dedicated developers + can focus on strategy and innovation.\",\"image\":\"https://www.sculptsoft.com/wp-content/uploads/2025/10/agentic_ai_in_action_how_autonomous_ai_agents_are_changing_software_development_in_2025.webp\",\"favicon\":\"https://www.sculptsoft.com/wp-content/uploads/2022/05/SculptSoft-Favicon.svg\"},{\"id\":\"https://www.researchgate.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks\",\"title\":\"The + Rise of Autonomous AI Agents: Automating Complex ...\",\"url\":\"https://www.researchgate.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks\",\"publishedDate\":\"2025-07-11T00:00:00.000Z\",\"author\":\"\",\"text\":\"- + [Home] \\n- [Computing] \\n- [Computer Science] \\n- [Autonomics] \\n\\nArticlePDF + Available\\n\\n# The Rise of Autonomous AI Agents: Automating Complex Tasks\\n\\n- + July 2025\\n- [International Journal of Artificial Intelligence for Science + (IJAI4S)] 1(2)\\n\\nDOI: [10.63619/ijai4s.v1i2.007] \\n\\n- License\\n- [CC + BY 4.0] \\n\\nAuthors:\\n\\n[Michael Negnevitsky] \\n\\n[Michael Negnevitsky] + \\n\\n- This person is not on ResearchGate, or hasn't claimed this research + yet.\\n\\n\\n[Download full-text PDF] \\n\\n[Read full-text] \\n\\n[Download + citation] \\n\\nCopy link Link copied\\n\\n[Read full-text] [Download citation] + \\nCopy link Link copied\\n\\n## Abstract and Figures\\n\\nThe emergence of + autonomous AI agents represents a transformative leap in the evolution of + artificial intelligence. These intelligent systems, capable of independently + perceiving environments, making decisions, learning from experience, and executing + multi-step actions without continuous human oversight, are redefining the + boundaries of what machines can accomplish. Unlike traditional rule-based + or supervised AI systems, autonomous agents integrate deep learning, reinforcement + learning, natural language processing, and multi-modal decision frameworks + to solve complex, dynamic, and often ambiguous real-world problems. This paper + explores the technological underpinnings, capabilities, applications, and + implications of autonomous AI agents. It critically examines their deployment + in sectors such as healthcare, finance, cybersecurity, logistics, manufacturing, + education, and scientific research. Furthermore, it addresses the ethical, + legal, and socio-technical challenges arising from the increasing autonomy + of machines, offering a roadmap for responsible innovation. Ultimately, autonomous + AI agents are not merely tools\u2014they are collaborators in a new era of + intelligent automation.\\n\\n[Training and deployment pipeline of autonomous + AI agents, from task specification to safe real-world execution.\\\\\\n\\\\\\n\u2026] + \\n\\n[Human-AI symbiosis: combining human intuition and ethics with AI scalability + and speed to enable collaborative discovery and decision-making.\\\\\\n\\\\\\n\u2026] + \\n\\nFigures - available via license: [Creative Commons Attribution 4.0 International] + \\n\\nContent may be subject to copyright.\\n\\n**Discover the world's research**\\n\\n- + 25+ million members\\n- 160+ million publication pages\\n- 2.3+ billion citations\\n\\n[Join + for free] \\n\\nAvailable via license: [CC BY 4.0] \\n\\nContent may be subject + to copyright.\\n\\nInternationalJournalofArti\uFB01cialIntelligenceforScienceAutomatingComplexTasks\\n\\nTheRiseofAutonomousAIAgents:\\n\\nAutomatingComplexTasks\\n\\nMichaelNegnevitsky1,\u2217\\n\\n1OxfordUniversity,OxfordOX12JD,UK\\n\\nCorrespondingauthor:MichaelNegnevitsky(e-mail:michaelnegnevitsky@gmail.com).\\n\\nDOI:https://doi.org/10.63619/ijai4s.v1i2.007\\n\\nThisworkislicensedunderaCreativeCommonsAttribution4.0InternationalLicense(CCBY4.0).Publishedbythe\\n\\nInternationalJournalofArti\uFB01cialIntelligenceforScience(IJAI4S).\\n\\nManuscriptreceivedMay30,2025;revisedJune13,2025;publishedJuly11,2025.\\n\\nAbstract:TheemergenceofautonomousAIagentsrepresentsatransformativeleapintheevolutionof\\n\\narti\uFB01cialintelligence.Theseintelligentsystems,capableofindependentlyperceivingenvironments,making\\n\\ndecisions,learningfromexperience,andexecutingmulti-stepactionswithoutcontinuoushumanoversight,are\\n\\nrede\uFB01ningtheboundariesofwhatmachinescanaccomplish.Unliketraditionalrule-basedorsupervisedAI\\n\\nsystems,autonomousagentsintegratedeeplearning,reinforcementlearning,naturallanguageprocessing,and\\n\\nmulti-modaldecisionframeworkstosolvecomplex,dynamic,andoftenambiguousreal-worldproblems.This\\n\\npaperexploresthetechnologicalunderpinnings,capabilities,applications,andimplicationsofautonomousAI\\n\\nagents.Itcriticallyexaminestheirdeploymentinsectorssuchashealthcare,\uFB01nance,cybersecurity,logistics,\\n\\nmanufacturing,education,andscienti\uFB01cresearch.Furthermore,itaddressestheethical,legal,andsocio-\\n\\ntechnicalchallengesarisingfromtheincreasingautonomyofmachines,offeringaroadmapforresponsible\\n\\ninnovation.Ultimately,autonomousAIagentsarenotmerelytools\u2014theyarecollaboratorsinaneweraof\\n\\nintelligentautomation.\\n\\nKeywords:AutonomousAIagents,intelligentautomation,reinforcementlearning,multi-agentsystems,task\\n\\nautomation,arti\uFB01cialgeneralintelligence,ethicalAI,autonomousdecision-making,AIplanning,agent-based\\n\\nmodeling.\\n\\n1.Introduction\\n\\nThe21stcenturyhaswitnessedunprecedentedadvancementsinarti\uFB01cialintelligence(AI),transforming\\n\\nindustries,economies,anddailylife.Amongtheseinnovations,theemergenceofautonomousAIagents\\n\\nmarksaparadigmshift\u2014notmerelyincomputationalcapability,butinthedelegationofcomplexcognitive\\n\\ntaskstomachines\\\\[1\\\\].Theseagents,whichcanindependentlyperceiveenvironments,makecontext-aware\\n\\ndecisions,learnfromexperience,andexecutegoal-directedactions,arerapidlyrede\uFB01ningwhatconstitutes\\n\\nautomationinthemodernworld\\\\[?\\\\].\\n\\nUnliketraditionalnarrowAIsystemsthatoperatewithinstatic,rule-basedframeworksorrequire\\n\\ncontinuoushumanoversight,autonomousagentsexhibitahighdegreeofautonomy,adaptability,and\\n\\ngeneralization.Theyarecapableofreal-timereasoning,dynamicplanning,andlifelonglearninginopen-\\n\\nended,unpredictableenvironments\\\\[2\\\\],\\\\[3\\\\].Forinstance,self-drivingvehiclesnavigatechaotictraf\uFB01c,\\n\\nroboticsurgeonsmakeintra-operativedecisions,andlanguageagentsengageincomplex,multi-turndia-\\n\\nlogues\u2014allwithminimalornohumanintervention\\\\[4\\\\].Thesedevelopmentsrepresentnotjustengineering\\n\\nmilestones,buttheinitialfoundationsofarti\uFB01cialgeneralintelligence(AGI)\u2014aformofintelligencethat\\n\\ncan\uFB02exiblyperformawiderangeofcognitivetasksacrossdomains\\\\[5\\\\],\\\\[6\\\\].\\n\\nTheriseofautonomousagentsalsore\uFB02ectsabroaderconvergenceofAIsub\uFB01elds,includingdeep\\n\\nreinforcementlearning,multi-agentsystems,neuro-symbolicreasoning,andlargelanguagemodels\\\\[7\\\\],\\n\\n\\\\[?\\\\].Theseintegrationshaveenabledagentstohandlenotonlyphysicaltasksinroboticsandlogistics,\\n\\nVol.01,No.02,June2025Page1\\n\\nInternationalJournalofArti\uFB01cialIntelligenceforScienceAutomatingComplexTasks\\n\\nbutalsoabstractreasoningtaskssuchaslegaldocumentdrafting,scienti\uFB01chypothesisgeneration,and\\n\\nadaptiveeducationdelivery.\\n\\nHowever,thistechnologicalleapalsobringsprofoundsocietalandethicalchallenges\\\\[?\\\\].Delegating\\n\\ndecisionstonon-humanentitiesraisescriticalconcerns:Howdoautonomousagentslearn,adapt,andmake\\n\\ndecisionsinhigh-stakesenvironments?Whatdomainsaretheymostsuitedfor\u2014andwhereshouldhuman\\n\\ncontrolremaincentral?Howcanweensuretheseagentsbehaveinwaysalignedwithhumanvalues,\\n\\nespeciallywhentheiractionsaffectsafety,justice,orequity?Whatregulatory,technical,andgovernance\\n\\nframeworksarerequiredtomanagethedeploymentofsuchintelligentsystems?\\n\\nThispaperoffersacomprehensiveexaminationofthearchitecture,applications,trainingmethodology,\\n\\nandethicalimplicationsofautonomousAIagents.Throughillustrativeusecases,comparativeanalyses,\\n\\nandfutureoutlooks,weaimtounderstandnotonlywhatthesesystemscando,butalsowhattheyshould\\n\\ndo\u2014asintelligentcollaboratorsinaworldincreasinglyshapedbymachineagency.\\n\\n2.TheArchitectureofAutonomousAgents\\n\\nAutonomousagentsareintelligentsystemscapableofperceivingtheirenvironment,makingdecisions,\\n\\ntakingactions,andadaptingovertimewithoutcontinuoushumanintervention\\\\[8\\\\].Theirarchitectureistyp-\\n\\nicallyorganizedintomodularandhierarchicallayers,eachresponsiblefordistinctaspectsoffunctionality\\n\\n\\\\[9\\\\].Thislayeredapproachenhancesinterpretability,modulardevelopment,andscalability.Wedescribethe\\n\\nfourprimarylayersofanautonomousagentsystem:perception,cognition,action,andmemory/adaptation\\n\\n\\\\[10\\\\].\\n\\nFig.1.LayeredarchitectureofanautonomousAIagentsystem,illustratingperception,cognition,action,and\\n\\nmemorycomponents.\\n\\nVol.01,No.02,June2025Page2\\n\\nInternationalJournalofArti\uFB01cialIntelligenceforScienceAutomatingComplexTasks\\n\\n2.1.PerceptionLayer\\n\\nTheperceptionlayerservesasthesensoryinterfacebetweentheagentanditsenvironment.Ittransforms\\n\\nrawdataintostructuredrepresentationsthathigher-levelmodulescaninterpretandreasonover\\\\[11\\\\].\\n\\n\u2022ComputerVision:Enablestheagenttounderstandvisualinput,includingobjectdetection,scene\\n\\nsegmentation,motiontracking,andspatiallayoutanalysis.Forexample,adronemayidentifyroads,\\n\\nhumans,orwildlifeusingYOLOorMaskR-CNN.\\n\\n\u2022NaturalLanguageProcessing(NLP):Allowstheagenttointerprettextualorspokeninstructions,\\n\\nconductdialogue,andextractsemanticmeaning.Applicationsincludelanguage-guidednavigation\\n\\nandcollaborativetaskexecutionwithhumans.\\n\\n\u2022SensorFusion:Combinesdatafrommultiplemodalities\u2014e.g.,LiDAR,RGBcameras,thermal\\n\\nsensors,radar,andmicrophones\u2014tobuildarobustandredundantperceptionsystemthatimproves\\n\\naccuracyinuncertainenvironments.\\n\\nThislayerensurestheagenthasacoherent,real-timeunderstandingofitssurroundings.\\n\\n2.2.CognitiveLayer\\n\\nThecognitivelayeristhe\u201Cbrain\u201Doftheagent.Itinterpretssensoryinputs,generatesinternalgoals,reasons\\n\\naboutconsequences,andchoosesactions\\\\[12\\\\].\\n\\n\u2022ReinforcementLearning(RL):Enablesagentstolearnoptimalpoliciesthroughtrial-and-error\\n\\ninteractionwiththeenvironment.Thisiswidelyusedinautonomousdriving,game-playing,and\\n\\nroboticcontrol.\\n\\n\u2022Meta-learning:Alsoknownas\u201Clearningtolearn,\u201Dthisallowsagentstorapidlyadapttonewtasks\\n\\norenvironmentswithminimaldata,enhancingtheirgeneralizationcapabilities.\\n\\n\u2022PlanningandScheduling:ClassicalAItechniquessuchasA\\\\*,MonteCarloTreeSearch(MCTS),\\n\\norPDDL-basedplannersareusedtogeneratemulti-stepactionplansunderconstraints.\\n\\n\u2022Neural-SymbolicIntegration:Combinesneuralnetworks(forperceptionandlearning)withsymbolic\\n\\nreasoning(e.g.,logicrules,knowledgegraphs)toachieveboth\uFB02exibilityandinterpretability.\\n\\nTogether,thesetechniquesenabletheagenttomakeinformed,strategicdecisionsindynamicenviron-\\n\\nments.\\n\\n2.3.ActionLayer\\n\\nOnceadecisionhasbeenmade,theactionlayerisresponsibleforphysicallyorvirtuallyexecutingthat\\n\\ndecision\\\\[13\\\\].Itactsastheinterfacebetweencognitionandtheexternalworld.\\n\\n\u2022RoboticActuation:Inembodiedagents,thisinvolvesmotorcommandstomanipulators,drones,or\\n\\nvehicles,enablinglocomotion,manipulation,andinteractionwithobjects.\\n\\n\u2022API-basedExecution:Insoftwareagents(e.\",\"image\":\"https://i1.rgstatic.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks/links/68715efc6e247f362b18bd9f/largepreview.png\",\"favicon\":\"https://c5.rgstatic.net/m/42199702882742/images/favicon/favicon-32x32.png\"},{\"id\":\"https://www.ml-science.com/blog/2025/4/17/developments-in-ai-agents-q1-2025-landscape-analysis\",\"title\":\"Developments + in AI Agents: Q1 2025 Landscape Analysis \u2014 The Science of Machine Learning + & AI\",\"url\":\"https://www.ml-science.com/blog/2025/4/17/developments-in-ai-agents-q1-2025-landscape-analysis\",\"publishedDate\":\"2025-04-17T00:00:00.000Z\",\"author\":\"\",\"text\":\"Developments + in AI Agents: Q1 2025 Landscape Analysis — The Science of Machine Learning + & AI\\n# [The Science of Machine Learning & AI] \\nMathematics - Data + Science - Computer Science\\n[**Blog**] **Special**:[**The Accelerating Evolution + of Artificial Intelligence**] \\n[ARC-AGI MODELS LEADERBOARD] \\nCopyright + \xA92016-2026[Don Cowan] All Rights Reserved\\nMathematical Notation Powered + by[CodeCogs] \\n[![Creative Commons License]] \\nThis work is licensed under + a[Creative Commons Attribution 4.0 International License].\\n[Blog RSS] \\n[\\n![] + \\n] \\n[About the Author] \\n[\\n] \\n![] \\n# [Developments in AI Agents: + Q1 2025 Landscape Analysis] \\n[April 29, 2025] in[Artificial Intelligence] + \\n## **Executive Summary**\\nThe first quarter of 2025 marked a period of + intense activity and significant evolution in the field of Artificial Intelligence + (AI) agents.\\n* Characterized by a**rapid transition from research concepts + to tangible products**, this timeframe solidified the notion of 2025 as a + pivotal \\\"year of the agent,\\\" albeit one accompanied by considerable + hype and persistent challenges. (1)\\n* The definition of**AI agents increasingly + centered on systems, typically built upon Large Language Models (LLMs)**, + capable of autonomous task execution through planning, reasoning, and the + crucial ability to utilize external tools. (4)\\n* **Major technology companies, + including OpenAI, Google, Microsoft, Anthropic, and Meta, accelerated the + push towards commercialization**, unveiling a suite of new agent platforms, + specialized agents targeting specific workflows (e.g., software development, + sales, security, research), and underlying infrastructure components. (7)\\n* + **Research simultaneously advanced**, focusing on enhancing core capabilities + like reasoning, fostering human-AI collaboration, and developing multi-agent + systems (MAS) capable of complex interactions. (12)\\n* Concurrently,**the + discourse surrounding the challenges, limitations, and ethical implications + intensified**. Concerns regarding reliability, safety, control, bias, and + privacy became more prominent as agent autonomy increased, prompting calls + for robust governance frameworks and dedicated agent infrastructure. (16)\\n* + While the potential for AI agents to automate tasks and augment human capabilities + is increasingly evident,**significant hurdles in achieving dependable, safe, + and ethically aligned performance remained a primary focus**for the industry + moving forward. (20)\\n## **Defining the AI Agent Landscape (Early 2025)**\\n### + **Conceptual Evolution and Definitions**\\nThe understanding and definition + of AI agents underwent a notable transformation leading into early 2025, largely + catalyzed by advancements in LLMs. Historically, the concept of an agent was + rooted in computer science and cognitive science, often defined broadly as + anything capable of perceiving its environment through sensors and acting + upon it through actuators. Early work emphasized properties like autonomy, + social ability, reactivity, and proactivity. (4)\\nHowever, by the first quarter + of 2025, the prevailing conceptualization became intrinsically linked to the + capabilities offered by modern LLMs. (4)\\nWhile no single, universally accepted + definition existed (24), a strong consensus emerged viewing AI agents as systems + or programs capable of*autonomously performing tasks*on behalf of a user or + another system. This autonomy involved designing their own workflows and utilizing + available tools to achieve specified goals. (1)\\nThis modern understanding + clearly differentiated AI agents from simpler AI applications like chatbots + or traditional automation. Unlike chatbots primarily focused on conversational + responses based on pre-existing data (6), or traditional systems executing + predefined algorithms within constraints (4), early 2025 agents were characterized + by their ability to plan, reason, learn, and interact dynamically with their + environment to accomplish complex, often multi-step objectives. (4) They leveraged + LLMs as core reasoning or processing components but augmented them with essential + capabilities like memory, planning modules, and mechanisms for tool use. (4) + This augmentation was crucial, addressing inherent LLM limitations such as + statelessness (lack of memory between interactions) and the inability to directly + interact with or affect the external world. (17) The term \\\"agentic AI\\\" + also gained traction, often referring to systems composed of multiple, potentially + collaborative, AI agents working towards a common objective. (28)\\nThe evolution + of the definition directly mirrors the enabling power of LLMs. While the theoretical + foundations of agency existed for decades (4), the practical ability of LLMs + to understand complex instructions, reason about tasks, and generate plausible + action steps provided the foundation for building agents that could move beyond + theoretical constructs and execute tasks in the real world (primarily digital + environments). The focus shifted from defining agency based on abstract properties + to defining it based on the functional capabilities enabled by integrating + LLMs with planning, memory, and tool-using mechanisms. This practical turn, + driven by LLM advancements, underpins the surge in agent development and deployment + observed in early 2025.\\n### **Core Capabilities**\\nThe AI agents emerging + in early 2025 were defined by a set of core capabilities that enabled their + autonomous and goal-directed behavior. These capabilities, often working in + concert and built upon LLM foundations, distinguished them from prior AI systems:\\n* + **Autonomy:**This was perhaps the most central characteristic. Agents were + expected to operate with minimal human intervention or continuous oversight, + capable of initiating and completing tasks independently.1 The degree of autonomy + existed on a spectrum. Some systems were semi-autonomous, requiring human + oversight for critical decisions (26), while others aimed for, or raised concerns + about, higher levels of autonomy, potentially even \\\"fully autonomous\\\" + operation where agents could act without predefined constraints, a prospect + met with significant safety concerns. (18)\\n* **Learning & Adaptation:**Agents + were designed to improve their performance over time. This involved learning + from environmental feedback, past interactions, and accumulated experience.3 + Mechanisms like reinforcement learning allowed agents to refine strategies + based on outcomes. (26) The ability to self-evaluate and correct errors was + also highlighted. (32) Memory, both short-term (for maintaining context within + a task) and long-term (for retaining historical data and preferences), was + crucial for adaptation and personalization. (26)\\n* **Reasoning & Decision-Making:**Agents + needed to process perceived information, analyze data, make judgments, and + select appropriate actions to achieve their goals. (1) LLMs often served as + the core \\\"reasoning engine\\\" (32), interpreting inputs and generating + potential action plans. The period saw the emergence of specialized models + explicitly optimized for reasoning tasks, aiming to enhance agent decision-making + capabilities. (12)\\n* **Interaction:**Agents were defined by their ability + to engage with their environment. This involved perception \u2013gathering + information through various inputs like text, voice, images, sensors, data + feeds, or APIs (1) \u2013and action \u2013influencing the environment through + outputs like generating text, sending commands to software, executing code, + or making API calls. (1) Crucially, this interaction extended beyond simple + user dialogue to include engagement with external systems, tools, data sources, + and potentially other agents or humans. (1)\\n* **Planning & Goal-Directedness:**Agents + were designed to be proactive and goal-oriented. (24) This required the ability + to understand a high-level objective, decompose it into smaller, manageable + steps (task decomposition), formulate a sequence of actions (planning), and + then execute that plan. (5) The capability for long-term planning, reasoning + about extended sequences of actions, was also an area of development and concern. + (12)\\n* **Tool Usage:**A key enabler of agent functionality was the ability + to utilize external tools. These could include APIs for accessing external + services (like weather data, booking systems, or databases), web search capabilities, + code interpreters, calculators, or even invoking other AI models or agents. + (4) Tool use allowed agents to overcome the inherent limitations of their + base LLMs (e.g., accessing real-time information, performing precise calculations, + interacting with specific software) and act effectively in the environment. + (6)\\nThe confluence of these capabilities defined the ambition for AI agents + in early 2025. The objective was clear: to transition AI from being passive + responders or pattern recognizers to becoming active, autonomous participants + capable of understanding goals, formulating plans, interacting with the digital + world through tools, learning from outcomes, and ultimately achieving complex + objectives with reduced human guidance. This represented a significant functional + leap, enabling the automation of tasks previously requiring human cognition + and intervention.\\n### **C. Agent Architectures and Components**\\nThe design + of AI agents in early 2025 commonly followed specific architectural patterns, + reflecting attempts to harness the power of LLMs while mitigating their weaknesses. + Agents were frequently described as*compound systems*. (24) This meant they + were not monolithic entities but rather architectures built around a foundation + model (typically an LLM) augmented by external resources or modules, often + referred to as \\\"scaffolding\\\". (5) This scaffolding provided capabilities + the base LLM lacked, such as persistent memory, structured planning, and interfaces + for tool\",\"image\":\"http://static1.squarespace.com/static/5acbdd3a25bf024c12f4c8b4/t/68111511dca49a402d94e879/1745949980281/AI+Agents.png?format=1500w\",\"favicon\":\"https://images.squarespace-cdn.com/content/v1/5acbdd3a25bf024c12f4c8b4/1600024348668-O52EU3A2Y2J0LZUUTBMZ/favicon.ico?format=100w\"},{\"id\":\"https://news.constructconnect.com/2025-buildings-show-ai-agents-become-talk-of-the-town\",\"title\":\"2025 + Buildings Show: AI agents Become 'Talk of the Town'\",\"url\":\"https://news.constructconnect.com/2025-buildings-show-ai-agents-become-talk-of-the-town\",\"publishedDate\":\"2025-12-11T00:00:00.000Z\",\"author\":\"Don + Wall\",\"text\":\"2025 Buildings Show: AI agents Become \u2018Talk of the + Town\u2019\\n[![ConstructConnect News]] \\n[Follow us on LinkedIn] [Watch + us on YouTube] [Follow us on Instagram] [Follow us on Facebook] [Follow us + on X] \\n[Find Projects Now] \\n[![ConstructConnect News]] \\n****[****] \\n## + [![ConstructConnect News]] \\n[Follow us on LinkedIn] [Watch us on YouTube] + [Follow us on Instagram] [Follow us on Facebook] [Follow us on X] \\n[\u2715] + \\n* [Featured] \\n* [Economy] \\n* [Industry] \\n* [Project Spotlight] \\n* + [About] \\n[**] \\n[Industry News & Trends] [Featured] # 2025 Buildings + Show: AI agents Become \u2018Talk of the Town\u2019\\nBy Don Wall**December + 11, 2025\\nShare\\n* [\\n] \\n* [\\n] \\n* [\\n] \\n#### KEY POINTS\\n* **Agentic + AI is revolutionizing construction by enabling autonomous systems to plan, + monitor, and update tasks, offering multi-step reasoning and decision-making + beyond traditional AI workflows and Large Language Models (LLMs).**\\n* **Tools + simplify software development by generating, modifying, and testing code from + written requirements, reducing manual coding time and empowering non-coders + to create apps and automations.**\\n* **While agentic AI accelerates innovation, + one expert suggested that users must maintain a \u201Chuman in the loop.\u201D**\\nWhat + a difference a year makes in the world of[construction tech].\\nDelegates + attending day two of the recent Buildings Show in Toronto heard how agentic + AI is rapidly emerging as construction\u2019s third wave of[AI], moving beyond + Large Language Models (LLMs) and[automated workflows] to create a world of + autonomous systems that plan, monitor, and update[project tasks], as well + aswrite code.\\nDigital innovation specialist Mahir Dheendsa of[Amrize] (formerly + Lafarge), host of a session titled \u201CAI Agents and the Robotic Workforce: + Reimagining the Future of Construction,\u201Dsaid the use of agentic AI is + accelerating. So rapidly that the preview of the session he wrote for the + show program months ago was already out of date.\\n![DW-buildings-show-AI-agencyweb] + \\nDigital innovation specialist Mahir Dheendsa of Amrize spoke about AI agents + at the Buildings Show in Toronto, Dec. 4. Dheendsa took a moment to outline + how Amrize represents the new brand for Lafarge. Image: Don Wall\\n\u201CAgentic + AI has been the talk of the town and the major headline in the AI space this + year,\u201D said Dheendsa. \u201CWe haven\u2019t really seen any major advancements + in robotics, so I\u2019m willing to sort of skip over that.\u201D\\nPlatforms + offering AI agents, such as n8n, Gemini Agent Studio, Make.com and Zapier, + democratize agent development, Dheendsa said, enabling[construction professionals] + to build custom automations without extensive coding expertise.\\nGenerative + development tools like Lovable AI and Cursor take it a step further by generating, + modifying, and testing software directly from written requirements, dramatically[reducing + the time] typically spent on planning and manual coding.\\n[![Econ Ads_Banners3.jpg.-2]] + \\n#### Multi-Step Reasoning\\nLLMs offer a single-step response with no tools + or actions, he explained. AI workflows follow a fixed sequence of steps and + can connect multiple apps, but with no reasoning or judgment. An AI agent + acts as autonomous intelligence, utilizing multi-step reasoning and tools + and APIs to evaluate, retry, and create.\\nDheendsa offered an example of + a supply chain coordination function with a problem: a crew of 20 is standing + around doing nothing because the steel beams haven\u2019t arrived.\\nThe agent + can track supplier GPS data and manufacturing status updates in real-time, + potentially predicting a delay three days in advance and suggesting rescheduling + the crane rental.\\nAnother example is with a change order. A[subcontractor] + might send bills for \u201Cextra work,\u201D and nobody remembers if it was + approved, said Dheendsa. The agent scans thousands of emails, contracts, and + meeting minutes automatically and verifies if the extra work was actually + part of the original contract scope.\\n\u201CAn AI agent has the ability to + make decisions,\u201D said Dheendsa. \u201CThe[AI agent] monitors and reports. + It will detect an issue, and it will decide if this is an issue worthy of + being brought up based on a criteria that you can provide it.\\n\u201CAnd + then it will figure out which team or which person needs to be notified.\u201D\\nHe + has used Replit, Lovable, and Cursor, and all can work to generate responses + to RFPs, he said.\\n\u201CYou just copy the[RFP] into the Lovable app, and + the software is done,\u201D said Dheendsa.\\n\u201CIf you have an idea, if + you want to build an app, you don\u2019t even need to know how to code. You + just put your idea in there, and it\u2019ll just generate you an app or a + web app, website, whatever you need.\u201D\\nHuman in the Loop\\nDheendsa + stressed that users still need to keep a \u201Chuman in the loop.\u201D\\n\u201CIt + is prone to making mistakes. It is prone to hallucinations, (but) it\u2019s + getting better.\u201D\\nWith Gemini 3, for example, \u201CI\u2019d say it\u2019s + 85 per cent there. It\u2019s very good.\u201D\\nAnother caveat is that users + should not venture into AI solutions if the original process is flawed: \u201CDon\u2019t + automate broken processes,\u201D Dheendsa said.\\nAmong various risks he identified, + \u201Cover-trust risk\u201D is highly dangerous, he said. Al confidence does + not equal correctness. Always verify technical outputs and cross-check against + standards and codes, he said.\\nMicrosoft has reported[construction jobs] + in the field are among the least threatened by AI, Dheendsa noted in an interview, + for a couple of reasons.\\nFor one, dredge operators or crane operators represent + a lot of \u201Cmoving parts.\u201D Second, he said, there hasn\u2019t been + enough data collected in the field for[AI to automate].\\n\u201CThe low-hanging + fruit is here with the administrative stuff, like data entry and all the[repetitive + workflows]. So let\u2019s just start there,\u201D said Dheendsa.\\n\u201CThe + barrier to entry is lower than ever before\u2026I encourage people to start + doing that, because even if it\u2019s not going to get you 100 per cent of + the way there, it\u2019ll at least get you 80 per cent.\u201D\\n#### About + ConstructConnect\\nAt[ConstructConnect], our software solutions provide the + information construction professionals need to start every project on a solid + foundation. For more than 100 years, our insights and market intelligence + have empowered commercial firms, manufacturers, trade contractors, and architects + to make data-driven decisions and maximize productivity.\\nConstructConnect + is a business unit of[Roper Technologies] (Nasdaq: ROP), part of the Nasdaq + 100, S&P 500, and Fortune 1000.\\nFor more information, visit[constructconnect.com] + \\n![Don Wall] \\n[Don Wall] \\nDon Wall is a freelance journalist that covers + construction including economics, architecture, resources, infrastructure, + health and safety and new projects. He has worked for over 35 years as a journalist, + including as a staff writer for over nine years with Canada's Daily Commercial + News by ConstructConnect.\\n[Previous] \\n### [$150M Deal Fails at HyperGrid + AI Project] \\n[Next] \\n### [Clarksville, Tennessee's Business Boom Grows + with New $90 Million Investment] \\n### Subscribe Now.\\nConstructConnect + News. Construction news. Built for you, daily.\\n### ### Topics\\n* [Industry + News & Trends] \\n* [Featured] \\n* [Economy] \\n* [Project Spotlight] + \\n* [Construction Starts] \\n* [Tariffs] \\n* [Construction Starts Forecast] + \\n* [Project Stress Index] \\n* [Building Product Manufacturing] \\n* [Construction + Economics & Finance] \\n#### Trending News\\n[![image]] \\n### [$150M + Deal Fails at HyperGrid AI Project] \\n[![image]] \\n### [2025 Buildings Show: + AI agents Become \u2018Talk of the Town\u2019] \\n[![image]] \\n### [Clarksville, + Tennessee's Business Boom Grows with New $90 Million Investment] \\n#### Trending + News\\n1. [![image]] \\n### [$150M Deal Fails at HyperGrid AI Project] \\n2. + [![image]] \\n### [2025 Buildings Show: AI agents Become \u2018Talk of the + Town\u2019] \\n3. [![image]] \\n### [Clarksville, Tennessee's Business Boom + Grows with New $90 Million Investment] \\n#### RELATED NEWS\\n[![image]] [Project + Spotlight] \\n### [Wastewater Facility in Vero Beach to Break Ground] \\nConstruction + on a new $164 million wastewater treatment facility at Vero Beach, Florida\u2019s + regional...\\nBy[Marshall Benveniste] \\nMay 29 2025\\n[![image]] [Featured] + \\n### [The More Things Change Portion of the Construction Marketplace] \\nAlex + Carrick explores the changes in nonresidential construction as new sources + of demand for...\\nBy[Alex Carrick] \\nMay 14 2025\\n[![image]] [Industry + News & Trends] \\n### [Autumn Construction Starts Forecast: Civil Construction + to Lead, Megaprojects Lift Outlook] \\nIn our Autumn 2025 Construction Starts + Forecast, ConstructConnect has slightly upgraded our 2025...\\nBy[Michael + Guckes, Chief Economist] \\nAug 19 2025\",\"image\":\"https://news.constructconnect.com/hubfs/AI%20agent%20supporting%20business%20processes%20with%20automation%20technology.%20Generative%20artificial%20intelligence%20analyzing%20data%20and%20predicting%20future.%20Businessman%20using%20laptop..jpg\",\"favicon\":\"https://news.constructconnect.com/hubfs/assets/images/ccfav.png\"},{\"id\":\"https://arxiv.org/abs/2502.07056\",\"title\":\"Autonomous + Deep Agent\",\"url\":\"https://arxiv.org/abs/2502.07056\",\"publishedDate\":\"2025-02-10T00:00:00.000Z\",\"author\":\"[Submitted + on 10 Feb 2025]\",\"text\":\"[2502.07056] Autonomous Deep Agent\\n[Skip to + main content] \\n[![Cornell University]] \\nWe gratefully acknowledge support + from the Simons Foundation,[member institutions], and all contributors.[Donate] + \\n[] \\n[![arxiv logo]] >[cs] >arXiv:2502.07056\\n[Help] |[Advanced + Search] \\nAll fieldsTitleAuthorAbstractCommentsJournal referenceACM classificationMSC + classificationReport numberarXiv identifierDOIORCIDarXiv author IDHelp pagesFull + text\\nSearch\\n[![arXiv logo]] \\n[![Cornell University Logo]] \\nopen search\\nGO\\nopen + navigation menu\\n# Computer Science \\\\> Artificial Intelligence\\n**arXiv:2502.07056**(cs)\\n[Submitted + on 10 Feb 2025]\\n# Title:Autonomous Deep Agent\\nAuthors:[Amy Yu],[Erik Lebedev],[Lincoln + Everett],[Xiaoxin Chen],[Terry Chen] \\nView a PDF of the paper titled Autonomous + Deep Agent, by Amy Yu and 3 other authors\\n[View PDF] [HTML (experimental)] + > > Abstract:\\n> This technical brief introduces Deep Agent, an advanced + autonomous AI system designed to manage complex multi-phase tasks through + a novel hierarchical task management architecture. The system's foundation + is built on our Hierarchical Task DAG (HTDAG) framework, which dynamically + decomposes high-level objectives into manageable sub-tasks while rigorously + maintaining dependencies and execution coherence. Deep Agent advances beyond + traditional agent systems through three key innovations: First, it implements + a recursive two-stage planner-executor architecture that enables continuous + task refinement and adaptation as circumstances change. Second, it features + an Autonomous API & Tool Creation (AATC) system that automatically generates + reusable components from UI interactions, substantially reducing operational + costs for similar tasks. Third, it incorporates Prompt Tweaking Engine and + Autonomous Prompt Feedback Learning components that optimize Large Language + Model prompts for specific scenarios, enhancing both inference accuracy and + operational stability. These components are integrated to form a service infrastructure + that manages user contexts, handles complex task dependencies, and orchestrates + end-to-end agentic workflow execution. Through this sophisticated architecture, + Deep Agent establishes a novel paradigm in self-governing AI systems, demonstrating + robust capability to independently handle intricate, multi-step tasks while + maintaining consistent efficiency and reliability through continuous self-optimization. + Subjects:|Artificial Intelligence (cs.AI); Machine Learning (cs.LG)|\\nACMclasses:|I.2.6; + I.2.7|\\nCite as:|[arXiv:2502.07056] [cs.AI]|\\n|(or[arXiv:2502.07056v1] [cs.AI]for + this version)|\\n|[https://doi.org/10.48550/arXiv.2502.07056] \\nFocus to + learn more\\narXiv-issued DOI via DataCite\\n|\\n## Submission history\\nFrom: + Amy Yu [[view email]]\\n**[v1]**Mon, 10 Feb 2025 21:46:54 UTC (2,085 KB)\\nFull-text + links:## Access Paper:\\nView a PDF of the paper titled Autonomous Deep Agent, + by Amy Yu and 3 other authors\\n* [View PDF] \\n* [HTML (experimental)] \\n* + [TeX Source] \\n[![license icon] view license] \\nCurrent browse context:\\ncs.AI\\n[<<prev] + | [next>>] \\n[new] |[recent] |[2025-02] \\nChange to browse by:\\n[cs] + \\n[cs.LG] \\n### References & Citations\\n* [NASA ADS] \\n* [Google Scholar] + \\n* [Semantic Scholar] \\nexport BibTeX citationLoading...\\n## BibTeX formatted + citation\\n×\\nloading...\\nData provided by:\\n### Bookmark\\n[![BibSonomy + logo]] [![Reddit logo]] \\nBibliographic Tools\\n# Bibliographic and Citation + Tools\\nBibliographic Explorer Toggle\\nBibliographic Explorer*([What is the + Explorer?])*\\nConnected Papers Toggle\\nConnected Papers*([What is Connected + Papers?])*\\nLitmaps Toggle\\nLitmaps*([What is Litmaps?])*\\nscite.ai Toggle\\nscite + Smart Citations*([What are Smart Citations?])*\\nCode, Data, Media\\n# Code, + Data and Media Associated with this Article\\nalphaXiv Toggle\\nalphaXiv*([What + is alphaXiv?])*\\nLinks to Code Toggle\\nCatalyzeX Code Finder for Papers*([What + is CatalyzeX?])*\\nDagsHub Toggle\\nDagsHub*([What is DagsHub?])*\\nGotitPub + Toggle\\nGotit.pub*([What is GotitPub?])*\\nHuggingface Toggle\\nHugging Face*([What + is Huggingface?])*\\nLinks to Code Toggle\\nPapers with Code*([What is Papers + with Code?])*\\nScienceCast Toggle\\nScienceCast*([What is ScienceCast?])*\\nDemos\\n# + Demos\\nReplicate Toggle\\nReplicate*([What is Replicate?])*\\nSpaces Toggle\\nHugging + Face Spaces*([What is Spaces?])*\\nSpaces Toggle\\nTXYZ.AI*([What is TXYZ.AI?])*\\nRelated + Papers\\n# Recommenders and Search Tools\\nLink to Influence Flower\\nInfluence + Flower*([What are Influence Flowers?])*\\nCore recommender toggle\\nCORE Recommender*([What + is CORE?])*\\n* Author\\n* Venue\\n* Institution\\n* Topic\\nAbout arXivLabs\\n# + arXivLabs: experimental projects with community collaborators\\narXivLabs + is a framework that allows collaborators to develop and share new arXiv features + directly on our website.\\nBoth individuals and organizations that work with + arXivLabs have embraced and accepted our values of openness, community, excellence, + and user data privacy. arXiv is committed to these values and only works with + partners that adhere to them.\\nHave an idea for a project that will add value + for arXiv's community?[**Learn more about arXivLabs**].\\n[Which authors of + this paper are endorsers?] |[Disable MathJax] ([What is MathJax?])\",\"image\":\"/static/browse/0.3.4/images/arxiv-logo-fb.png\",\"favicon\":\"https://arxiv.org/static/browse/0.3.4/images/icons/favicon-32x32.png\"}],\"searchTime\":1848.5,\"costDollars\":{\"total\":0.015,\"search\":{\"neural\":0.005},\"contents\":{\"text\":0.01}}}" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json; charset=utf-8 + Date: + - Fri, 06 Feb 2026 21:27:38 GMT + Nel: + - '{"report_to":"cf-nel","success_fraction":0.0,"max_age":604800}' + Report-To: + - '{"group":"cf-nel","max_age":604800,"endpoints":[{"url":"https://a.nel.cloudflare.com/report/v4?s=8qRdrIdAWlhoTFMDBageehJCq5TNEpeHiORSh3hYm2ZljbGfpnZ58qytlo0rgVknX5vUadzLwyoRcNJGxX2KRunh4q7LH%2BtWLA%3D%3D"}]}' + Server: + - cloudflare + Transfer-Encoding: + - chunked + access-control-allow-credentials: + - 'true' + cf-cache-status: + - DYNAMIC + content-security-policy: + - CSP-FILTERED + cross-origin-opener-policy: + - same-origin + cross-origin-resource-policy: + - same-origin + etag: + - ETAG-XXX + origin-agent-cluster: + - ?1 + referrer-policy: + - REFERRER-POLICY-XXX + strict-transport-security: + - STS-XXX + vary: + - Origin + x-content-type-options: + - X-CONTENT-TYPE-XXX + x-dns-prefetch-control: + - 'off' + x-download-options: + - noopen + x-frame-options: + - X-FRAME-OPTIONS-XXX + x-permitted-cross-domain-policies: + - X-PERMITTED-XXX + x-ratelimit-limit: + - '450' + x-ratelimit-remaining: + - '428' + x-ratelimit-reset: + - '1770413257' + x-xss-protection: + - X-XSS-PROTECTION-XXX + status: + code: 200 + message: OK +- request: + body: "{\"messages\":[{\"role\":\"system\",\"content\":\"You are Research Analyst. + You are a research analyst who searches the web for information, identifies + key findings, and produces structured research summaries.\\n\\nYour goal: Research + topics using search tools and produce structured summaries\\n\\nYou are executing + a specific step in a multi-step plan. Focus ONLY on completing\\nthe current + step. Do not plan ahead or worry about future steps.\\n\\nBefore acting, briefly + reason about what you need to do and which approach\\nor tool would be most + helpful for this specific step.\"},{\"role\":\"user\",\"content\":\"## Current + Step\\nConduct research on recent developments in autonomous AI agents in 2025.\\n\\nSuggested + tool: exa_search_tool\\n\\nComplete this step and provide your result.\"},{\"role\":\"assistant\",\"content\":null,\"tool_calls\":[{\"id\":\"call_NpXIbiQ1fcI2sBA84tVHMIRu\",\"type\":\"function\",\"function\":{\"name\":\"exa_search_tool\",\"arguments\":\"{\\\"search_query\\\":\\\"autonomous + AI agents developments 2025\\\",\\\"start_published_date\\\":\\\"2025-01-01\\\",\\\"end_published_date\\\":\\\"2025-12-31\\\",\\\"include_domains\\\":null}\"}}]},{\"role\":\"tool\",\"tool_call_id\":\"call_NpXIbiQ1fcI2sBA84tVHMIRu\",\"name\":\"exa_search_tool\",\"content\":\"Title: + AI Agents in 2025: Expectations vs. Reality\\nURL: https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality\\nID: + https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality\\nScore: + None\\nPublished Date: 2025-02-26T00:00:00.000Z\\nAuthor: None\\nImage: https://www.ibm.com/content/dam/connectedassets-adobe-cms/worldwide-content/gbs/cf/ul/g/5a/88/336.jpg/_jcr_content/renditions/cq5dam.thumbnail.1280.1280.png\\nFavicon: + https://www.ibm.com/content/dam/adobe-cms/default-images/icon-32x32.png\\nExtras: + None\\nSubpages: None\\nText: AI Agents in 2025: Expectations vs. Reality | + IBM\\nTags\\n[Artificial Intelligence] \\n# AI agents in 2025: Expectations + vs. reality\\n![the same person doing multiple jobs] \\n## Authors\\n[Ivan Belcic] + \\nStaff writer\\n[Cole Stryker] \\nStaff Editor, AI Models\\nIBM Think\\n## + AI agents in 2025: Expectations vs. reality\\nIt\u2019s impossible to take two + steps across the tech media landscape without stumbling over an article hailing + 2025 as the year of the[AI agent]. Agents, we\u2019re told, will transform the + way work is done, impacting every facet of our lives, personal and professional.\\nWe\u2019d + barely surfaced from a landslide of NFT and crypto hype that characterized the + early 2020s, and the metaverse bubble that followed, before media voices began + singing the praises of[generative AI] (gen AI) in the wake of releases such + as OpenAI\u2019s[GPT] model family, Anthropic\u2019s[Claude] and Microsoft\u2019s + Copilot.\\nWhile the chorus hasn\u2019t moved on entirely, the focus in 2025 + has shifted from[large language models (LLMs)] to advancements in the ostensibly + autonomous[artificial intelligence (AI)] agents ushering in the future of work.\\nDespite + a momentary surge in gen AI interest around[Deepseek] \u2019s R1, which promised + significant performance improvements over ChatGPT, the dominant innovation narrative + in 2025 is the AI agent.\\nMedia coverage highlights the promises of innovation,[automation] + and efficiency agents will bring, but how much of the conversation is click-hungry + hype?\\nThe ad-supported media world thrives on clicks, and it\u2019s reasonable + to expect sensational, attention-grabbing headlines crafted to garner yours. + But what can we realistically expect from[agentic AI] in 2025, and how will + it affect our lives?\\nWe spoke with several IBM experts to cut through the + hype, with the goal of holding a more reasonable conversation about AI agents + and what they\u2019re going to do. Our team of informed insiders includes:\\n* + [Maryam Ashoori, PhD]: Director of Product Management,[IBM\xAE watsonx.ai\u2122] + \\n* [Marina Danilevsky]: Senior Research Scientist, Language Technologies\\n* + [Vyoma Gajjar]: AI Technical Solutions Architect\\n* [Chris Hay]: Distinguished + Engineer\\nIndustry newsletter\\n### The latest AI trends, brought to you by + experts\\nGet curated insights on the most important\u2014and intriguing\u2014AI + news. Subscribe to our weekly Think newsletter. See the[IBM Privacy Statement].\\n### + Thank you! You are subscribed.\\nYour subscription will be delivered in English. + You will find an unsubscribe link in every newsletter. You can manage your subscriptions + or unsubscribe[here]. Refer to our[IBM Privacy Statement] for more information.\\n## + What are AI agents?\\nAn AI agent is a software program capable of acting autonomously + to understand, plan and execute tasks. AI agents are powered by LLMs and can + interface with tools, other models and other aspects of a system or network + as needed to fulfill user goals.\\nWe\u2019re going beyond asking a chatbot + to suggest a dinner recipe based on the available ingredients in the fridge. + Agents are more than automated[customer experience] emails that inform you it\u2019ll + be a few days until a real-world human can get to your inquiry.\\n[AI agents + differ from traditional AI assistants] that need a prompt each time they generate + a response. In theory, a user gives an agent a high-level task, and the agent + figures out how to complete it.\\nCurrent offerings are still in the early stages + of approaching this idea. \u201CWhat\u2019s commonly referred to as \u2018agents\u2019 + in the market is the addition of rudimentary planning and tool-calling (sometimes + called function calling) capabilities to LLMs,\u201D says Ashoori. \u201CThese + enable the LLM to break down complex tasks into smaller steps that the LLM can + perform.\u201D\\nHay is optimistic that more robust agents are on the way: \u201CYou + wouldn\u2019t need any further progression in models today to build[future AI + agents],\u201D he says.\\nWith that out of the way, what\u2019s the conversation + about agents over the coming year, and how much of it can we take seriously?\\nAI + agents\\n### What are AI agents?\\nFrom monolithic models to compound AI systems, + discover how AI agents integrate with databases and external tools to enhance + problem-solving capabilities and adaptability.\\n[\\nLearn more\\n] \\n## Narrative + 1: 2025 is the year of the AI agent\\n\u201CMore and better agents\u201D are + on the way, predicts Time.[1] \u201CAutonomous \u2018agents\u2019 and profitability + are likely to dominate the artificial intelligence agenda,\u201D reports Reuters.[2] + \u201CThe age of agentic AI has arrived,\u201D promises Forbes, in response + to a claim from Nvidia\u2019s Jensen Huang.[3] \\nTech media is awash with assurances + that our lives are on the verge of a total transformation. Autonomous agents + are poised to streamline and alter our jobs, drive optimization and accompany + us in our daily lives, handling our mundanities in real time and freeing us + up for creative pursuits and other higher-level tasks.\\n### 2025 as the year + of agentic exploration\\n\u201CIBM and Morning Consult did a survey of 1,000 + developers who are building AI applications for enterprise, and 99% of them + said they are exploring or developing AI agents,\u201D\_explains Ashoori. \u201CSo + yes, the answer is that 2025 is going to be the year of the agent.\u201D However, + that declaration is not without nuance.\\nAfter establishing the current market + conception of agents as LLMs with function calling, Ashoori draws a distinction + between that idea and truly autonomous agents. \u201CThe true definition [of + an AI agent] is an intelligent entity with reasoning and planning capabilities + that can autonomously take action. Those reasoning and planning capabilities + are up for discussion. It depends on how you define that.\u201D\\n\u201CI definitely + see AI agents heading in this direction, but we\u2019re not fully there yet,\u201D + says Gajjar. \u201CRight now, we\u2019re seeing early glimpses\u2014AI agents + can already analyze data, predict trends and automate workflows to some extent. + But building AI agents that can autonomously handle complex decision-making + will take more than just better algorithms. We\u2019ll need big leaps in contextual + reasoning and testing for edge cases,\u201D she adds.\\nDanilevsky isn\u2019t + convinced that this is anything new. \u201CI'm still struggling to truly + believe that this is all that different from just orchestration,\u201D she says. + \u201CYou've renamed orchestration, but now it's called agents, because + that's the cool word. But orchestration is something that we've been + doing in programming forever.\u201D\\nWith regard to 2025 being the year of + the agent, Danilevsky is skeptical. \u201CIt depends on what you say an agent + is, what you think an agent is going to accomplish and what kind of value you + think it will bring,\u201D she says. \u201CIt's quite a statement to make + when we haven't even yet figured out ROI (return on investment) on LLM technology + more generally.\u201D\\nAnd it\u2019s not just the business side that has her + hedging her bets. \u201CThere's the hype of imagining if this thing could + think for you and make all these decisions and take actions on your computer. + Realistically, that's terrifying.\u201D\\nDanilevsky frames the disconnect + as one of miscommunication. \u201C[Agents] tend to be very ineffective because + humans are very bad communicators. We still can't get chat agents to interpret + what you want correctly all the time.\u201D\\nStill, the forthcoming year holds + a lot of promise as an era of experimentation. \u201CI'm a big believer + in [2025 as the year of the agent],\u201D says Hay excitedly.\\nEvery large + tech company and hundreds of startups are now experimenting with agents. Salesforce, + for example, has released their Agentforce platform, which enables users to + create agents that are easily integrated within the Salesforce app ecosystem.\\n\u201CThe + wave is coming and we're going to have a lot of agents. It's still a + very nascent ecosystem, so I think a lot of people are going to build agents, + and they're going to have a lot of fun.\u201D\\n## Narrative 2: Agents can + handle highly complex tasks on their own\\nThis narrative assumes that today\u2019s + agents meet the theoretical definition outlined in the introduction to this + piece. 2025\u2019s agents will be fully autonomous AI programs that can scope + out a project and complete it with all the necessary tools they need and with + no help from human partners. But what\u2019s missing from this narrative is + nuance.\\n### Today\u2019s models are more than enough\\nHay believes that the + groundwork has already been laid for such developments. \u201CThe big thing + about agents is that they have the ability to plan,\u201D he outlines. \u201CThey + have the ability to reason, to use tools and perform tasks, and they need to + do it at speed and scale.\u201D\\nHe cites 4 developments that, compared to + the best models of 12 to 18 months ago, mean that the models of early 2025 can + power the agents envisioned by the proponents of this narrative:\\n* Better, + faster, smaller models\\n* Chain-of-thought (COT) training\\n* Increased context + windows\\n* Function calling\\n\u201CNow, most of these things are in play,\u201D + Hay continues. \u201CYou can have the AI call tools. It can plan. It can reason + and come back with good answers. It can use inference-time compute. You\u2019ll + have better chains of thought and more memory to work with. It's going to + run fast. It\u2019s going to be cheap. That leads you to a structure where I + think you can have agents. The models are improving and they're getting + better, so that's only going to accelerate.\u201D\\n### Realistic expectations + are a must\\nAshoori is careful to differentiate between what agents will be + able to do later, and what they can do now. \u201CThere is the promise, and + there is what the agent's capable of doing today,\u201D she says. \u201CI + would say the answer depends on the use case. For simple use cases, the agents + are capable of [choosing the correct tool], but for more sophisticated use cases, + the technology\\nSummary: None\\n\\n\\nTitle: AWS re:Invent 2025 - Autonomous + AI agents you can truly ...\\nURL: https://www.youtube.com/watch?v=RhRJsjr-I3I\\nID: + https://www.youtube.com/watch?v=RhRJsjr-I3I\\nScore: None\\nPublished Date: + 2025-12-06T21:27:38.021Z\\nAuthor: AWS Events\\nImage: https://i.ytimg.com/vi/RhRJsjr-I3I/hqdefault.jpg\\nFavicon: + None\\nExtras: None\\nSubpages: None\\nText: Learn how to power accurate and + trusted agents at scale with AI-ready data and governance. This presentation + is brought to you by IBM, an AWS Partner.\\\\n\\\\nLearn more:\\\\nMore AWS + events: https://go.aws/3kss9CP \\\\n\\\\nSubscribe:\\\\nMore AWS videos: http://bit.ly/2O3zS75\\\\nMore + AWS events videos: http://bit.ly/316g9t4\\\\n\\\\nABOUT AWS:\\\\nAmazon Web + Services (AWS) hosts events, both online and in-person, bringing the cloud computing + community together to connect, collaborate, and learn from AWS experts.\\\\n + AWS is the world's most comprehensive and broadly adopted cloud platform, offering + over 200 fully featured services from data centers globally. Millions of customers\u2014including + the fastest-growing startups, largest enterprises, and leading government agencies\u2014are + using AWS to lower costs, become more agile, and innovate faster.\\\\n\\\\n#AWSreInvent + #AWSreInvent2025 #AWS\\n| view_count: 128 views | short_view_count: 128 views + | num_likes: 1 like | num_subscribers: 170 thousand | duration: 19 minutes 10 + seconds\\nSummary: None\\n\\n\\nTitle: Autonomous AI Agents Are Reshaping the + Business ...\\nURL: https://kpmg.com/sk/en/insights/2025/06/autonomous-ai-agents-reshape-business-landscape.html\\nID: + https://kpmg.com/sk/en/insights/2025/06/autonomous-ai-agents-reshape-business-landscape.html\\nScore: + None\\nPublished Date: 2025-06-25T00:00:00.000Z\\nAuthor: None\\nImage: /content/dam/kpmgsites/sk/images/2025/Agentic_AI_cover.jpg\\nFavicon: + https://kpmg.com/etc.clientlibs/kpmg/clientlibs/clientlib-site/resources/images/favicons/favicon-32x32.svg\\nExtras: + None\\nSubpages: None\\nText: Autonomous AI Agents Are Reshaping the Business + Landscape\\n[Skip to main content] \\n[contact\\\\_mail\\nContact us\\n] \\n#### + ###### [\\n] \\n[tips\\\\_and\\\\_updates\\nStay informed with our newsletters\\n] + \\nlanguage\\nSlovakia (EN)\\nexpand\\\\_more\\nclose\\nSearchcancel\\nmenu\\n[![KPMG + logo]] \\n[contact\\\\_mail\\nContact us\\n] \\n#### ###### [\\n] \\n[tips\\\\_and\\\\_updates\\nStay + informed with our newsletters\\n] \\nlanguage\\nSlovakia (EN)\\nexpand\\\\_more\\nsearch\\nSearch\\nsearch\\nYour + search term was too short.\\nSearch terms must include 3 or more characters.\\nPrevious + search terms\\nclose\\nCancel\\n# Autonomous AI Agents Are Reshaping the Business + Landscape\\nArtificial intelligence that can think and act independently is + no longer just a vision of the future \u2014it\u2019s becoming a competitive + advantage\\nevent25-06-2025\\nShare\\n* [] \\n* [] \\n![3D Rendering futuristic + robot technology development, artificial intelligence AI, and machine learning + concept. Global robotic bionic science research for future of human life.] \\n* + [From GenAI to AgAI] \\n* [Technical Foundations of AgAI] \\n* [AI Agents as + Digital Workforce] \\nAgentic artificial intelligence\u2014autonomous systems + capable of perceiving, reasoning, and acting independently\u2014is no longer + just a vision of the future. This rapidly evolving technology is already transforming + how businesses operate, innovate, and compete to stay ahead.\\nIn our two-part + blog series,***The Age of Agentic AI***, we explore the rise of agentic AI (AgAI) + from both a business and a technological perspective. We\u2019ll trace its development, + examine its technical foundations, highlight real-world applications, and assess + its impact across industries\u2014along with the risks and opportunities it + presents. Through expert insights and case studies, we\u2019ll show that agentic + AI isn\u2019t just another tool\u2014it\u2019s a new kind of digital workforce, + reshaping the future of work, decision-making, and enterprise strategy.\\n### + Author\\n##### **\\nAlexander Zagnetko\\n***Manager,*Process Organization and + Improvement\\n[Contact us for more information] \\n#### From Generative to Agentic + AI\\nOver the past five years, we\u2019ve witnessed a dramatic leap in AI capabilities. + From the launch of GPT-3 in 2020 to the widespread adoption of models like GPT-4, + Claude, and Gemini between 2023 and 2024, generative AI has already revolutionized + numerous industries. But as impressive as these developments are, they represent + just the beginning of a much broader transformation.\\nThe next major milestone + is the rise of autonomous agents\u2014systems that can now carry out complex + workflows and collaborate with humans or other agents. These systems set their + own goals, plan their actions, use tools, and adapt to changing conditions. + Unlike traditional AI agents designed for narrow, specific tasks, agentic AI + systems operate with much greater independence. They are built to:\\n* Perceive + their environment through both structured and unstructured data,\\n* Plan actions + using logic and decision-making frameworks,\\n* Execute tasks using tools, APIs, + or physical devices,\\n* Learn from feedback and improve over time,\\n* Collaborate + with humans and other agents.\\nThe shift from generative to agentic AI marks + a fundamental leap in artificial intelligence. It stems from the integration + of multiple capabilities: large language models (LLMs), memory systems, planning + algorithms, tool integration, and multi-agent coordination. Together, these + elements enable AI systems to function as digital workers\u2014capable of performing + tasks that once required human judgment and coordination. Agentic AI is opening + up entirely new possibilities and will have a profound impact on business strategy, + organizational design, and competitive advantage.\\n#### Technical Foundations + of Agentic AI\\nAgAI systems are built on a layered architecture that combines + several advanced technologies. Let\u2019s break down the core components. At + the heart of most agentic systems is a**large language model (LLM)**such as + GPT-4, Claude, or Gemini. These models are trained on massive datasets and can + understand and generate human-like language. They provide the linguistic and + reasoning capabilities that underpin agent behaviour. Unlike traditional chatbots, + agentic systems need to**remember past interactions**, decisions, and outcomes. + Memory enables agents to learn from experience, personalize interactions, and + maintain continuity over time.\\nFor agents to act autonomously, they must be + able to plan several steps ahead and consider the consequences of their decisions. + This is why they combine reasoning with tool use\u2014a method known as ReAct + (Reasoning + Acting)\u2014and explore multiple logical pathways before choosing + a course of action, a technique referred to as Tree-of-Thoughts. Another well-known + approach to evaluating potential actions is Monte Carlo Tree Search, originally + developed for game-playing AI. These techniques enable agents to break down + complex goals into manageable tasks and respond flexibly to feedback.\\nAs agents + become more autonomous, ensuring their responsible and ethical behavior becomes + increasingly important. This is essential for building trust and deploying agentic + AI in a safe and accountable way. Key mechanisms include:\\n* **Human-in-the-loop**\u2013 + giving humans the ability to approve or override an agent\u2019s decisions.\\n* + **Audit trails**\u2013 tracking the agent\u2019s actions to ensure accountability.\\n* + **Ethical constraints**\u2013 rules that prevent harmful or unethical behavior.\\n* + **Value alignment**\u2013 ensuring agents act in accordance with human values.\\n#### + AI Agents as the New Digital Workforce\\nAgentic AI is opening up new possibilities + for businesses in terms of efficiency, scalability, and innovation. Whether + it\u2019s automating routine tasks or orchestrating complex workflows, agentic + AI is finding broad applications across sectors\u2014from financial services + and healthcare to education, retail, and software development.\\naccount\\\\_balance\\\\_wallet## + Financial Services\\nThe financial sector has long been a frontrunner in AI + adoption, and agentic AI takes automation to the next level. Autonomous agents + can monitor market conditions in real time, assess risks, and execute trades + with minimal human oversight. They continuously scan transactions for suspicious + activity, track regulatory updates, and ensure compliance. In customer service + and advisory roles, agents analyze client data and recommend tailored financial + products, effectively acting as digital financial advisors.\\nmedical\\\\_services## + Healthcare\\nHealthcare is one of the most complex and data-intensive sectors, + making it a natural fit for agentic AI. Agents can analyze medical records, + test results, and imaging data, flag potential drug interactions, suggest additional + diagnostics, and assist with documentation. They also coordinate care by scheduling + appointments, managing referrals, and reducing redundancies. On the administrative + side, agents streamline tasks like billing, claims processing, and regulatory + compliance, allowing medical professionals to focus more on patient care.\\ngavel## + Legal and Compliance\\nThe legal sector is undergoing a major transformation + thanks to agentic AI. Autonomous agents can read, interpret, and generate legal + documents, analyze contracts, highlight key clauses, flag risks, suggest edits, + and draft standard agreements based on templates and client requirements. They + also monitor regulatory changes across jurisdictions, evaluate their impact, + and recommend compliance actions. In litigation, they assist with legal research, + summarizing precedents, and preparing arguments.\\ncast\\\\_for\\\\_education## + Education and Training\\nDigital transformation has reached education as well. + AI agents serve as interactive tutors, answering questions in real time, providing + feedback, and adapting to individual learning styles. They can even detect emotional + cues to offer support when motivation is low. In assessments, agents automatically + grade assignments, generate tests, and provide detailed feedback, while identifying + performance trends and recommending timely interventions.\\nstorefront## Retail + and E-commerce\\nIn the retail sector, companies use agentic AI to enhance customer + experience, optimize operations, and boost sales. In logistics, agents forecast + demand, manage inventory, coordinate with suppliers, and respond to market changes + in real time\u2014for example, by rerouting deliveries or adjusting pricing. + In marketing, they design targeted campaigns, segment audiences, and optimize + content, improving product placement and promotional strategies based on customer + behavior.\\nwebhook## Software Development\\nAgentic AI is also revolutionizing + the entire software development lifecycle\u2014from design to deployment. Agents + can generate code from natural language, detect and fix bugs, optimize performance, + and create documentation and tests. In product management, they gather user + feedback, prioritize features, draft specifications, and coordinate team collaboration, + accelerating innovation and improving product quality.\\n#### Speed of Implementation + Is the Competitive Advantage\\nAgentic AI is not just about boosting efficiency\u2014it\u2019s + a fundamental shift in how companies operate, make decisions, and create value. + By combining large language models, memory systems, planning algorithms, and + the ability to interact with external tools, agentic AI becomes an active participant + in business processes. Its impact goes beyond any single department\u2014it\u2019s + transforming entire industries, from finance and healthcare to software development.\\nOrganizations + that understand and implement this technology early will gain a competitive + edge\u2014not only through speed and precision but also through their ability + to respond to market changes, personalize services, and innovate faster than + ever before.\\n## Contact our experts\\nShould you wish more information on + how we can help your business or to arrange a meeting for personal presentation\\nSummary: + None\\n\\n\\nTitle: The State of AI Agents - Cobus Greyling\\nURL: https://cobusgreyling.medium.com/the-state-of-ai-agents-3e13fa123ab8\\nID: + https://cobusgreyling.medium.com/the-state-of-ai-agents-3e13fa123ab8\\nScore: + None\\nPublished Date: None\\nAuthor: Cobus Greyling\\nImage: https://miro.medium.com/v2/resize:fit:1200/1*1FdlyvCKYt0nc3rGwqZ6dA.png\\nFavicon: + https://miro.medium.com/v2/5d8de952517e8160e40ef9841c781cdc14a5db313057fa3c3de41c6f5b494b19\\nExtras: + None\\nSubpages: None\\nText: The State of AI Agents. AI Agents burst on the + scene not too\u2026 | by Cobus Greyling | Medium\\n[Sitemap] \\n[Open in app] + \\nSign up\\n[Sign in] \\n[Medium Logo] \\n[\\nWrite\\n] \\n[\\nSearch\\n] \\nSign + up\\n[Sign in] \\n![] \\nPress enter or click to view image in full size\\n![] + \\n# The State of AI Agents\\n## AI Agents burst on the scene not too long ago\u2026the + potential of AI Agents is real\u2026but the hype has been met with real-world + impediments when it comes to implementation.\\n[\\n![Cobus Greyling] \\n] \\n[Cobus + Greyling] \\n6 min read\\n\xB7Nov 3, 2025\\n[\\n] \\n--\\n2\\n[] \\nListen\\nShare\\nThere + has been a number of[***sobering studies***] when it comes to the realities + of implementing AI Agents\u2026\\nThe existence of the[***Pareto Frontier***] + ***(cost vs accuracy)***has been well examined and documented.\\n[NVIDIA] has + been taking the lead in moving away from the idea of an AI Agent being underpinned + by a single LLM.\\nThey are advocating the use of multiple Small Language Models + (SLMs) and orchestrating the SLMs in an[Agentic workflow].\\n> Latency and cost + have also been impediments to AI agent use.\\nCost is being addressed by using + and orchestrating multiple[***open-sourced SLMs***] which is***fine-tuned***for + specific tasks within an AI Agent.\\n***Latency***is addressed with[parallel + running task] s within an AI Agent.\\nOr having multiple AI Agents within a + Agentic workflow, and executing those AI Agents in parallel for complex tasks.\\n[Lighter + AI Agents] \u2026NVIDIA has also been backing this growing trend of**agentic + workflows**where the orchestrating AI Agent (the \u201Cconductors\u201D that + plan, route tasks, and synthesise outputs) are indeed***leaning toward being + more***[***lightweight***] ***and less code-intensive (AI Agents)***to implement.\\n***Agentic + Workflows***are easy to build in general, especially when presented via a GUI + builder, that is no-code and handles tasks like versioning, collaboration and + more.\\nDrift and continuous improvement are addressed with what[NVIDIA refers + to as a***data-flywhee***] ***l***.\\nWhere a***continuous feedback loop***is + established for training data to improve the different models orchestrated in + the AI Agent/Agentic Workflow.\\n> Most of these challenges I detailed in an + [**> earlier post\\n**] > .\\nAlso, for a long time there has been this notion + of different[**levels of AI Agents**].\\nTypically there are[**5 or 6 levels + defined**, in a number of studies].\\n[**HuggingFace**also published a study] + where they detail the shifting relationship between humans and AI Agents\u2026\\nThis + is why I found this recent study interesting (*the image below*) where they + extended the levels from what was typically**5 levels**, to**6 levels**.\\nThe + one extreme is Level 0, here the human dominates and AI plays no role. And on + the other extreme, Level 5, where the human has no role and the AI takes full + control.\\nAnd in between are these steps of going from one level to another.\\n> + This approach really resonates with me as it so sober in dissecting the gradual + approach in task automation transition.\\n***Where are we now, based on this + chart?***\\nPress enter or click to view image in full size\\n![] \\nWe have + been introducing assistance from the days of Chatbots and NLU models\u2026\\n\u2026currently + we are in the process of transferring tasks from human dominance to AI dominance. + This is happening on a***task-by-task basis.***\\nHence this***movement***must + not be seen as a single wave hitting enterprises and our way-of-work.\\nBut + rather a detailed approach of identifying tasks suited (for now) to move along + this trajectory.\\nConsidering the image below, it breaks down the***traditional***four + pillars of AI Agents.\\n*In this article I will stay with this for the sake + of simplicity. But to some degree this approach has been upended by Agentic + Workflows and multiple smaller language models.*\\n> A recent study also introduced + a data component\u2026data lakes.\\nSo not only is there focus on context / + memory and[tools]. But also giving the AI Agent access to vast amounts of data + to help the AI Agent perform user requests.\\nPress enter or click to view image + in full size\\n![] \\nThe table below (from the study) shows the contrast between**traditional + AI Agents**and what the study terms**Data Agents**.\\nPress enter or click to + view image in full size\\n![] \\n***The future?***\\n[For AI Agents to be truly + autonomous and able to address virtually any task], it needs to be able to write + code.\\n> On this front [> OpenAI\\n] > and [> xAI\\n] > introduced functions + / tools which are able to write code.\\nHence the AI Agent gets an instruction, + there is no pre-existing tool or function to perform the request, but the AI + Agent has the ability to write code to achieve a goal.\\n***There exist limitations\u2026***\\nSandbox + Restrictions for Security\u2026AI models operate within tightly controlled sandboxes + to mitigate risks, prohibiting arbitrary software installs or direct internet + access.\\nThey\u2019re confined to pre-loaded libraries, forcing clever workarounds + like built-in proxies.\\nWhile this setup enables basic functionality, it blocks + the creation of truly innovative tools that demand unrestricted resources \u2014limiting + adaptability to predefined, bounded tasks.\\nDebugging remains a weak spot, + models often hallucinate subtle bugs in intricate logic and finite token limits***(coupled + with cost constraints)***make exhaustive testing impractical.\\nThey excel at + 70\u201380% of straightforward objectives but stumble on edge cases, frequently + demanding human intervention.\\nIn my experience, AI Agents generate code that***looks***flawless, + only for execution in notebooks or live environments to reveal flaws \u2014necessitating + rounds of re-prompting.\\nThe bigger the code block, the steeper the challenge. + Autocomplete and highly contextual use in IDE\u2019s work quite well.\\nAmbiguous + instructions derail progress, with agents chasing***tangential sub-goals instead + of the core intent***.\\nThey\u2019re far from self-directed; precision is essential + to keep them aligned.\\nThis is why detailed, step-by-step breakdowns are non-negotiable.\\nTools + like Gemini thrive in Colab notebooks precisely because they leverage rich contextual + anchors \u2014iterating on proven, working code cell by cell, rather than from + scratch.\\n***If you are still reading, thank you\u2026I hope it was helpful, + any feedback will be welcome\u2026***\\nBelow are a number of resources that + might be of help on this topic.\\nPress enter or click to view image in full + size\\n![] \\n[***Chief Evangelist***] ***@***[*Kore.ai*] *| I\u2019m passionate + about exploring the intersection of AI and language. Language Models, AI Agents, + Agentic Apps, Dev Frameworks & Data-Driven Tools shaping tomorrow.*\\nPress + enter or click to view image in full size\\n![] \\n[\\n## A Survey of Data Agents: + Emerging Paradigm or Overstated Hype?\\n### The rapid advancement of large language + models (LLMs) has spurred the emergence of data agents--autonomous systems\u2026\\narxiv.org\\n] + \\n[\\n## COBUS GREYLING\\n### Where AI Meets Language | Language Models, AI + Agents, Agentic Applications, Development Frameworks & Data-Centric\u2026\\nwww.cobusgreyling.com\\n] + \\n[\\n## Five Levels Of AI Agents\\n### AI Agents are defined as artificial + entities which can perceive their environment, make decisions and take actions\u2026\\ncobusgreyling.medium.com\\n] + \\n[\\n## 5 Levels Of AI Agents (Updated)\\n### \U0001D5D4\U0001D602\U0001D601\U0001D5FC\U0001D5FB\U0001D5FC\U0001D5FA\U0001D5FC\U0001D602\U0001D600\U0001D5D4\U0001D5DC\U0001D5D4\U0001D5F4\U0001D5F2\U0001D5FB\U0001D601\U0001D600are + AI systems capable of performing a series of complex tasks independently to\u2026\\ncobusgreyling.medium.com\\n] + \\n[\\n## Fully Autonomous AI Agents Should Not Be Developed\\n### Research + from HuggingFace\u2026and they coined a new definition for what an AI Agent + is.\\ncobusgreyling.medium.com\\n] \\n[\\n## The Evolution of AI Agents\\n### + \u2026it is helpful to consider how far we have come in few years\u2019 time\u2026\\ncobusgreyling.medium.com\\n] + \\n[\\n## GitHub - HKUSTDial/awesome-data-agents: Continuously updated paper + list on advancements in Data\u2026\\n### Continuously updated paper list on + advancements in Data Agents. Companion repo to our paper "A Survey of Data + Agents\u2026\\ngithub.com\\n] \\n[\\nAI\\n] \\n[\\nArtificial Intelligence\\n] + \\n[\\nMachine Learning\\n] \\n[\\nSoftware Development\\n] \\n[\\nSoftware + Engineering\\n] \\n[\\n![Cobus Greyling] \\n] \\n[\\n![Cobus Greyling] \\n] + \\n[## Written byCobus Greyling\\n] \\n[29K followers] \\n\xB7[0 following] + \\nat the intersection of AI & language.[www.cobusgreyling.com] \\n## Responses + (2)\\n[] \\nSee all responses\\n[\\nHelp\\n] \\n[\\nStatus\\n] \\n[\\nAbout\\n] + \\n[\\nCareers\\n] \\n[\\nPress\\n] \\n[\\nBlog\\n] \\n[\\nPrivacy\\n] \\n[\\nRules\\n] + \\n[\\nTerms\\n] \\n[\\nText to speech\\n]\\nSummary: None\\n\\n\\nTitle: AI + agents arrived in 2025 \u2013 here's what happened and ...\\nURL: https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325\\nID: + https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325\\nScore: + None\\nPublished Date: 2025-12-29T00:00:00.000Z\\nAuthor: Thomas \u015Eerban + von Davier\\nImage: https://images.theconversation.com/files/709953/original/file-20251219-66-te6uyi.jpg?ixlib=rb-4.1.0&rect=0%2C250%2C8000%2C4000&q=45&auto=format&w=1356&h=668&fit=crop\\nFavicon: + https://cdn.theconversation.com/static/tc/logos/web-app-logo-192x192-2d05bdd6de6328146de80245d4685946.png\\nExtras: + None\\nSubpages: None\\nText: AI agents arrived in 2025 \u2013here\u2019s what + happened and the challenges ahead in 2026\\n[] [] \\n[![The Conversation]] \\nAcademic + rigour, journalistic flair\\n![a couple dozen robot face emojis floating between + two human hands] \\nAI agents have emerged from the lab, bringing promise and + peril.[tadamichi/iStock via Getty Images] \\n# **AI agents arrived in 2025 \u2013here\u2019s + what happened and the challenges ahead in2026**\\nPublished: December 29, 2025 + 4.35pm CET\\n[****Thomas \u015Eerban von Davier,*Carnegie Mellon University*] + \\n### Author\\n1. [![] Thomas \u015Eerban von Davier] \\nAffiliated Faculty + Member, Carnegie Mellon Institute for Strategy and Technology, Carnegie Mellon + University\\n### Disclosure statement\\nThomas \u015Eerban von Davier does not + work for, consult, own shares in or receive funding from any company or organisation + that would benefit from this article, and has disclosed no relevant affiliations + beyond their academic appointment.\\n### Partners\\n[] \\n[Carnegie Mellon University] + provides funding as a member of The Conversation US.\\n[View all partners] \\n### + DOI\\n[https://doi.org/10.64628/AAI.maxh7d4en] \\nhttps://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325\\nhttps://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325\\nLink + copied\\nShare article\\nShare article\\nCopy link[Email] \\n[Bluesky] [Facebook] + [WhatsApp] [Messenger] [LinkedIn] [X (Twitter)] \\nPrint article\\nIn artificial + intelligence, 2025 marked a decisive shift. Systems once confined to research + labs and prototypes began to appear as everyday tools. At the center of this + transition was the rise of AI agents \u2013AI systems that can use other software + tools and act on their own.\\nWhile researchers have studied AI for more than + 60 years, and the term \u201Cagent\u201D has long been part of the field\u2019s + vocabulary, 2025 was the year the concept became concrete for developers and + consumers alike.\\nAI agents moved from theory to infrastructure, reshaping + how people interact with large language models, the systems that power chatbots + like ChatGPT.\\nIn 2025, the definition of AI agent shifted from the[academic + framing] of systems that perceive, reason and act to AI company[Anthropic\u2019s + description] of large language models that are capable of using software tools + and taking autonomous action. While large language models have long excelled + at text-based responses, the recent change is their expanding capacity to act, + using tools, calling[APIs], coordinating with other systems and completing tasks + independently.\\nThis shift did not happen overnight. A key inflection point + came in late 2024, when Anthropic released the[Model Context Protocol]. The + protocol allowed developers to connect large language models to external tools + in a standardized way, effectively giving models the ability to act beyond generating + text. With that, the stage was set for 2025 to become the year of AI agents.\\n[![Embedded + YouTube video]] \\nAI agents are a whole new ballgame compared with generative + AI.## The milestones that defined 2025\\nThe momentum accelerated quickly. In + January, the release of Chinese model[DeepSeek-R1] as an[open-weight] model + disrupted assumptions about who could build high-performing large language models, + briefly rattling markets and intensifying global competition. An open-weight + model is an AI model whose training, reflected in values called weights, is + publicly available. Throughout 2025, major U.S. labs such as[OpenAI],[Anthropic],[Google] + and[xAI] released larger, high-performance models, while Chinese tech companies + including[Alibaba],[Tencent], and[DeepSeek] expanded the open-model ecosystem + to the point where the Chinese models have been[downloaded more than American + models].\\n##### Another turning point came in April, when Google introduced + its[Agent2Agent protocol]. While Anthropic\u2019s Model Context Protocol focused + on how agents use tools, Agent2Agent addressed how agents communicate with each + other. Crucially, the two protocols were designed to work together. Later in + the year, both[Anthropic] and[Google] donated their protocols to the open-source + software nonprofit Linux Foundation, cementing them as open standards rather + than proprietary experiments.\\nThese developments quickly found their way into + consumer products. By mid-2025, \u201Cagentic browsers\u201D began to appear. + Tools such as[Perplexity\u2019s Comet],[Browser Company\u2019s Dia],[OpenAI\u2019s + GPT Atlas],[Copilot in Microsoft\u2019s Edge],[ASI X Inc.\u2019s Fellou],[MainFunc.ai\u2019s + Genspark],[Opera\u2019s Opera Neon] and others reframed the browser as an active + participant rather than a passive interface. For example, rather than helping + you search for vacation details, it plays a part in booking the vacation.\\nAt + the same time, workflow builders like[n8n] and[Google\u2019s Antigravity] lowered + the technical barrier for creating custom agent systems beyond what has already + happened with coding agents like[Cursor] and[GitHub Copilot].\\n## New power, + new risks\\nAs agents became more capable, their risks became harder to ignore. + In November, Anthropic disclosed how its Claude Code agent[had been misused] + to automate parts of a cyberattack. The incident illustrated a broader concern: + By automating repetitive, technical work, AI agents can also lower the barrier + for malicious activity.\\nThis tension defined much of 2025. AI agents expanded + what individuals and organizations could do, but they also[amplified existing + vulnerabilities]. Systems that were once isolated text generators became interconnected, + tool-using actors operating with little human oversight.\\n[![Embedded YouTube + video]] \\nThe business community is gearing up for multiagent systems.## What + to watch for in 2026\\nLooking ahead, several open questions are likely to shape + the next phase of AI agents.\\nOne is benchmarks. Traditional benchmarks, which + are like a structured exam with a series of questions and standardized scoring, + work well for single models, but[agents are composite systems] made up of models, + tools, memory and decision logic. Researchers increasingly want to evaluate[not + just outcomes, but processes]. This would be like asking students to show their + work, not just provide an answer.\\nProgress here will be critical for improving + reliability and trust, and ensuring that an AI agent will perform the task at + hand. One method is establishing clear definitions around[AI agents and AI workflows]. + Organizations will need to map out exactly where AI will[integrate into workflows + or introduce new ones].\\nAnother development to watch is governance. In late + 2025, the Linux Foundation announced the creation of the[Agentic AI Foundation], + signaling an effort to establish shared standards and best practices. If successful, + it could play a role like the[World Wide Web Consortium] in shaping an open, + interoperable agent ecosystem.\\nThere is also a growing debate over model size. + While large, general-purpose models dominate headlines, smaller and more specialized + models are often[better suited to specific tasks]. As agents become configurable + consumer and business tools, whether through browsers or workflow management + software, the power to choose the right model increasingly shifts to users rather + than labs or corporations.\\n## The challenges ahead\\nDespite the optimism, + significant socio-technical challenges remain. Expanding data center infrastructure[strains + energy grids] and affects local communities. In workplaces, agents raise concerns + about automation,[job displacement] and surveillance.\\nFrom a security perspective, + connecting models to tools and stacking agents together[multiplies risks] that + are already unresolved in standalone large language models. Specifically, AI + practitioners are addressing the dangers of[indirect prompt injections], where + prompts are hidden in open web spaces that are readable by AI agents and result + in harmful or unintended actions.\\nRegulation is another unresolved issue. + Compared with[Europe] and[China], the United States has relatively limited oversight + of algorithmic systems. As AI agents become embedded across digital life, questions + about access, accountability and limits remain largely unanswered.\\nMeeting + these challenges will require more than technical breakthroughs. It demands[rigorous + engineering practices], careful design and clear documentation of how systems + work and fail. Only by treating AI agents as socio-technical systems rather + than mere software components, I believe, can we build an AI ecosystem that + is both innovative and safe.\\n**\\n* [Artificial intelligence (AI)] \\n* [Google] + \\n* [Technology] \\n* [OpenAI] \\n* [Anthropic] \\n* [AI safety] \\n* [AI agents] + \\n### Events\\n[More events] \\n### Jobs\\n* ##### [Deputy Editor] \\n* ##### + [Director of Professional Development] \\n* ##### [Research Fellow in Cybersecurity + for Cyber-physical systems] \\n* ##### [University Librarian] \\n* ##### [Video + Commissioning Editor] \\n[More jobs]\\nSummary: None\\n\\n\\nTitle: How Autonomous + AI Agents Are Changing Software ...\\nURL: https://www.sculptsoft.com/agentic-ai-in-action-how-autonomous-ai-agents-are-changing-software-development-in-2025/\\nID: + https://www.sculptsoft.com/agentic-ai-in-action-how-autonomous-ai-agents-are-changing-software-development-in-2025/\\nScore: + None\\nPublished Date: 2025-10-15T00:00:00.000Z\\nAuthor: Hiloni Mehta\\nImage: + https://www.sculptsoft.com/wp-content/uploads/2025/10/agentic_ai_in_action_how_autonomous_ai_agents_are_changing_software_development_in_2025.webp\\nFavicon: + https://www.sculptsoft.com/wp-content/uploads/2022/05/SculptSoft-Favicon.svg\\nExtras: + None\\nSubpages: None\\nText: Agentic AI in Action: How Autonomous AI Agents + Are Changing Software Development in 2025 - SculptSoft\\n[![logo]![logo]] \\n**\\n**\\n**\\n[\\nBusiness + Inquiry\\n] \\n[\\nHire Our Experts\\n] \\n# Agentic AI in Action: How Autonomous + AI Agents Are Changing Software Development in 2025\\nPublished By:**[Hiloni + Mehta] **\\nPublished On:**[October 15, 2025] **\\nPublished In:**[AI] **\\n[Introduction] + \\n[What Is Agentic AI and How Does It Work?] \\n[How Autonomous AI Agents Are + Transforming Software Development in 2025] \\n[Key Benefits of Agentic AI in + Software Development] \\n[Use Cases of Agentic AI in Software Development] \\n[The + Future of Software Development: From Teams to AI Ecosystems] \\n[How SculptSoft + Builds Agentic AI Solutions for Modern Businesses] \\n[Conclusion] \\n[FAQs] + \\n#### Introduction\\nSoftware development is entering a new era where artificial + intelligence doesn\u2019t just assist developers, but actively builds, tests, + and improves applications on its own.\\nIn 2025, this evolution is being led + by a new class of intelligent systems known as Agentic AI – autonomous + AI agents capable of perceiving, reasoning, and taking actions independently. + Unlike traditional AI tools that require constant human input, Agentic AI can + analyze goals, make decisions, and execute tasks end-to-end transforming how + businesses create and manage software.\\nFor years, companies have struggled + with familiar challenges:\\n* Slow development cycles that delay releases.\\n* + Rising engineering costs with limited resources.\\n* Complex integrations across + legacy systems.\\n* Dependency on manual processes that limit scalability.\\nAgentic + AI changes this dynamic. These intelligent agents can automate entire parts + of the development lifecycle from writing and reviewing code to running tests + and deploying updates, freeing teams to focus on innovation rather than repetitive + work.\\nAs a result, organizations adopting Agentic AI aren\u2019t just accelerating + development speed; they\u2019re reimagining what software teams can achieve + with autonomous systems that think, act, and learn continuously.\\nIn this blog, + we\u2019ll explore how Agentic AI is transforming software development in 2025, + the real-world impact it\u2019s already making, and how businesses can strategically + implement it to gain a long-term competitive edge.\\n#### What Is Agentic AI + and How Does It Work?\\nBefore diving into how Agentic AI is transforming software + development, it\u2019s important to understand what it actually is and why it\u2019s + considered the next major leap in[artificial intelligence].\\nAgentic AI refers + to a new generation of autonomous AI systems that can perceive information, + reason through objectives, and take actions independently all without requiring + continuous human instructions. Unlike traditional AI tools that respond to commands + or prompts, Agentic AI Agents operate based on goals. They can interpret context, + plan tasks, make decisions, and execute actions on their own much like how a + skilled developer might handle an assignment from start to finish.\\n##### How + Agentic AI Works\\nAgentic AI works through a continuous cognitive loop that + mirrors human-like thinking. Each agent follows a process of perception, reasoning, + action, and reflection, enabling it to learn and improve over time.\\nHere\u2019s + how this plays out in practice:\\n* **Perception:**The AI Agent gathers context + such as project requirements, past code commits, or bug reports from connected + systems.\\n* **Reasoning:**It analyzes this information using its internal logic, + memory, and large language models to determine the best next steps.\\n* **Action:**The + agent performs the required tasks like generating code, testing modules, or + deploying builds through integrated tools or APIs.\\n* **Reflection:**It evaluates + the outcome, identifies what worked or failed, and refines its next decisions + accordingly.\\n##### How Agentic AI Differs from Traditional AI\\nTraditional + AI models are reactive, they provide answers when prompted. In contrast, Agentic + AI is proactive. It acts on defined goals, manages tasks across multiple systems, + and even collaborates with other AI agents to complete larger objectives.\\nFor + instance, while a conventional AI coding assistant can help generate snippets + of code, an Agentic AI system can oversee the entire process from analyzing + requirements to writing, testing, and deploying code autonomously.\\nThis shift + marks the beginning of AI-driven software ecosystems where intelligent agents + work alongside human teams to optimize speed, quality, and scalability.\\nTo + understand**what Agentic AI is, how it functions, and the complete lifecycle + of an AI agent**, read our detailed guide here:[**What Is Agentic AI? A Complete + Guide to Its Lifecycle and Functions**].\\n#### How Autonomous AI Agents Are + Transforming Software Development in 2025\\nSoftware development has always + been about collaboration of designers, developers, testers, and project managers + working together to bring an idea to life. But in 2025, a new kind of collaborator + has joined the team:**Autonomous AI Agents**.\\nThese intelligent, self-operating + AI systems go beyond simple automation. They understand objectives, make independent + decisions, and execute end-to-end workflows within the[software development + lifecycle] (SDLC) transforming how modern teams plan, build, test, and deploy + software.\\nLet\u2019s explore how Agentic AI is reshaping every stage of development:\\n1. + **Planning and Requirement Analysis**\\nIn traditional software planning, teams + spend extensive time gathering requirements, analyzing dependencies, and setting + milestones. Agentic AI accelerates this process through intelligent automation.\\nAI + Agents can read documentation, project briefs, and performance data to extract + accurate requirements automatically. They detect dependencies, forecast development + timelines, and assess risks using real-time data insights.\\nBy leveraging AI-driven + project analysis, businesses gain faster planning cycles, better decision-making, + and improved resource utilization – all key to delivering successful, + high-quality products.\\n2. **Code Generation and Review**\\nThe coding stage + is where autonomous AI Agents truly revolutionize AI software development. Unlike + basic AI coding assistants, these agents can generate entire modules, optimize + logic, and ensure adherence to industry standards.\\nThey interpret requirements, + create efficient algorithms, and perform automated code reviews to identify + syntax errors, security gaps, or performance bottlenecks. This not only enhances + code quality but also reduces development time and human dependency.\\nBy integrating + AI-powered code generation and validation, development teams achieve faster + release cycles, improved maintainability, and greater consistency across projects.\\n3. + **Testing and Quality Assurance**\\nAI in software testing has evolved dramatically + with the rise of Agentic AI. Instead of relying on manual QA teams, AI Agents + autonomously execute testing workflows from creating test cases to identifying + and fixing bugs.\\nThese agents continuously learn from past results, improving + their accuracy and predicting potential failures before they occur. They validate + builds, monitor regression, and ensure that every release meets high performance + and reliability standards.\\nThe outcome: shorter testing cycles, higher accuracy, + and more stable software deployments.\\n4. **Deployment and CI/CD Automation**\\nDeployment + is one of the most complex and time-sensitive stages of software delivery. Autonomous + AI agents simplify this through end-to-end[CI/CD automation.] \\nThey integrate + seamlessly with DevOps pipelines, handling build management, environment setup, + and deployment tasks with precision. During live deployments, AI Agents monitor + performance, detect anomalies, and automatically roll back faulty builds ensuring + zero downtime and continuous delivery stability.\\nThis creates a self-managing + DevOps ecosystem that enhances release reliability and operational efficiency.\\n5. + **Continuous Improvement and Maintenance**\\nWith Agentic AI, software maintenance + transforms from a reactive task into a proactive, self-optimizing process.\\nAI + agents constantly monitor application performance, analyze user behavior, and + identify inefficiencies in real time. They recommend performance optimizations, + manage dependency updates, and even handle minor issue resolutions autonomously.\\nThis + continuous learning loop ensures that your systems evolve intelligently improving + scalability, stability, and efficiency over time without manual intervention.\\nBy + adopting Agentic AI solutions, organizations move beyond static automation into + a new era of AI-driven software development where every stage of the lifecycle + is optimized, intelligent, and self-improving.\\n[SculptSoft], a trusted custom + AI software development company, helps enterprises implement these autonomous + AI systems to accelerate innovation, reduce time-to-market, and achieve higher-quality + outcomes in 2025 and beyond.\\n#### Key Benefits of Agentic AI in Software Development\\nAgentic + AI is transforming software development from a manual, time-consuming process + into a dynamic, intelligent workflow driven by autonomous AI Agents. These systems + don\u2019t just assist developers, they think, plan, and act on their own, making + development faster, smarter, and more efficient.\\nLet\u2019s look at the key + benefits of Agentic AI in software development that are reshaping how modern + businesses build and scale their digital products in 2025.\\n1. **Accelerated + Development Cycles**\\nTraditional development teams spend weeks (sometimes + months) moving from planning to deployment. With autonomous AI Agents, that + process is drastically reduced.\\nAgentic AI automates repetitive tasks like + code generation, testing, and bug fixing so dedicated developers can focus on + strategy and innovation.\\nSummary: None\\n\\n\\nTitle: The Rise of Autonomous + AI Agents: Automating Complex ...\\nURL: https://www.researchgate.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks\\nID: + https://www.researchgate.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks\\nScore: + None\\nPublished Date: 2025-07-11T00:00:00.000Z\\nAuthor: \\nImage: https://i1.rgstatic.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks/links/68715efc6e247f362b18bd9f/largepreview.png\\nFavicon: + https://c5.rgstatic.net/m/42199702882742/images/favicon/favicon-32x32.png\\nExtras: + None\\nSubpages: None\\nText: - [Home] \\n- [Computing] \\n- [Computer Science] + \\n- [Autonomics] \\n\\nArticlePDF Available\\n\\n# The Rise of Autonomous AI + Agents: Automating Complex Tasks\\n\\n- July 2025\\n- [International Journal + of Artificial Intelligence for Science (IJAI4S)] 1(2)\\n\\nDOI: [10.63619/ijai4s.v1i2.007] + \\n\\n- License\\n- [CC BY 4.0] \\n\\nAuthors:\\n\\n[Michael Negnevitsky] \\n\\n[Michael + Negnevitsky] \\n\\n- This person is not on ResearchGate, or hasn't claimed this + research yet.\\n\\n\\n[Download full-text PDF] \\n\\n[Read full-text] \\n\\n[Download + citation] \\n\\nCopy link Link copied\\n\\n[Read full-text] [Download citation] + \\nCopy link Link copied\\n\\n## Abstract and Figures\\n\\nThe emergence of + autonomous AI agents represents a transformative leap in the evolution of artificial + intelligence. These intelligent systems, capable of independently perceiving + environments, making decisions, learning from experience, and executing multi-step + actions without continuous human oversight, are redefining the boundaries of + what machines can accomplish. Unlike traditional rule-based or supervised AI + systems, autonomous agents integrate deep learning, reinforcement learning, + natural language processing, and multi-modal decision frameworks to solve complex, + dynamic, and often ambiguous real-world problems. This paper explores the technological + underpinnings, capabilities, applications, and implications of autonomous AI + agents. It critically examines their deployment in sectors such as healthcare, + finance, cybersecurity, logistics, manufacturing, education, and scientific + research. Furthermore, it addresses the ethical, legal, and socio-technical + challenges arising from the increasing autonomy of machines, offering a roadmap + for responsible innovation. Ultimately, autonomous AI agents are not merely + tools\u2014they are collaborators in a new era of intelligent automation.\\n\\n[Training + and deployment pipeline of autonomous AI agents, from task specification to + safe real-world execution.\\\\\\n\\\\\\n\u2026] \\n\\n[Human-AI symbiosis: combining + human intuition and ethics with AI scalability and speed to enable collaborative + discovery and decision-making.\\\\\\n\\\\\\n\u2026] \\n\\nFigures - available + via license: [Creative Commons Attribution 4.0 International] \\n\\nContent + may be subject to copyright.\\n\\n**Discover the world's research**\\n\\n- 25+ + million members\\n- 160+ million publication pages\\n- 2.3+ billion citations\\n\\n[Join + for free] \\n\\nAvailable via license: [CC BY 4.0] \\n\\nContent may be subject + to copyright.\\n\\nInternationalJournalofArti\uFB01cialIntelligenceforScienceAutomatingComplexTasks\\n\\nTheRiseofAutonomousAIAgents:\\n\\nAutomatingComplexTasks\\n\\nMichaelNegnevitsky1,\u2217\\n\\n1OxfordUniversity,OxfordOX12JD,UK\\n\\nCorrespondingauthor:MichaelNegnevitsky(e-mail:michaelnegnevitsky@gmail.com).\\n\\nDOI:https://doi.org/10.63619/ijai4s.v1i2.007\\n\\nThisworkislicensedunderaCreativeCommonsAttribution4.0InternationalLicense(CCBY4.0).Publishedbythe\\n\\nInternationalJournalofArti\uFB01cialIntelligenceforScience(IJAI4S).\\n\\nManuscriptreceivedMay30,2025;revisedJune13,2025;publishedJuly11,2025.\\n\\nAbstract:TheemergenceofautonomousAIagentsrepresentsatransformativeleapintheevolutionof\\n\\narti\uFB01cialintelligence.Theseintelligentsystems,capableofindependentlyperceivingenvironments,making\\n\\ndecisions,learningfromexperience,andexecutingmulti-stepactionswithoutcontinuoushumanoversight,are\\n\\nrede\uFB01ningtheboundariesofwhatmachinescanaccomplish.Unliketraditionalrule-basedorsupervisedAI\\n\\nsystems,autonomousagentsintegratedeeplearning,reinforcementlearning,naturallanguageprocessing,and\\n\\nmulti-modaldecisionframeworkstosolvecomplex,dynamic,andoftenambiguousreal-worldproblems.This\\n\\npaperexploresthetechnologicalunderpinnings,capabilities,applications,andimplicationsofautonomousAI\\n\\nagents.Itcriticallyexaminestheirdeploymentinsectorssuchashealthcare,\uFB01nance,cybersecurity,logistics,\\n\\nmanufacturing,education,andscienti\uFB01cresearch.Furthermore,itaddressestheethical,legal,andsocio-\\n\\ntechnicalchallengesarisingfromtheincreasingautonomyofmachines,offeringaroadmapforresponsible\\n\\ninnovation.Ultimately,autonomousAIagentsarenotmerelytools\u2014theyarecollaboratorsinaneweraof\\n\\nintelligentautomation.\\n\\nKeywords:AutonomousAIagents,intelligentautomation,reinforcementlearning,multi-agentsystems,task\\n\\nautomation,arti\uFB01cialgeneralintelligence,ethicalAI,autonomousdecision-making,AIplanning,agent-based\\n\\nmodeling.\\n\\n1.Introduction\\n\\nThe21stcenturyhaswitnessedunprecedentedadvancementsinarti\uFB01cialintelligence(AI),transforming\\n\\nindustries,economies,anddailylife.Amongtheseinnovations,theemergenceofautonomousAIagents\\n\\nmarksaparadigmshift\u2014notmerelyincomputationalcapability,butinthedelegationofcomplexcognitive\\n\\ntaskstomachines\\\\[1\\\\].Theseagents,whichcanindependentlyperceiveenvironments,makecontext-aware\\n\\ndecisions,learnfromexperience,andexecutegoal-directedactions,arerapidlyrede\uFB01ningwhatconstitutes\\n\\nautomationinthemodernworld\\\\[?\\\\].\\n\\nUnliketraditionalnarrowAIsystemsthatoperatewithinstatic,rule-basedframeworksorrequire\\n\\ncontinuoushumanoversight,autonomousagentsexhibitahighdegreeofautonomy,adaptability,and\\n\\ngeneralization.Theyarecapableofreal-timereasoning,dynamicplanning,andlifelonglearninginopen-\\n\\nended,unpredictableenvironments\\\\[2\\\\],\\\\[3\\\\].Forinstance,self-drivingvehiclesnavigatechaotictraf\uFB01c,\\n\\nroboticsurgeonsmakeintra-operativedecisions,andlanguageagentsengageincomplex,multi-turndia-\\n\\nlogues\u2014allwithminimalornohumanintervention\\\\[4\\\\].Thesedevelopmentsrepresentnotjustengineering\\n\\nmilestones,buttheinitialfoundationsofarti\uFB01cialgeneralintelligence(AGI)\u2014aformofintelligencethat\\n\\ncan\uFB02exiblyperformawiderangeofcognitivetasksacrossdomains\\\\[5\\\\],\\\\[6\\\\].\\n\\nTheriseofautonomousagentsalsore\uFB02ectsabroaderconvergenceofAIsub\uFB01elds,includingdeep\\n\\nreinforcementlearning,multi-agentsystems,neuro-symbolicreasoning,andlargelanguagemodels\\\\[7\\\\],\\n\\n\\\\[?\\\\].Theseintegrationshaveenabledagentstohandlenotonlyphysicaltasksinroboticsandlogistics,\\n\\nVol.01,No.02,June2025Page1\\n\\nInternationalJournalofArti\uFB01cialIntelligenceforScienceAutomatingComplexTasks\\n\\nbutalsoabstractreasoningtaskssuchaslegaldocumentdrafting,scienti\uFB01chypothesisgeneration,and\\n\\nadaptiveeducationdelivery.\\n\\nHowever,thistechnologicalleapalsobringsprofoundsocietalandethicalchallenges\\\\[?\\\\].Delegating\\n\\ndecisionstonon-humanentitiesraisescriticalconcerns:Howdoautonomousagentslearn,adapt,andmake\\n\\ndecisionsinhigh-stakesenvironments?Whatdomainsaretheymostsuitedfor\u2014andwhereshouldhuman\\n\\ncontrolremaincentral?Howcanweensuretheseagentsbehaveinwaysalignedwithhumanvalues,\\n\\nespeciallywhentheiractionsaffectsafety,justice,orequity?Whatregulatory,technical,andgovernance\\n\\nframeworksarerequiredtomanagethedeploymentofsuchintelligentsystems?\\n\\nThispaperoffersacomprehensiveexaminationofthearchitecture,applications,trainingmethodology,\\n\\nandethicalimplicationsofautonomousAIagents.Throughillustrativeusecases,comparativeanalyses,\\n\\nandfutureoutlooks,weaimtounderstandnotonlywhatthesesystemscando,butalsowhattheyshould\\n\\ndo\u2014asintelligentcollaboratorsinaworldincreasinglyshapedbymachineagency.\\n\\n2.TheArchitectureofAutonomousAgents\\n\\nAutonomousagentsareintelligentsystemscapableofperceivingtheirenvironment,makingdecisions,\\n\\ntakingactions,andadaptingovertimewithoutcontinuoushumanintervention\\\\[8\\\\].Theirarchitectureistyp-\\n\\nicallyorganizedintomodularandhierarchicallayers,eachresponsiblefordistinctaspectsoffunctionality\\n\\n\\\\[9\\\\].Thislayeredapproachenhancesinterpretability,modulardevelopment,andscalability.Wedescribethe\\n\\nfourprimarylayersofanautonomousagentsystem:perception,cognition,action,andmemory/adaptation\\n\\n\\\\[10\\\\].\\n\\nFig.1.LayeredarchitectureofanautonomousAIagentsystem,illustratingperception,cognition,action,and\\n\\nmemorycomponents.\\n\\nVol.01,No.02,June2025Page2\\n\\nInternationalJournalofArti\uFB01cialIntelligenceforScienceAutomatingComplexTasks\\n\\n2.1.PerceptionLayer\\n\\nTheperceptionlayerservesasthesensoryinterfacebetweentheagentanditsenvironment.Ittransforms\\n\\nrawdataintostructuredrepresentationsthathigher-levelmodulescaninterpretandreasonover\\\\[11\\\\].\\n\\n\u2022ComputerVision:Enablestheagenttounderstandvisualinput,includingobjectdetection,scene\\n\\nsegmentation,motiontracking,andspatiallayoutanalysis.Forexample,adronemayidentifyroads,\\n\\nhumans,orwildlifeusingYOLOorMaskR-CNN.\\n\\n\u2022NaturalLanguageProcessing(NLP):Allowstheagenttointerprettextualorspokeninstructions,\\n\\nconductdialogue,andextractsemanticmeaning.Applicationsincludelanguage-guidednavigation\\n\\nandcollaborativetaskexecutionwithhumans.\\n\\n\u2022SensorFusion:Combinesdatafrommultiplemodalities\u2014e.g.,LiDAR,RGBcameras,thermal\\n\\nsensors,radar,andmicrophones\u2014tobuildarobustandredundantperceptionsystemthatimproves\\n\\naccuracyinuncertainenvironments.\\n\\nThislayerensurestheagenthasacoherent,real-timeunderstandingofitssurroundings.\\n\\n2.2.CognitiveLayer\\n\\nThecognitivelayeristhe\u201Cbrain\u201Doftheagent.Itinterpretssensoryinputs,generatesinternalgoals,reasons\\n\\naboutconsequences,andchoosesactions\\\\[12\\\\].\\n\\n\u2022ReinforcementLearning(RL):Enablesagentstolearnoptimalpoliciesthroughtrial-and-error\\n\\ninteractionwiththeenvironment.Thisiswidelyusedinautonomousdriving,game-playing,and\\n\\nroboticcontrol.\\n\\n\u2022Meta-learning:Alsoknownas\u201Clearningtolearn,\u201Dthisallowsagentstorapidlyadapttonewtasks\\n\\norenvironmentswithminimaldata,enhancingtheirgeneralizationcapabilities.\\n\\n\u2022PlanningandScheduling:ClassicalAItechniquessuchasA\\\\*,MonteCarloTreeSearch(MCTS),\\n\\norPDDL-basedplannersareusedtogeneratemulti-stepactionplansunderconstraints.\\n\\n\u2022Neural-SymbolicIntegration:Combinesneuralnetworks(forperceptionandlearning)withsymbolic\\n\\nreasoning(e.g.,logicrules,knowledgegraphs)toachieveboth\uFB02exibilityandinterpretability.\\n\\nTogether,thesetechniquesenabletheagenttomakeinformed,strategicdecisionsindynamicenviron-\\n\\nments.\\n\\n2.3.ActionLayer\\n\\nOnceadecisionhasbeenmade,theactionlayerisresponsibleforphysicallyorvirtuallyexecutingthat\\n\\ndecision\\\\[13\\\\].Itactsastheinterfacebetweencognitionandtheexternalworld.\\n\\n\u2022RoboticActuation:Inembodiedagents,thisinvolvesmotorcommandstomanipulators,drones,or\\n\\nvehicles,enablinglocomotion,manipulation,andinteractionwithobjects.\\n\\n\u2022API-basedExecution:Insoftwareagents(e.\\nSummary: + None\\n\\n\\nTitle: Developments in AI Agents: Q1 2025 Landscape Analysis \u2014 + The Science of Machine Learning & AI\\nURL: https://www.ml-science.com/blog/2025/4/17/developments-in-ai-agents-q1-2025-landscape-analysis\\nID: + https://www.ml-science.com/blog/2025/4/17/developments-in-ai-agents-q1-2025-landscape-analysis\\nScore: + None\\nPublished Date: 2025-04-17T00:00:00.000Z\\nAuthor: \\nImage: http://static1.squarespace.com/static/5acbdd3a25bf024c12f4c8b4/t/68111511dca49a402d94e879/1745949980281/AI+Agents.png?format=1500w\\nFavicon: + https://images.squarespace-cdn.com/content/v1/5acbdd3a25bf024c12f4c8b4/1600024348668-O52EU3A2Y2J0LZUUTBMZ/favicon.ico?format=100w\\nExtras: + None\\nSubpages: None\\nText: Developments in AI Agents: Q1 2025 Landscape Analysis + — The Science of Machine Learning & AI\\n# [The Science of Machine + Learning & AI] \\nMathematics - Data Science - Computer Science\\n[**Blog**] + **Special**:[**The Accelerating Evolution of Artificial Intelligence**] \\n[ARC-AGI + MODELS LEADERBOARD] \\nCopyright \xA92016-2026[Don Cowan] All Rights Reserved\\nMathematical + Notation Powered by[CodeCogs] \\n[![Creative Commons License]] \\nThis work + is licensed under a[Creative Commons Attribution 4.0 International License].\\n[Blog + RSS] \\n[\\n![] \\n] \\n[About the Author] \\n[\\n] \\n![] \\n# [Developments + in AI Agents: Q1 2025 Landscape Analysis] \\n[April 29, 2025] in[Artificial + Intelligence] \\n## **Executive Summary**\\nThe first quarter of 2025 marked + a period of intense activity and significant evolution in the field of Artificial + Intelligence (AI) agents.\\n* Characterized by a**rapid transition from research + concepts to tangible products**, this timeframe solidified the notion of 2025 + as a pivotal \\\"year of the agent,\\\" albeit one accompanied by considerable + hype and persistent challenges. (1)\\n* The definition of**AI agents increasingly + centered on systems, typically built upon Large Language Models (LLMs)**, capable + of autonomous task execution through planning, reasoning, and the crucial ability + to utilize external tools. (4)\\n* **Major technology companies, including OpenAI, + Google, Microsoft, Anthropic, and Meta, accelerated the push towards commercialization**, + unveiling a suite of new agent platforms, specialized agents targeting specific + workflows (e.g., software development, sales, security, research), and underlying + infrastructure components. (7)\\n* **Research simultaneously advanced**, focusing + on enhancing core capabilities like reasoning, fostering human-AI collaboration, + and developing multi-agent systems (MAS) capable of complex interactions. (12)\\n* + Concurrently,**the discourse surrounding the challenges, limitations, and ethical + implications intensified**. Concerns regarding reliability, safety, control, + bias, and privacy became more prominent as agent autonomy increased, prompting + calls for robust governance frameworks and dedicated agent infrastructure. (16)\\n* + While the potential for AI agents to automate tasks and augment human capabilities + is increasingly evident,**significant hurdles in achieving dependable, safe, + and ethically aligned performance remained a primary focus**for the industry + moving forward. (20)\\n## **Defining the AI Agent Landscape (Early 2025)**\\n### + **Conceptual Evolution and Definitions**\\nThe understanding and definition + of AI agents underwent a notable transformation leading into early 2025, largely + catalyzed by advancements in LLMs. Historically, the concept of an agent was + rooted in computer science and cognitive science, often defined broadly as anything + capable of perceiving its environment through sensors and acting upon it through + actuators. Early work emphasized properties like autonomy, social ability, reactivity, + and proactivity. (4)\\nHowever, by the first quarter of 2025, the prevailing + conceptualization became intrinsically linked to the capabilities offered by + modern LLMs. (4)\\nWhile no single, universally accepted definition existed + (24), a strong consensus emerged viewing AI agents as systems or programs capable + of*autonomously performing tasks*on behalf of a user or another system. This + autonomy involved designing their own workflows and utilizing available tools + to achieve specified goals. (1)\\nThis modern understanding clearly differentiated + AI agents from simpler AI applications like chatbots or traditional automation. + Unlike chatbots primarily focused on conversational responses based on pre-existing + data (6), or traditional systems executing predefined algorithms within constraints + (4), early 2025 agents were characterized by their ability to plan, reason, + learn, and interact dynamically with their environment to accomplish complex, + often multi-step objectives. (4) They leveraged LLMs as core reasoning or processing + components but augmented them with essential capabilities like memory, planning + modules, and mechanisms for tool use. (4) This augmentation was crucial, addressing + inherent LLM limitations such as statelessness (lack of memory between interactions) + and the inability to directly interact with or affect the external world. (17) + The term \\\"agentic AI\\\" also gained traction, often referring to systems + composed of multiple, potentially collaborative, AI agents working towards a + common objective. (28)\\nThe evolution of the definition directly mirrors the + enabling power of LLMs. While the theoretical foundations of agency existed + for decades (4), the practical ability of LLMs to understand complex instructions, + reason about tasks, and generate plausible action steps provided the foundation + for building agents that could move beyond theoretical constructs and execute + tasks in the real world (primarily digital environments). The focus shifted + from defining agency based on abstract properties to defining it based on the + functional capabilities enabled by integrating LLMs with planning, memory, and + tool-using mechanisms. This practical turn, driven by LLM advancements, underpins + the surge in agent development and deployment observed in early 2025.\\n### + **Core Capabilities**\\nThe AI agents emerging in early 2025 were defined by + a set of core capabilities that enabled their autonomous and goal-directed behavior. + These capabilities, often working in concert and built upon LLM foundations, + distinguished them from prior AI systems:\\n* **Autonomy:**This was perhaps + the most central characteristic. Agents were expected to operate with minimal + human intervention or continuous oversight, capable of initiating and completing + tasks independently.1 The degree of autonomy existed on a spectrum. Some systems + were semi-autonomous, requiring human oversight for critical decisions (26), + while others aimed for, or raised concerns about, higher levels of autonomy, + potentially even \\\"fully autonomous\\\" operation where agents could act without + predefined constraints, a prospect met with significant safety concerns. (18)\\n* + **Learning & Adaptation:**Agents were designed to improve their performance + over time. This involved learning from environmental feedback, past interactions, + and accumulated experience.3 Mechanisms like reinforcement learning allowed + agents to refine strategies based on outcomes. (26) The ability to self-evaluate + and correct errors was also highlighted. (32) Memory, both short-term (for maintaining + context within a task) and long-term (for retaining historical data and preferences), + was crucial for adaptation and personalization. (26)\\n* **Reasoning & Decision-Making:**Agents + needed to process perceived information, analyze data, make judgments, and select + appropriate actions to achieve their goals. (1) LLMs often served as the core + \\\"reasoning engine\\\" (32), interpreting inputs and generating potential + action plans. The period saw the emergence of specialized models explicitly + optimized for reasoning tasks, aiming to enhance agent decision-making capabilities. + (12)\\n* **Interaction:**Agents were defined by their ability to engage with + their environment. This involved perception \u2013gathering information through + various inputs like text, voice, images, sensors, data feeds, or APIs (1) \u2013and + action \u2013influencing the environment through outputs like generating text, + sending commands to software, executing code, or making API calls. (1) Crucially, + this interaction extended beyond simple user dialogue to include engagement + with external systems, tools, data sources, and potentially other agents or + humans. (1)\\n* **Planning & Goal-Directedness:**Agents were designed to + be proactive and goal-oriented. (24) This required the ability to understand + a high-level objective, decompose it into smaller, manageable steps (task decomposition), + formulate a sequence of actions (planning), and then execute that plan. (5) + The capability for long-term planning, reasoning about extended sequences of + actions, was also an area of development and concern. (12)\\n* **Tool Usage:**A + key enabler of agent functionality was the ability to utilize external tools. + These could include APIs for accessing external services (like weather data, + booking systems, or databases), web search capabilities, code interpreters, + calculators, or even invoking other AI models or agents. (4) Tool use allowed + agents to overcome the inherent limitations of their base LLMs (e.g., accessing + real-time information, performing precise calculations, interacting with specific + software) and act effectively in the environment. (6)\\nThe confluence of these + capabilities defined the ambition for AI agents in early 2025. The objective + was clear: to transition AI from being passive responders or pattern recognizers + to becoming active, autonomous participants capable of understanding goals, + formulating plans, interacting with the digital world through tools, learning + from outcomes, and ultimately achieving complex objectives with reduced human + guidance. This represented a significant functional leap, enabling the automation + of tasks previously requiring human cognition and intervention.\\n### **C. Agent + Architectures and Components**\\nThe design of AI agents in early 2025 commonly + followed specific architectural patterns, reflecting attempts to harness the + power of LLMs while mitigating their weaknesses. Agents were frequently described + as*compound systems*. (24) This meant they were not monolithic entities but + rather architectures built around a foundation model (typically an LLM) augmented + by external resources or modules, often referred to as \\\"scaffolding\\\". + (5) This scaffolding provided capabilities the base LLM lacked, such as persistent + memory, structured planning, and interfaces for tool\\nSummary: None\\n\\n\\nTitle: + 2025 Buildings Show: AI agents Become 'Talk of the Town'\\nURL: https://news.constructconnect.com/2025-buildings-show-ai-agents-become-talk-of-the-town\\nID: + https://news.constructconnect.com/2025-buildings-show-ai-agents-become-talk-of-the-town\\nScore: + None\\nPublished Date: 2025-12-11T00:00:00.000Z\\nAuthor: Don Wall\\nImage: + https://news.constructconnect.com/hubfs/AI%20agent%20supporting%20business%20processes%20with%20automation%20technology.%20Generative%20artificial%20intelligence%20analyzing%20data%20and%20predicting%20future.%20Businessman%20using%20laptop..jpg\\nFavicon: + https://news.constructconnect.com/hubfs/assets/images/ccfav.png\\nExtras: None\\nSubpages: + None\\nText: 2025 Buildings Show: AI agents Become \u2018Talk of the Town\u2019\\n[![ConstructConnect + News]] \\n[Follow us on LinkedIn] [Watch us on YouTube] [Follow us on Instagram] + [Follow us on Facebook] [Follow us on X] \\n[Find Projects Now] \\n[![ConstructConnect + News]] \\n****[****] \\n## [![ConstructConnect News]] \\n[Follow us on LinkedIn] + [Watch us on YouTube] [Follow us on Instagram] [Follow us on Facebook] [Follow + us on X] \\n[\u2715] \\n* [Featured] \\n* [Economy] \\n* [Industry] \\n* [Project + Spotlight] \\n* [About] \\n[**] \\n[Industry News & Trends] [Featured] # + 2025 Buildings Show: AI agents Become \u2018Talk of the Town\u2019\\nBy Don + Wall**December 11, 2025\\nShare\\n* [\\n] \\n* [\\n] \\n* [\\n] \\n#### KEY + POINTS\\n* **Agentic AI is revolutionizing construction by enabling autonomous + systems to plan, monitor, and update tasks, offering multi-step reasoning and + decision-making beyond traditional AI workflows and Large Language Models (LLMs).**\\n* + **Tools simplify software development by generating, modifying, and testing + code from written requirements, reducing manual coding time and empowering non-coders + to create apps and automations.**\\n* **While agentic AI accelerates innovation, + one expert suggested that users must maintain a \u201Chuman in the loop.\u201D**\\nWhat + a difference a year makes in the world of[construction tech].\\nDelegates attending + day two of the recent Buildings Show in Toronto heard how agentic AI is rapidly + emerging as construction\u2019s third wave of[AI], moving beyond Large Language + Models (LLMs) and[automated workflows] to create a world of autonomous systems + that plan, monitor, and update[project tasks], as well aswrite code.\\nDigital + innovation specialist Mahir Dheendsa of[Amrize] (formerly Lafarge), host of + a session titled \u201CAI Agents and the Robotic Workforce: Reimagining the + Future of Construction,\u201Dsaid the use of agentic AI is accelerating. So + rapidly that the preview of the session he wrote for the show program months + ago was already out of date.\\n![DW-buildings-show-AI-agencyweb] \\nDigital + innovation specialist Mahir Dheendsa of Amrize spoke about AI agents at the + Buildings Show in Toronto, Dec. 4. Dheendsa took a moment to outline how Amrize + represents the new brand for Lafarge. Image: Don Wall\\n\u201CAgentic AI has + been the talk of the town and the major headline in the AI space this year,\u201D + said Dheendsa. \u201CWe haven\u2019t really seen any major advancements in robotics, + so I\u2019m willing to sort of skip over that.\u201D\\nPlatforms offering AI + agents, such as n8n, Gemini Agent Studio, Make.com and Zapier, democratize agent + development, Dheendsa said, enabling[construction professionals] to build custom + automations without extensive coding expertise.\\nGenerative development tools + like Lovable AI and Cursor take it a step further by generating, modifying, + and testing software directly from written requirements, dramatically[reducing + the time] typically spent on planning and manual coding.\\n[![Econ Ads_Banners3.jpg.-2]] + \\n#### Multi-Step Reasoning\\nLLMs offer a single-step response with no tools + or actions, he explained. AI workflows follow a fixed sequence of steps and + can connect multiple apps, but with no reasoning or judgment. An AI agent acts + as autonomous intelligence, utilizing multi-step reasoning and tools and APIs + to evaluate, retry, and create.\\nDheendsa offered an example of a supply chain + coordination function with a problem: a crew of 20 is standing around doing + nothing because the steel beams haven\u2019t arrived.\\nThe agent can track + supplier GPS data and manufacturing status updates in real-time, potentially + predicting a delay three days in advance and suggesting rescheduling the crane + rental.\\nAnother example is with a change order. A[subcontractor] might send + bills for \u201Cextra work,\u201D and nobody remembers if it was approved, said + Dheendsa. The agent scans thousands of emails, contracts, and meeting minutes + automatically and verifies if the extra work was actually part of the original + contract scope.\\n\u201CAn AI agent has the ability to make decisions,\u201D + said Dheendsa. \u201CThe[AI agent] monitors and reports. It will detect an issue, + and it will decide if this is an issue worthy of being brought up based on a + criteria that you can provide it.\\n\u201CAnd then it will figure out which + team or which person needs to be notified.\u201D\\nHe has used Replit, Lovable, + and Cursor, and all can work to generate responses to RFPs, he said.\\n\u201CYou + just copy the[RFP] into the Lovable app, and the software is done,\u201D said + Dheendsa.\\n\u201CIf you have an idea, if you want to build an app, you don\u2019t + even need to know how to code. You just put your idea in there, and it\u2019ll + just generate you an app or a web app, website, whatever you need.\u201D\\nHuman + in the Loop\\nDheendsa stressed that users still need to keep a \u201Chuman + in the loop.\u201D\\n\u201CIt is prone to making mistakes. It is prone to hallucinations, + (but) it\u2019s getting better.\u201D\\nWith Gemini 3, for example, \u201CI\u2019d + say it\u2019s 85 per cent there. It\u2019s very good.\u201D\\nAnother caveat + is that users should not venture into AI solutions if the original process is + flawed: \u201CDon\u2019t automate broken processes,\u201D Dheendsa said.\\nAmong + various risks he identified, \u201Cover-trust risk\u201D is highly dangerous, + he said. Al confidence does not equal correctness. Always verify technical outputs + and cross-check against standards and codes, he said.\\nMicrosoft has reported[construction + jobs] in the field are among the least threatened by AI, Dheendsa noted in an + interview, for a couple of reasons.\\nFor one, dredge operators or crane operators + represent a lot of \u201Cmoving parts.\u201D Second, he said, there hasn\u2019t + been enough data collected in the field for[AI to automate].\\n\u201CThe low-hanging + fruit is here with the administrative stuff, like data entry and all the[repetitive + workflows]. So let\u2019s just start there,\u201D said Dheendsa.\\n\u201CThe + barrier to entry is lower than ever before\u2026I encourage people to start + doing that, because even if it\u2019s not going to get you 100 per cent of the + way there, it\u2019ll at least get you 80 per cent.\u201D\\n#### About ConstructConnect\\nAt[ConstructConnect], + our software solutions provide the information construction professionals need + to start every project on a solid foundation. For more than 100 years, our insights + and market intelligence have empowered commercial firms, manufacturers, trade + contractors, and architects to make data-driven decisions and maximize productivity.\\nConstructConnect + is a business unit of[Roper Technologies] (Nasdaq: ROP), part of the Nasdaq + 100, S&P 500, and Fortune 1000.\\nFor more information, visit[constructconnect.com] + \\n![Don Wall] \\n[Don Wall] \\nDon Wall is a freelance journalist that covers + construction including economics, architecture, resources, infrastructure, health + and safety and new projects. He has worked for over 35 years as a journalist, + including as a staff writer for over nine years with Canada's Daily Commercial + News by ConstructConnect.\\n[Previous] \\n### [$150M Deal Fails at HyperGrid + AI Project] \\n[Next] \\n### [Clarksville, Tennessee's Business Boom Grows with + New $90 Million Investment] \\n### Subscribe Now.\\nConstructConnect News. Construction + news. Built for you, daily.\\n### ### Topics\\n* [Industry News & Trends] + \\n* [Featured] \\n* [Economy] \\n* [Project Spotlight] \\n* [Construction Starts] + \\n* [Tariffs] \\n* [Construction Starts Forecast] \\n* [Project Stress Index] + \\n* [Building Product Manufacturing] \\n* [Construction Economics & Finance] + \\n#### Trending News\\n[![image]] \\n### [$150M Deal Fails at HyperGrid AI + Project] \\n[![image]] \\n### [2025 Buildings Show: AI agents Become \u2018Talk + of the Town\u2019] \\n[![image]] \\n### [Clarksville, Tennessee's Business Boom + Grows with New $90 Million Investment] \\n#### Trending News\\n1. [![image]] + \\n### [$150M Deal Fails at HyperGrid AI Project] \\n2. [![image]] \\n### [2025 + Buildings Show: AI agents Become \u2018Talk of the Town\u2019] \\n3. [![image]] + \\n### [Clarksville, Tennessee's Business Boom Grows with New $90 Million Investment] + \\n#### RELATED NEWS\\n[![image]] [Project Spotlight] \\n### [Wastewater Facility + in Vero Beach to Break Ground] \\nConstruction on a new $164 million wastewater + treatment facility at Vero Beach, Florida\u2019s regional...\\nBy[Marshall Benveniste] + \\nMay 29 2025\\n[![image]] [Featured] \\n### [The More Things Change Portion + of the Construction Marketplace] \\nAlex Carrick explores the changes in nonresidential + construction as new sources of demand for...\\nBy[Alex Carrick] \\nMay 14 2025\\n[![image]] + [Industry News & Trends] \\n### [Autumn Construction Starts Forecast: Civil + Construction to Lead, Megaprojects Lift Outlook] \\nIn our Autumn 2025 Construction + Starts Forecast, ConstructConnect has slightly upgraded our 2025...\\nBy[Michael + Guckes, Chief Economist] \\nAug 19 2025\\nSummary: None\\n\\n\\nTitle: Autonomous + Deep Agent\\nURL: https://arxiv.org/abs/2502.07056\\nID: https://arxiv.org/abs/2502.07056\\nScore: + None\\nPublished Date: 2025-02-10T00:00:00.000Z\\nAuthor: [Submitted on 10 Feb + 2025]\\nImage: /static/browse/0.3.4/images/arxiv-logo-fb.png\\nFavicon: https://arxiv.org/static/browse/0.3.4/images/icons/favicon-32x32.png\\nExtras: + None\\nSubpages: None\\nText: [2502.07056] Autonomous Deep Agent\\n[Skip to + main content] \\n[![Cornell University]] \\nWe gratefully acknowledge support + from the Simons Foundation,[member institutions], and all contributors.[Donate] + \\n[] \\n[![arxiv logo]] >[cs] >arXiv:2502.07056\\n[Help] |[Advanced Search] + \\nAll fieldsTitleAuthorAbstractCommentsJournal referenceACM classificationMSC + classificationReport numberarXiv identifierDOIORCIDarXiv author IDHelp pagesFull + text\\nSearch\\n[![arXiv logo]] \\n[![Cornell University Logo]] \\nopen search\\nGO\\nopen + navigation menu\\n# Computer Science \\\\> Artificial Intelligence\\n**arXiv:2502.07056**(cs)\\n[Submitted + on 10 Feb 2025]\\n# Title:Autonomous Deep Agent\\nAuthors:[Amy Yu],[Erik Lebedev],[Lincoln + Everett],[Xiaoxin Chen],[Terry Chen] \\nView a PDF of the paper titled Autonomous + Deep Agent, by Amy Yu and 3 other authors\\n[View PDF] [HTML (experimental)] + > > Abstract:\\n> This technical brief introduces Deep Agent, an advanced autonomous + AI system designed to manage complex multi-phase tasks through a novel hierarchical + task management architecture. The system's foundation is built on our Hierarchical + Task DAG (HTDAG) framework, which dynamically decomposes high-level objectives + into manageable sub-tasks while rigorously maintaining dependencies and execution + coherence. Deep Agent advances beyond traditional agent systems through three + key innovations: First, it implements a recursive two-stage planner-executor + architecture that enables continuous task refinement and adaptation as circumstances + change. Second, it features an Autonomous API & Tool Creation (AATC) system + that automatically generates reusable components from UI interactions, substantially + reducing operational costs for similar tasks. Third, it incorporates Prompt + Tweaking Engine and Autonomous Prompt Feedback Learning components that optimize + Large Language Model prompts for specific scenarios, enhancing both inference + accuracy and operational stability. These components are integrated to form + a service infrastructure that manages user contexts, handles complex task dependencies, + and orchestrates end-to-end agentic workflow execution. Through this sophisticated + architecture, Deep Agent establishes a novel paradigm in self-governing AI systems, + demonstrating robust capability to independently handle intricate, multi-step + tasks while maintaining consistent efficiency and reliability through continuous + self-optimization. Subjects:|Artificial Intelligence (cs.AI); Machine Learning + (cs.LG)|\\nACMclasses:|I.2.6; I.2.7|\\nCite as:|[arXiv:2502.07056] [cs.AI]|\\n|(or[arXiv:2502.07056v1] + [cs.AI]for this version)|\\n|[https://doi.org/10.48550/arXiv.2502.07056] \\nFocus + to learn more\\narXiv-issued DOI via DataCite\\n|\\n## Submission history\\nFrom: + Amy Yu [[view email]]\\n**[v1]**Mon, 10 Feb 2025 21:46:54 UTC (2,085 KB)\\nFull-text + links:## Access Paper:\\nView a PDF of the paper titled Autonomous Deep Agent, + by Amy Yu and 3 other authors\\n* [View PDF] \\n* [HTML (experimental)] \\n* + [TeX Source] \\n[![license icon] view license] \\nCurrent browse context:\\ncs.AI\\n[<<prev] + | [next>>] \\n[new] |[recent] |[2025-02] \\nChange to browse by:\\n[cs] + \\n[cs.LG] \\n### References & Citations\\n* [NASA ADS] \\n* [Google Scholar] + \\n* [Semantic Scholar] \\nexport BibTeX citationLoading...\\n## BibTeX formatted + citation\\n×\\nloading...\\nData provided by:\\n### Bookmark\\n[![BibSonomy + logo]] [![Reddit logo]] \\nBibliographic Tools\\n# Bibliographic and Citation + Tools\\nBibliographic Explorer Toggle\\nBibliographic Explorer*([What is the + Explorer?])*\\nConnected Papers Toggle\\nConnected Papers*([What is Connected + Papers?])*\\nLitmaps Toggle\\nLitmaps*([What is Litmaps?])*\\nscite.ai Toggle\\nscite + Smart Citations*([What are Smart Citations?])*\\nCode, Data, Media\\n# Code, + Data and Media Associated with this Article\\nalphaXiv Toggle\\nalphaXiv*([What + is alphaXiv?])*\\nLinks to Code Toggle\\nCatalyzeX Code Finder for Papers*([What + is CatalyzeX?])*\\nDagsHub Toggle\\nDagsHub*([What is DagsHub?])*\\nGotitPub + Toggle\\nGotit.pub*([What is GotitPub?])*\\nHuggingface Toggle\\nHugging Face*([What + is Huggingface?])*\\nLinks to Code Toggle\\nPapers with Code*([What is Papers + with Code?])*\\nScienceCast Toggle\\nScienceCast*([What is ScienceCast?])*\\nDemos\\n# + Demos\\nReplicate Toggle\\nReplicate*([What is Replicate?])*\\nSpaces Toggle\\nHugging + Face Spaces*([What is Spaces?])*\\nSpaces Toggle\\nTXYZ.AI*([What is TXYZ.AI?])*\\nRelated + Papers\\n# Recommenders and Search Tools\\nLink to Influence Flower\\nInfluence + Flower*([What are Influence Flowers?])*\\nCore recommender toggle\\nCORE Recommender*([What + is CORE?])*\\n* Author\\n* Venue\\n* Institution\\n* Topic\\nAbout arXivLabs\\n# + arXivLabs: experimental projects with community collaborators\\narXivLabs is + a framework that allows collaborators to develop and share new arXiv features + directly on our website.\\nBoth individuals and organizations that work with + arXivLabs have embraced and accepted our values of openness, community, excellence, + and user data privacy. arXiv is committed to these values and only works with + partners that adhere to them.\\nHave an idea for a project that will add value + for arXiv's community?[**Learn more about arXivLabs**].\\n[Which authors of + this paper are endorsers?] |[Disable MathJax] ([What is MathJax?])\\nSummary: + None\\n\\nResolved Search Type: neural\\nCostDollars: total=0.015\\n - search: + {'neural': 0.005}\\n - contents: {'text': 0.01}\"}],\"model\":\"gpt-4o-mini\",\"tool_choice\":\"auto\",\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"exa_search_tool\",\"description\":\"Search + the internet using Exa\",\"strict\":true,\"parameters\":{\"properties\":{\"search_query\":{\"description\":\"Mandatory + search query you want to use to search the internet\",\"title\":\"Search Query\",\"type\":\"string\"},\"start_published_date\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}],\"default\":null,\"description\":\"Start + date for the search\",\"title\":\"Start Published Date\"},\"end_published_date\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}],\"default\":null,\"description\":\"End + date for the search\",\"title\":\"End Published Date\"},\"include_domains\":{\"anyOf\":[{\"items\":{\"type\":\"string\"},\"type\":\"array\"},{\"type\":\"null\"}],\"default\":null,\"description\":\"List + of domains to include in the search\",\"title\":\"Include Domains\"}},\"required\":[\"search_query\",\"start_published_date\",\"end_published_date\",\"include_domains\"],\"type\":\"object\",\"additionalProperties\":false}}}]}" + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '90929' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NeV4GuqRpgmKe6nQIszVRGXkIah\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413259,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"### Recent Developments in Autonomous + AI Agents (2025)\\n\\nThis summary compiles recent findings on the developments + in autonomous AI agents for the year 2025, highlighting their evolution, applications, + challenges, and implications across various industries.\\n\\n#### Key Articles + and Insights:\\n\\n1. **AI Agents in 2025: Expectations vs. Reality** \\n + \ - **Source**: [IBM](https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality) + \ \\n - **Published**: February 26, 2025 \\n - **Summary**: The article + discusses the transition from large language models (LLMs) to agentic AI, + emphasizing autonomous agents' potential to reshape work dynamics. Experts + from IBM note that while there is enthusiasm about AI agents, many current + applications are still in their infancy, requiring refinement in planning + and reasoning capabilities to reach true autonomy.\\n\\n2. **Autonomous AI + Agents Are Reshaping the Business Landscape** \\n - **Source**: [KPMG](https://kpmg.com/sk/en/insights/2025/06/autonomous-ai-agents-reshape-business-landscape.html) + \ \\n - **Published**: June 25, 2025 \\n - **Summary**: This report outlines + how agentic AI is transitioning from theory into practical use, impacting + sectors like finance and healthcare. It describes agentic AI as a \\\"digital + workforce\\\" that enhances operational efficiency and decision-making processes + through autonomous task management. \\n\\n3. **The Rise of Autonomous AI Agents: + Automating Complex Tasks** \\n - **Source**: [ResearchGate](https://www.researchgate.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks) + \ \\n - **Published**: July 11, 2025 \\n - **Summary**: This academic + paper explores the technological foundations and applications of autonomous + agents in various fields, noting their ability to autonomously resolve multi-step + tasks. It addresses the ethical considerations linked to the increasing autonomy + of these systems.\\n\\n4. **Developments in AI Agents: Q1 2025 Landscape Analysis** + \ \\n - **Source**: [The Science of Machine Learning & AI](https://www.ml-science.com/blog/2025/4/17/developments-in-ai-agents-q1-2025-landscape-analysis) + \ \\n - **Published**: April 17, 2025 \\n - **Summary**: Highlights significant + advancements in AI agent technology, including their capability for autonomous + task achievement. The article underscores ongoing challenges related to reliability, + safety, and ethical governance in AI performance.\\n\\n5. **Agentic AI in + Software Development** \\n - **Source**: [SculptSoft](https://www.sculptsoft.com/agentic-ai-in-action-how-autonomous-ai-agents-are-changing-software-development-in-2025/) + \ \\n - **Published**: October 15, 2025 \\n - **Summary**: Discusses + how autonomous AI agents are revolutionizing software development by automating + the development lifecycle\u2014planning, coding, testing, and deployment\u2014thus + enabling teams to innovate faster while reducing manual workloads.\\n\\n6. + **AI Agents Arrived in 2025 \u2013 Here\u2019s What Happened and the Challenges + Ahead** \\n - **Source**: [The Conversation](https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325) + \ \\n - **Published**: December 29, 2025 \\n - **Summary**: The article + reviews the major milestones reached in 2025, including the introduction of + protocols to enable agents to communicate and use external tools. It also + addresses the societal implications, such as the need for enhanced governance + and ethical considerations in AI deployment.\\n\\n7. **2025 Buildings Show: + AI Agents Become 'Talk of the Town'** \\n - **Source**: [ConstructConnect](https://news.constructconnect.com/2025-buildings-show-ai-agents-become-talk-of-the-town) + \ \\n - **Published**: December 11, 2025 \\n - **Summary**: At the Buildings + Show, digital innovation in construction technology showcased agentic AI's + capability to monitor and manage project tasks autonomously, streamlining + operations in the sector.\\n\\n### Conclusion\\nThe advancements in autonomous + AI agents in 2025 exhibit a shift towards integrating AI into real-world applications, + enhancing productivity across various industries. While there is significant + potential for these agents to improve efficiency and decision-making, challenges + related to ethical considerations, governance, and reliability remain pivotal + in guiding the future of AI development.\",\n \"refusal\": null,\n + \ \"annotations\": []\n },\n \"logprobs\": null,\n \"finish_reason\": + \"stop\"\n }\n ],\n \"usage\": {\n \"prompt_tokens\": 19713,\n \"completion_tokens\": + 966,\n \"total_tokens\": 20679,\n \"prompt_tokens_details\": {\n \"cached_tokens\": + 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_6c0d1490cb\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:28:22 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '42906' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: "{\"messages\":[{\"role\":\"system\",\"content\":\"You are a Planning Agent + observing execution progress. After each step completes, you analyze what happened + and decide whether the remaining plan is still valid.\\n\\nReason step-by-step + about:\\n1. What new information was learned from this step's result\\n2. Whether + the remaining steps still make sense given this new information\\n3. What refinements, + if any, are needed for upcoming steps\\n4. Whether the overall goal has already + been achieved\\n\\nBe conservative about triggering full replans \u2014 only + do so when the remaining plan is fundamentally wrong, not just suboptimal.\"},{\"role\":\"user\",\"content\":\"## + Original task\\n\\n\\n## Expected output\\n\\n\\n\\n## Just completed step 1\\nDescription: + Conduct research on recent developments in autonomous AI agents in 2025.\\nResult: + ### Recent Developments in Autonomous AI Agents (2025)\\n\\nThis summary compiles + recent findings on the developments in autonomous AI agents for the year 2025, + highlighting their evolution, applications, challenges, and implications across + various industries.\\n\\n#### Key Articles and Insights:\\n\\n1. **AI Agents + in 2025: Expectations vs. Reality** \\n - **Source**: [IBM](https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality) + \ \\n - **Published**: February 26, 2025 \\n - **Summary**: The article + discusses the transition from large language models (LLMs) to agentic AI, emphasizing + autonomous agents' potential to reshape work dynamics. Experts from IBM note + that while there is enthusiasm about AI agents, many current applications are + still in their infancy, requiring refinement in planning and reasoning capabilities + to reach true autonomy.\\n\\n2. **Autonomous AI Agents Are Reshaping the Business + Landscape** \\n - **Source**: [KPMG](https://kpmg.com/sk/en/insights/2025/06/autonomous-ai-agents-reshape-business-landscape.html) + \ \\n - **Published**: June 25, 2025 \\n - **Summary**: This report outlines + how agentic AI is transitioning from theory into practical use, impacting sectors + like finance and healthcare. It describes agentic AI as a \\\"digital workforce\\\" + that enhances operational efficiency and decision-making processes through autonomous + task management. \\n\\n3. **The Rise of Autonomous AI Agents: Automating Complex + Tasks** \\n - **Source**: [ResearchGate](https://www.researchgate.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks) + \ \\n - **Published**: July 11, 2025 \\n - **Summary**: This academic paper + explores the technological foundations and applications of autonomous agents + in various fields, noting their ability to autonomously resolve multi-step tasks. + It addresses the ethical considerations linked to the increasing autonomy of + these systems.\\n\\n4. **Developments in AI Agents: Q1 2025 Landscape Analysis** + \ \\n - **Source**: [The Science of Machine Learning & AI](https://www.ml-science.com/blog/2025/4/17/developments-in-ai-agents-q1-2025-landscape-analysis) + \ \\n - **Published**: April 17, 2025 \\n - **Summary**: Highlights significant + advancements in AI agent technology, including their capability for autonomous + task achievement. The article underscores ongoing challenges related to reliability, + safety, and ethical governance in AI performance.\\n\\n5. **Agentic AI in Software + Development** \\n - **Source**: [SculptSoft](https://www.sculptsoft.com/agentic-ai-in-action-how-autonomous-ai-agents-are-changing-software-development-in-2025/) + \ \\n - **Published**: October 15, 2025 \\n - **Summary**: Discusses how + autonomous AI agents are revolutionizing software development by automating + the development lifecycle\u2014planning, coding, testing, and deployment\u2014thus + enabling teams to innovate faster while reducing manual workloads.\\n\\n6. **AI + Agents Arrived in 2025 \u2013 Here\u2019s What Happened and the Challenges Ahead** + \ \\n - **Source**: [The Conversation](https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325) + \ \\n - **Published**: December 29, 2025 \\n - **Summary**: The article + reviews the major milestones reached in 2025, including the introduction of + protocols to enable agents to communicate and use external tools. It also addresses + the societal implications, such as the need for enhanced governance and ethical + considerations in AI deployment.\\n\\n7. **2025 Buildings Show: AI Agents Become + 'Talk of the Town'** \\n - **Source**: [ConstructConnect](https://news.constructconnect.com/2025-buildings-show-ai-agents-become-talk-of-the-town) + \ \\n - **Published**: December 11, 2025 \\n - **Summary**: At the Buildings + Show, digital innovation in construction technology showcased agentic AI's capability + to monitor and manage project tasks autonomously, streamlining operations in + the sector.\\n\\n### Conclusion\\nThe advancements in autonomous AI agents in + 2025 exhibit a shift towards integrating AI into real-world applications, enhancing + productivity across various industries. While there is significant potential + for these agents to improve efficiency and decision-making, challenges related + to ethical considerations, governance, and reliability remain pivotal in guiding + the future of AI development.\\n\\n## Remaining plan steps:\\n Step 2: Summarize + the key findings from the research conducted on autonomous AI agents.\\n\\nAnalyze + this step's result and provide your observation.\"}],\"model\":\"gpt-4o-mini\",\"response_format\":{\"type\":\"json_schema\",\"json_schema\":{\"schema\":{\"description\":\"Planner's + observation after a step execution completes.\\n\\nReturned by the PlannerObserver + after EVERY step \u2014 not just failures.\\nThe Planner uses this to decide + whether to continue, refine, or replan.\\n\\nBased on PLAN-AND-ACT (Section + 3.3): the Planner observes what the Executor\\ndid and incorporates new information + into the remaining plan.\\n\\nAttributes:\\n step_completed_successfully: + Whether the step achieved its objective.\\n key_information_learned: New + information revealed by this step\\n (e.g., \\\"Found 3 products: A, + B, C\\\"). Used to refine upcoming steps.\\n remaining_plan_still_valid: + Whether pending todos still make sense\\n given the new information. + True does NOT mean no refinement needed.\\n suggested_refinements: Minor + tweaks to upcoming step descriptions.\\n These are lightweight in-place + updates, not a full replan.\\n Example: [\\\"Step 3 should select product + B instead of 'best product'\\\"]\\n needs_full_replan: The remaining plan + is fundamentally wrong and must\\n be regenerated from scratch. Mutually + exclusive with\\n remaining_plan_still_valid (if this is True, that should + be False).\\n replan_reason: Explanation of why a full replan is needed (None + if not).\\n goal_already_achieved: The overall task goal has been satisfied + early.\\n No more steps needed \u2014 skip remaining todos and finalize.\",\"properties\":{\"step_completed_successfully\":{\"description\":\"Whether + the step achieved what it was asked to do\",\"title\":\"Step Completed Successfully\",\"type\":\"boolean\"},\"key_information_learned\":{\"default\":\"\",\"description\":\"What + new information this step revealed\",\"title\":\"Key Information Learned\",\"type\":\"string\"},\"remaining_plan_still_valid\":{\"default\":true,\"description\":\"Whether + the remaining pending todos still make sense given new information\",\"title\":\"Remaining + Plan Still Valid\",\"type\":\"boolean\"},\"suggested_refinements\":{\"anyOf\":[{\"items\":{\"type\":\"string\"},\"type\":\"array\"},{\"type\":\"null\"}],\"description\":\"Minor + tweaks to descriptions of upcoming steps (lightweight, no full replan)\",\"title\":\"Suggested + Refinements\"},\"needs_full_replan\":{\"default\":false,\"description\":\"The + remaining plan is fundamentally wrong and must be regenerated\",\"title\":\"Needs + Full Replan\",\"type\":\"boolean\"},\"replan_reason\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}],\"description\":\"Explanation + of why a full replan is needed\",\"title\":\"Replan Reason\"},\"goal_already_achieved\":{\"default\":false,\"description\":\"The + overall task goal has been satisfied early; no more steps needed\",\"title\":\"Goal + Already Achieved\",\"type\":\"boolean\"}},\"required\":[\"step_completed_successfully\",\"key_information_learned\",\"remaining_plan_still_valid\",\"suggested_refinements\",\"needs_full_replan\",\"replan_reason\",\"goal_already_achieved\"],\"title\":\"StepObservation\",\"type\":\"object\",\"additionalProperties\":false},\"name\":\"StepObservation\",\"strict\":true}},\"stream\":false}" + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '8418' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-helper-method: + - beta.chat.completions.parse + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NfCUwdK5hKLF9blS0pKpCPxWFF5\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413302,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"{\\\"step_completed_successfully\\\":true,\\\"key_information_learned\\\":\\\"Recent + developments in autonomous AI agents indicate significant progress across + various industries, demonstrating their potential to enhance productivity + and operational efficiency, though challenges in ethical governance and reliability + persist.\\\",\\\"remaining_plan_still_valid\\\":true,\\\"suggested_refinements\\\":[\\\"In + Step 2, emphasize the emerging challenges related to ethical governance and + reliability alongside the key findings.\\\"],\\\"needs_full_replan\\\":false,\\\"replan_reason\\\":null,\\\"goal_already_achieved\\\":false}\",\n + \ \"refusal\": null,\n \"annotations\": []\n },\n \"logprobs\": + null,\n \"finish_reason\": \"stop\"\n }\n ],\n \"usage\": {\n \"prompt_tokens\": + 1733,\n \"completion_tokens\": 103,\n \"total_tokens\": 1836,\n \"prompt_tokens_details\": + {\n \"cached_tokens\": 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_6c0d1490cb\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:28:25 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '3510' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: '{"messages":[{"role":"system","content":"You are refining upcoming plan + steps based on new information. Update the step descriptions to be more specific + and actionable given what was learned. Keep the same step numbers.\n\nRespond + with one line per step in the format:\nStep N: "},{"role":"user","content":"## + New information learned\nRecent developments in autonomous AI agents indicate + significant progress across various industries, demonstrating their potential + to enhance productivity and operational efficiency, though challenges in ethical + governance and reliability persist.\n\n## Suggested refinements\nIn Step 2, + emphasize the emerging challenges related to ethical governance and reliability + alongside the key findings.\n\n## Current pending steps\nStep 2: Summarize the + key findings from the research conducted on autonomous AI agents.\n\nUpdate + the step descriptions to incorporate the new information."}],"model":"gpt-4o-mini"}' + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '962' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NfFXJZKoImIRE7AeKGxjf4bybQ8\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413305,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"Step 2: Summarize the key findings + from the research conducted on autonomous AI agents, highlighting their potential + to enhance productivity and operational efficiency while addressing emerging + challenges related to ethical governance and reliability.\",\n \"refusal\": + null,\n \"annotations\": []\n },\n \"logprobs\": null,\n + \ \"finish_reason\": \"stop\"\n }\n ],\n \"usage\": {\n \"prompt_tokens\": + 157,\n \"completion_tokens\": 39,\n \"total_tokens\": 196,\n \"prompt_tokens_details\": + {\n \"cached_tokens\": 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:28:26 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '1102' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: "{\"messages\":[{\"role\":\"system\",\"content\":\"You are Research Analyst. + You are a research analyst who searches the web for information, identifies + key findings, and produces structured research summaries.\\n\\nYour goal: Research + topics using search tools and produce structured summaries\\n\\nYou are executing + a specific step in a multi-step plan. Focus ONLY on completing\\nthe current + step. Do not plan ahead or worry about future steps.\\n\\nBefore acting, briefly + reason about what you need to do and which approach\\nor tool would be most + helpful for this specific step.\"},{\"role\":\"user\",\"content\":\"## Current + Step\\nSummarize the key findings from the research conducted on autonomous + AI agents, highlighting their potential to enhance productivity and operational + efficiency while addressing emerging challenges related to ethical governance + and reliability.\\n\\n## Context from previous steps:\\nStep 1 result: ### Recent + Developments in Autonomous AI Agents (2025)\\n\\nThis summary compiles recent + findings on the developments in autonomous AI agents for the year 2025, highlighting + their evolution, applications, challenges, and implications across various industries.\\n\\n#### + Key Articles and Insights:\\n\\n1. **AI Agents in 2025: Expectations vs. Reality** + \ \\n - **Source**: [IBM](https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality) + \ \\n - **Published**: February 26, 2025 \\n - **Summary**: The article + discusses the transition from large language models (LLMs) to agentic AI, emphasizing + autonomous agents' potential to reshape work dynamics. Experts from IBM note + that while there is enthusiasm about AI agents, many current applications are + still in their infancy, requiring refinement in planning and reasoning capabilities + to reach true autonomy.\\n\\n2. **Autonomous AI Agents Are Reshaping the Business + Landscape** \\n - **Source**: [KPMG](https://kpmg.com/sk/en/insights/2025/06/autonomous-ai-agents-reshape-business-landscape.html) + \ \\n - **Published**: June 25, 2025 \\n - **Summary**: This report outlines + how agentic AI is transitioning from theory into practical use, impacting sectors + like finance and healthcare. It describes agentic AI as a \\\"digital workforce\\\" + that enhances operational efficiency and decision-making processes through autonomous + task management. \\n\\n3. **The Rise of Autonomous AI Agents: Automating Complex + Tasks** \\n - **Source**: [ResearchGate](https://www.researchgate.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks) + \ \\n - **Published**: July 11, 2025 \\n - **Summary**: This academic paper + explores the technological foundations and applications of autonomous agents + in various fields, noting their ability to autonomously resolve multi-step tasks. + It addresses the ethical considerations linked to the increasing autonomy of + these systems.\\n\\n4. **Developments in AI Agents: Q1 2025 Landscape Analysis** + \ \\n - **Source**: [The Science of Machine Learning & AI](https://www.ml-science.com/blog/2025/4/17/developments-in-ai-agents-q1-2025-landscape-analysis) + \ \\n - **Published**: April 17, 2025 \\n - **Summary**: Highlights significant + advancements in AI agent technology, including their capability for autonomous + task achievement. The article underscores ongoing challenges related to reliability, + safety, and ethical governance in AI performance.\\n\\n5. **Agentic AI in Software + Development** \\n - **Source**: [SculptSoft](https://www.sculptsoft.com/agentic-ai-in-action-how-autonomous-ai-agents-are-changing-software-development-in-2025/) + \ \\n - **Published**: October 15, 2025 \\n - **Summary**: Discusses how + autonomous AI agents are revolutionizing software development by automating + the development lifecycle\u2014planning, coding, testing, and deployment\u2014thus + enabling teams to innovate faster while reducing manual workloads.\\n\\n6. **AI + Agents Arrived in 2025 \u2013 Here\u2019s What Happened and the Challenges Ahead** + \ \\n - **Source**: [The Conversation](https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325) + \ \\n - **Published**: December 29, 2025 \\n - **Summary**: The article + reviews the major milestones reached in 2025, including the introduction of + protocols to enable agents to communicate and use external tools. It also addresses + the societal implications, such as the need for enhanced governance and ethical + considerations in AI deployment.\\n\\n7. **2025 Buildings Show: AI Agents Become + 'Talk of the Town'** \\n - **Source**: [ConstructConnect](https://news.constructconnect.com/2025-buildings-show-ai-agents-become-talk-of-the-town) + \ \\n - **Published**: December 11, 2025 \\n - **Summary**: At the Buildings + Show, digital innovation in construction technology showcased agentic AI's capability + to monitor and manage project tasks autonomously, streamlining operations in + the sector.\\n\\n### Conclusion\\nThe advancements in autonomous AI agents in + 2025 exhibit a shift towards integrating AI into real-world applications, enhancing + productivity across various industries. While there is significant potential + for these agents to improve efficiency and decision-making, challenges related + to ethical considerations, governance, and reliability remain pivotal in guiding + the future of AI development.\\n\\nComplete this step and provide your result.\"}],\"model\":\"gpt-4o-mini\",\"tool_choice\":\"auto\",\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"exa_search_tool\",\"description\":\"Search + the internet using Exa\",\"strict\":true,\"parameters\":{\"properties\":{\"search_query\":{\"description\":\"Mandatory + search query you want to use to search the internet\",\"title\":\"Search Query\",\"type\":\"string\"},\"start_published_date\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}],\"default\":null,\"description\":\"Start + date for the search\",\"title\":\"Start Published Date\"},\"end_published_date\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}],\"default\":null,\"description\":\"End + date for the search\",\"title\":\"End Published Date\"},\"include_domains\":{\"anyOf\":[{\"items\":{\"type\":\"string\"},\"type\":\"array\"},{\"type\":\"null\"}],\"default\":null,\"description\":\"List + of domains to include in the search\",\"title\":\"Include Domains\"}},\"required\":[\"search_query\",\"start_published_date\",\"end_published_date\",\"include_domains\"],\"type\":\"object\",\"additionalProperties\":false}}}]}" + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '6327' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NfHqzDAkS5GLBqefBdkiXo6gEh9\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413307,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": null,\n \"tool_calls\": [\n {\n + \ \"id\": \"call_eMVIqV0hW7fOCNVzPGvarvNv\",\n \"type\": + \"function\",\n \"function\": {\n \"name\": \"exa_search_tool\",\n + \ \"arguments\": \"{\\\"search_query\\\": \\\"autonomous AI agents + productivity operational efficiency 2025\\\", \\\"start_published_date\\\": + \\\"2025-01-01\\\", \\\"end_published_date\\\": \\\"2025-12-31\\\", \\\"include_domains\\\": + null}\"\n }\n },\n {\n \"id\": \"call_QoeOUYfW09KlW4WYa0jHps9N\",\n + \ \"type\": \"function\",\n \"function\": {\n \"name\": + \"exa_search_tool\",\n \"arguments\": \"{\\\"search_query\\\": + \\\"autonomous AI agents ethical governance reliability challenges 2025\\\", + \\\"start_published_date\\\": \\\"2025-01-01\\\", \\\"end_published_date\\\": + \\\"2025-12-31\\\", \\\"include_domains\\\": null}\"\n }\n }\n + \ ],\n \"refusal\": null,\n \"annotations\": []\n },\n + \ \"logprobs\": null,\n \"finish_reason\": \"tool_calls\"\n }\n + \ ],\n \"usage\": {\n \"prompt_tokens\": 1262,\n \"completion_tokens\": + 127,\n \"total_tokens\": 1389,\n \"prompt_tokens_details\": {\n \"cached_tokens\": + 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_6c0d1490cb\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:28:29 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '2056' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: '{"query": "autonomous AI agents productivity operational efficiency 2025", + "startPublishedDate": "2025-01-01", "endPublishedDate": "2025-12-31", "type": + "auto", "contents": {"text": {"maxCharacters": 10000}}}' + headers: + Accept: + - '*/*' + Connection: + - keep-alive + Content-Length: + - '208' + Content-Type: + - application/json + User-Agent: + - X-USER-AGENT-XXX + accept-encoding: + - ACCEPT-ENCODING-XXX + x-api-key: + - X-API-KEY-XXX + method: POST + uri: https://api.exa.ai/search + response: + body: + string: "{\"requestId\":\"bf6eb24264e127b65fa61f08dc496dce\",\"resolvedSearchType\":\"neural\",\"results\":[{\"id\":\"https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-agent-survey.html\",\"title\":\"PwC's + AI Agent Survey\",\"url\":\"https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-agent-survey.html\",\"publishedDate\":\"2025-05-16T00:00:00.000Z\",\"author\":\"PricewaterhouseCoopers\",\"text\":\"AI + agent survey: PwC\\n[] \\nSearch\\nMenu\\n[Featured insights] \\n[Featured + insights] \\nMenu\\n[Featured insights] \\n[2026 Global Digital Trust Insights + Survey] \\nMenu\\n[Featured insights] \\n[Board governance issues] \\nMenu\\n[Featured + insights] \\n[C-suite insights] \\nMenu\\n[Featured insights] \\n[Case studies] + \\nMenu\\n[Featured insights] \\n[Policy on Demand] \\nMenu\\n[Featured insights] + \\n[Podcasts] \\nMenu\\n[Featured insights] \\n[PwC Executive Pulse] \\nMenu\\n[Featured + insights] \\n[Tech Effect] \\nMenu\\n[Featured insights] \\n[Viewpoint] \\nMenu\\n[Featured + insights] \\n[Webcasts] \\nMenu\\n[Featured insights] \\n[All Research and + insights] \\nFeatured\\n[![] \\nAmerica in motion\\n] \\n[![] \\nExecutive + leadership hub - What\u2019s important to the C-suite?\\n] \\nMenu\\n[Capabilities] + \\n[Capabilities] \\nMenu\\n[Capabilities] \\n[Audit and Assurance] \\nMenu\\n[Capabilities] + \\n[Alliances and ecosystems] \\nMenu\\n[Capabilities] \\n[Artificial Intelligence + (AI)] \\nMenu\\n[Capabilities] \\n[Board governance issues] \\nMenu\\n[Capabilities] + \\n[Consulting] \\nMenu\\n[Capabilities] \\n[Cybersecurity, Risk and Regulatory] + \\nMenu\\n[Capabilities] \\n[Deals] \\nMenu\\n[Capabilities] \\n[Digital assets + and crypto] \\nMenu\\n[Capabilities] \\n[Digital assurance and transparency] + \\nMenu\\n[Capabilities] \\n[Engineering & AI] \\nMenu\\n[Capabilities] + \\n[Enterprise Strategy] \\nMenu\\n[Capabilities] \\n[Financial Markets & + Real Estate] \\nMenu\\n[Capabilities] \\n[Finance and Accounting] \\nMenu\\n[Capabilities] + \\n[Financial statement audit] \\nMenu\\n[Capabilities] \\n[Front Office] + \\nMenu\\n[Capabilities] \\n[Managed Services] \\nMenu\\n[Capabilities] \\n[Metaverse] + \\nMenu\\n[Capabilities] \\n[Operations] \\nMenu\\n[Capabilities] \\n[PwC + Private] \\nMenu\\n[Capabilities] \\n[Risk Modeling Services] \\nMenu\\n[Capabilities] + \\n[Sustainability and ESG] \\nMenu\\n[Capabilities] \\n[Tax services] \\nMenu\\n[Capabilities] + \\n[Workforce] \\nMenu\\n[Capabilities] \\n[All capabilities] \\nMenu\\n[Industries] + \\n[Industries] \\nMenu\\n[Industries] \\n[Aerospace and defense] \\nMenu\\n[Industries] + \\n[Asset and wealth management] \\nMenu\\n[Industries] \\n[Automotive] \\nMenu\\n[Industries] + \\n[Banking and capital markets] \\nMenu\\n[Industries] \\n[Chemicals] \\nMenu\\n[Industries] + \\n[Consumer markets] \\nMenu\\n[Industries] \\n[Consumer packaged goods] + \\nMenu\\n[Industries] \\n[Energy] \\nMenu\\n[Industries] \\n[Energy and industrials] + \\nMenu\\n[Industries] \\n[Engineering and construction] \\nMenu\\n[Industries] + \\n[Financial services] \\nMenu\\n[Industries] \\n[Gaming] \\nMenu\\n[Industries] + \\n[Health industries] \\nMenu\\n[Industries] \\n[Health services] \\nMenu\\n[Industries] + \\n[Hospitality and leisure] \\nMenu\\n[Industries] \\n[Industrial manufacturing] + \\nMenu\\n[Industries] \\n[Insurance] \\nMenu\\n[Industries] \\n[Media] \\nMenu\\n[Industries] + \\n[Medical technology] \\nMenu\\n[Industries] \\n[Pharmaceutical and life + sciences] \\nMenu\\n[Industries] \\n[Power and utilities] \\nMenu\\n[Industries] + \\n[Private equity] \\nMenu\\n[Industries] \\n[Real estate] \\nMenu\\n[Industries] + \\n[Restaurants, wholesale and agriculture] \\nMenu\\n[Industries] \\n[Retail] + \\nMenu\\n[Industries] \\n[Space] \\nMenu\\n[Industries] \\n[Sports] \\nMenu\\n[Industries] + \\n[Technology] \\nMenu\\n[Industries] \\n[Technology, media and telecommunications] + \\nMenu\\n[Industries] \\n[Telecommunications] \\nMenu\\n[Industries] \\n[Travel, + transportation and logistics] \\nMenu\\n[Technology] \\n[Technology] \\nMenu\\n[Technology] + \\n[Alliances and ecosystems] \\nMenu\\n[Technology] \\n[Delivery platforms] + \\nMenu\\n[Technology] \\n[Emerging technology] \\nMenu\\n[Technology] \\n[Engineering + & AI] \\nMenu\\n[Technology] \\n[Products] \\nMenu\\n[Technology] \\n[Tech-enabled + services] \\nMenu\\n[Technology] \\n[Tech Effect] \\nMenu\\n[About us] \\n[About + us] \\nMenu\\n[About us] \\n[Alumni] \\nMenu\\n[About us] \\n[Analyst relations] + \\nMenu\\n[About us] \\n[Investing in our people] \\nMenu\\n[About us] \\n[Newsroom] + \\nMenu\\n[About us] \\n[Offices] \\nMenu\\n[About us] \\n[Our leadership] + \\nMenu\\n[About us] \\n[Purpose and values] \\nFeatured\\n[![] \\nTech Effect\\n] + \\nMenu\\n[Careers] \\n[Careers] \\nMenu\\n[Careers] \\n[Why PwC] \\nMenu\\n[Careers] + \\n[Entry Level Careers] \\nMenu\\n[Careers] \\n[Experienced Careers] \\nMenu\\n[Careers] + \\n[University Relations] \\nFeatured\\n[![] \\nShared success benefits\\n] + \\nLoading Results\\nNo Match Found\\nView All Results\\n# PwC\u2019s AI Agent + Survey\\nCopy linkLink copied to clipboard\\n![Example pattern for mobile] + \\n![] \\n## Summary\\n* Companies are increasing AI investments and realizing + productivity gains, but few are transforming how work gets done.\\n* The biggest + barrier isn\u2019t the technology; it\u2019s mindset, change readiness and + workforce engagement.\\n* Trust lags for high-stakes use cases, underscoring + the need for responsible AI.\\n* Few businesses are connecting agents across + workflows and functions, yet that\u2019s where the real value lies.\\n### + Key trends about AI agents in the enterprise\\nAI agents aren\u2019t just + the future \u2014they\u2019re already on the job. Today, they\u2019re delivering + real results. US companies are ramping up investments, and many are already + seeing returns. But despite the momentum, most haven\u2019t made the bold, + strategic shifts needed to sustain that success.\\nOf the 300 senior executives + in our May 2025 survey, 88% say their team or business function plans to increase + AI-related budgets in the next 12 months due to agentic AI. Seventy-nine percent + say AI agents are already being adopted in their companies. And of those adopting + AI agents, two-thirds (66%) say that they\u2019re delivering measurable value + through increased productivity.\\nYet from what we\u2019re seeing in the field, + broad adoption doesn\u2019t always mean deep impact. Many employees are using + agentic features built into enterprise apps to speed up routine tasks \u2014surfacing + insights, updating records, answering questions. It\u2019s a meaningful boost + in productivity, but it stops short of transformation.\\n> > While embedded + agents from hyperscalers and model providers are seeing strong uptake, the + real opportunity is still ahead. Reports of full adoption often reflect excitement + about what agentic capabilities could enable \u2014not evidence of widespread + transformation.\\n> > That\u2019s beginning to shift. Multi-agent models are + emerging as a powerful next step \u2014systems of AI agents working in concert + to deliver tangible results. These agents can take on complex, cross-functional + workflows across finance, customer service, software development, R&D + and more.\\nAI isn\u2019t just another technology, as more and more executives + recognize. Three-quarters (75%) agree or strongly agree that AI agents will + reshape the workplace more than the internet did. What\u2019s more, 71% agree + that AI agents are advancing so quickly that artificial general intelligence + (AGI), in which AI can think, learn and solve problems as broadly and flexibly + as a human, will be a reality within two years.\\nWith the potential for disruption + growing, playing it safe isn\u2019t a strategy. To lead with AI, you need + to move with intention. Our survey reveals four areas that demand your attention.\\n## + 1. Budgets are surging\\nAlmost nine out of 10 executives surveyed (88%) say + their companies plan to up their AI-related budgets this year due to agentic + AI. Over a quarter plan increases of 26% or more, likely to pay for ambitious + plans.\\n> > Of those companies adopting AI agents, 35% say they\u2019re doing + so broadly, while another 17% say AI agents are being fully adopted in almost + all workflows and functions.\\n> > Still, most (68%) report that half or fewer + of their employees interact with agents in their everyday work. Few have the + kind of agentic workflows that a[leading hospitality company] we worked with + has already put in place. There, employees and customers engage with teams + of AI agents working across functions \u2014enhancing experiences, improving + speed and driving down costs.\\nThis is a real-world example of multi-agent + models in action \u2014AI agents working together to deliver meaningful outcomes. + While this is already delivering significant value for one company, it signals + what\u2019s next for many others.\\n## 2. Measurable value is here\\nIf your + company is like most, leadership (and other stakeholders) want AI results + now \u2014increased revenue, lower costs or other concrete value \u2014and + AI agents are delivering. Of those adopting AI agents, nearly two-thirds (66%) + report increased productivity. Over half (57%) report cost savings, faster + decision-making (55%) or improved customer experience (54%). As with AI generally, + early value often comes from internal use cases, but customer-facing cases + are rising fast.\\n> > 73% of survey respondents agree that\",\"image\":\"https://www.pwc.com/us/en/tech-effect/content/images/flagship/ai/ai-agents-survey.png\",\"favicon\":\"https://www.pwc.com/etc.clientlibs/pwc/clientlibs/css_common/resources/image/favicon.ico\"},{\"id\":\"https://www.intouchcx.com/thought-leadership/the-rise-of-ai-agents-redefining-automation-and-productivity-in-2025/\",\"title\":\"The + Rise of AI Agents: Redefining Automation and ...\",\"url\":\"https://www.intouchcx.com/thought-leadership/the-rise-of-ai-agents-redefining-automation-and-productivity-in-2025/\",\"publishedDate\":\"2025-04-08T00:00:00.000Z\",\"author\":\"IntouchCX + Team\",\"text\":\"The Rise of AI Agents: Redefining Automation and Productivity + in 2025 - IntouchCX\\nToggle menu[![Intouch logo]] \\n[![Intouch logo]] Toggle + menu\\n[Contact] [Login] Close menu\\n[Join Our Team] \\n[Contact] [Login] + [Join Our Team] \\n[![Facebook icon] Facebook] [![Twitter icon] Twitter] [![Instagram + icon] Linkedin] [![Pinterest icon] Instagram] [![YouTube icon] YouTube] [![TikTok + icon] TikTok] \\n* [Home] \\n* |[Blog Posts] \\n* |The Rise of AI Agents: + Redefining Automation and Productivity in 2025\\n![] \\n# Resources\\n# The + Rise of AI Agents: Redefining Automation and Productivity in 2025\\n* ![Picture + of IntouchCX Team] IntouchCX Team\\n* April 8, 2025\\n![] \\nThe integration + of AI into everyday business operations has reached a turning point. In 2024, + AI models became smarter and more efficient; in 2025, the focus shifts from + intelligence to action. AI agents are no longer just responsive tools\u2014they + are autonomous systems capable of executing complex workflows, making strategic + decisions, and seamlessly integrating into business operations to enhance + productivity and efficiency across industries, reshaping the way businesses + operate and setting new standards for automation and collaboration.\\nFrom + Chatbots to Fully Autonomous AI Agents\\nThe transition from conversational + AI to AI agents is a game-changer. While traditional AI models like chatbots + primarily generate text and answer questions, AI agents go further by taking + independent actions based on contextual understanding. These agents can schedule + meetings, manage workflows, analyze large datasets, and even navigate the + web to perform complex tasks without human intervention.\\n[**The future of + AI isn’t just smarter chatbots\u2014it’s fully autonomous AI agents + that anticipate needs, execute tasks, and continuously learn from interactions.**] + This shift is transforming the way businesses operate, integrating AI agents + into daily workflows and leveraging advancements in AI orchestration, where + multiple agents collaborate in enterprise settings. Instead of relying on + a single AI model, businesses will deploy specialized AI agents fine-tuned + for different roles\u2014whether handling customer service inquiries, managing + supply chains, or optimizing marketing campaigns.\\n**The Business Impact + of AI Agents**\\nThe adoption of AI agents will transform industries by boosting + efficiency and scalability. Companies deploying these autonomous systems expect + significant productivity gains, with early estimates suggesting up to a 30% + increase in operational efficiency.\\n**Key benefits include:**\\n* **Task + Automation:**AI agents will handle repetitive and time-consuming processes, + allowing employees to focus on strategic initiatives.\\n* **Scalability:**Businesses + can expand operations without proportionally increasing costs, as AI agents + streamline workflows.\\n* **Enhanced Multimodal Capabilities:**Beyond text-based + interactions, AI agents will process images, audio, and video to provide richer + insights and solutions.\\n* **AI-Powered Decision-Making:**These agents will + leverage real-time data analysis to make informed recommendations, reducing + the need for human oversight in routine decision-making.\\nThe Emergence of + Multi-Agent Systems\\nAs businesses scale AI adoption, they will transition + from using single AI agents to deploying coordinated teams of specialized + agents. These AI-powered “swarms” will work in tandem, breaking + down complex workflows into manageable steps.\\nFor example, an enterprise + AI ecosystem might consist of:\\n* A scheduling agent that coordinates meetings + based on team availability.\\n* A customer support agent that resolves inquiries + in real time.\\n* A research agent that scans market trends and generates + reports.\\nThis interconnected approach enhances efficiency while ensuring + that AI-driven processes remain dynamic and adaptable to real-world business + needs.\\nAI Agents and Human Oversight\\nDespite their autonomy, AI agents + will not operate in isolation. Human oversight remains crucial for handling + high-stakes decisions, ethical considerations, and complex problem-solving. + The challenge for businesses in 2025 will be striking the right balance between + automation and human intervention, ensuring that AI-driven processes remain + accountable and aligned with organizational goals.\\nThe Future of AI Agents + in Enterprise Workflows\\nLooking ahead, AI agents have the potential to become + an integral part of the workforce. As regulations catch up with rapid technological + advancements, businesses must navigate new frameworks for AI governance, data + security, and ethical AI deployment.\\nWith AI orchestration tools streamlining + operations and new protocols enabling seamless integration with third-party + technologies, AI agents will redefine what\u2019s possible in automation. + From handling routine administrative tasks to driving strategic decision-making, + these digital co-workers are set to revolutionize productivity, efficiency, + and innovation across industries.\\nHowever, AI cannot operate in isolation\u2014human + involvement remains essential to ensure accuracy, ethical decision-making, + and empathetic customer interactions. By automating repetitive processes, + AI frees up agents to focus on complex problem-solving and emotionally intelligent + engagements, enhancing both efficiency and customer satisfaction.\\nAI agents + are not just the next step in automation\u2014they are the future of work. + Find out how you can leverage[**AI solutions**] in your CX strategy now!\\n## + Recent Posts\\n[![Transforming Knowledge Management Through Strategic Sprints] + February 4, 2026.### Transforming Knowledge Management Through Strategic Sprints\\nCustomer + experience often comes down to one simple factor: how easily frontline agents + can find […]\\nRead more] \\n[![AI, Automation, and Customer-Centric + Travel: The Big Trends for the Transportation Industry] January 28, 2026.### + AI, Automation, and Customer-Centric Travel: The Big Trends for the Transportation + Industry\\nThe Future in Motion: How AI and Automation Are Shaping Travel + and Transportation \_Executive […]\\nRead more] \\n[![From Friction + to Flow: Solving Common CX Pain Points With Trust & Safety] January 21, + 2026.### From Friction to Flow: Solving Common CX Pain Points With Trust & + Safety\\nA flawless customer journey can be ruined in seconds if a user encounters + fraud, harassment, […]\\nRead more] \\n![graphic]![graphic]![graphic]![graphic]![graphic]\",\"image\":\"https://www.intouchcx.com/wp-content/uploads/2025/04/AdobeStock_1309149643_1024x580_V3-1.jpg\",\"favicon\":\"https://www.intouchcx.com/wp-content/themes/intouchcx/assets/favicon/favicon-32x32.png\"},{\"id\":\"https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality\",\"title\":\"AI + Agents in 2025: Expectations vs. Reality\",\"url\":\"https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality\",\"publishedDate\":\"2025-02-26T00:00:00.000Z\",\"author\":null,\"text\":\"AI + Agents in 2025: Expectations vs. Reality | IBM\\nTags\\n[Artificial Intelligence] + \\n# AI agents in 2025: Expectations vs. reality\\n![the same person doing + multiple jobs] \\n## Authors\\n[Ivan Belcic] \\nStaff writer\\n[Cole Stryker] + \\nStaff Editor, AI Models\\nIBM Think\\n## AI agents in 2025: Expectations + vs. reality\\nIt\u2019s impossible to take two steps across the tech media + landscape without stumbling over an article hailing 2025 as the year of the[AI + agent]. Agents, we\u2019re told, will transform the way work is done, impacting + every facet of our lives, personal and professional.\\nWe\u2019d barely surfaced + from a landslide of NFT and crypto hype that characterized the early 2020s, + and the metaverse bubble that followed, before media voices began singing + the praises of[generative AI] (gen AI) in the wake of releases such as OpenAI\u2019s[GPT] + model family, Anthropic\u2019s[Claude] and Microsoft\u2019s Copilot.\\nWhile + the chorus hasn\u2019t moved on entirely, the focus in 2025 has shifted from[large + language models (LLMs)] to advancements in the ostensibly autonomous[artificial + intelligence (AI)] agents ushering in the future of work.\\nDespite a momentary + surge in gen AI interest around[Deepseek] \u2019s R1, which promised significant + performance improvements over ChatGPT, the dominant innovation narrative in + 2025 is the AI agent.\\nMedia coverage highlights the promises of innovation,[automation] + and efficiency agents will bring, but how much of the conversation is click-hungry + hype?\\nThe ad-supported media world thrives on clicks, and it\u2019s reasonable + to expect sensational, attention-grabbing headlines crafted to garner yours. + But what can we realistically expect from[agentic AI] in 2025, and how will + it affect our lives?\\nWe spoke with several IBM experts to cut through the + hype, with the goal of holding a more reasonable conversation about AI agents + and what they\u2019re going to do. Our team of informed insiders includes:\\n* + [Maryam Ashoori, PhD]: Director of Product Management,[IBM\xAE watsonx.ai\u2122] + \\n* [Marina Danilevsky]: Senior Research Scientist, Language Technologies\\n* + [Vyoma Gajjar]: AI Technical Solutions Architect\\n* [Chris Hay]: Distinguished + Engineer\\nIndustry newsletter\\n### The latest AI trends, brought to you + by experts\\nGet curated insights on the most important\u2014and intriguing\u2014AI + news. Subscribe to our weekly Think newsletter. See the[IBM Privacy Statement].\\n### + Thank you! You are subscribed.\\nYour subscription will be delivered in English. + You will find an unsubscribe link in every newsletter. You can manage your + subscriptions or unsubscribe[here]. Refer to our[IBM Privacy Statement] for + more information.\\n## What are AI agents?\\nAn AI agent is a software program + capable of acting autonomously to understand, plan and execute tasks. AI agents + are powered by LLMs and can interface with tools, other models and other aspects + of a system or network as needed to fulfill user goals.\\nWe\u2019re going + beyond asking a chatbot to suggest a dinner recipe based on the available + ingredients in the fridge. Agents are more than automated[customer experience] + emails that inform you it\u2019ll be a few days until a real-world human can + get to your inquiry.\\n[AI agents differ from traditional AI assistants] that + need a prompt each time they generate a response. In theory, a user gives + an agent a high-level task, and the agent figures out how to complete it.\\nCurrent + offerings are still in the early stages of approaching this idea. \u201CWhat\u2019s + commonly referred to as \u2018agents\u2019 in the market is the addition of + rudimentary planning and tool-calling (sometimes called function calling) + capabilities to LLMs,\u201D says Ashoori. \u201CThese enable the LLM to break + down complex tasks into smaller steps that the LLM can perform.\u201D\\nHay + is optimistic that more robust agents are on the way: \u201CYou wouldn\u2019t + need any further progression in models today to build[future AI agents],\u201D + he says.\\nWith that out of the way, what\u2019s the conversation about agents + over the coming year, and how much of it can we take seriously?\\nAI agents\\n### + What are AI agents?\\nFrom monolithic models to compound AI systems, discover + how AI agents integrate with databases and external tools to enhance problem-solving + capabilities and adaptability.\\n[\\nLearn more\\n] \\n## Narrative 1: 2025 + is the year of the AI agent\\n\u201CMore and better agents\u201D are on the + way, predicts Time.[1] \u201CAutonomous \u2018agents\u2019 and profitability + are likely to dominate the artificial intelligence agenda,\u201D reports Reuters.[2] + \u201CThe age of agentic AI has arrived,\u201D promises Forbes, in response + to a claim from Nvidia\u2019s Jensen Huang.[3] \\nTech media is awash with + assurances that our lives are on the verge of a total transformation. Autonomous + agents are poised to streamline and alter our jobs, drive optimization and + accompany us in our daily lives, handling our mundanities in real time and + freeing us up for creative pursuits and other higher-level tasks.\\n### 2025 + as the year of agentic exploration\\n\u201CIBM and Morning Consult did a survey + of 1,000 developers who are building AI applications for enterprise, and 99% + of them said they are exploring or developing AI agents,\u201D\_explains Ashoori. + \u201CSo yes, the answer is that 2025 is going to be the year of the agent.\u201D + However, that declaration is not without nuance.\\nAfter establishing the + current market conception of agents as LLMs with function calling, Ashoori + draws a distinction between that idea and truly autonomous agents. \u201CThe + true definition [of an AI agent] is an intelligent entity with reasoning and + planning capabilities that can autonomously take action. Those reasoning and + planning capabilities are up for discussion. It depends on how you define + that.\u201D\\n\u201CI definitely see AI agents heading in this direction, + but we\u2019re not fully there yet,\u201D says Gajjar. \u201CRight now, we\u2019re + seeing early glimpses\u2014AI agents can already analyze data, predict trends + and automate workflows to some extent. But building AI agents that can autonomously + handle complex decision-making will take more than just better algorithms. + We\u2019ll need big leaps in contextual reasoning and testing for edge cases,\u201D + she adds.\\nDanilevsky isn\u2019t convinced that this is anything new. \u201CI'm + still struggling to truly believe that this is all that different from just + orchestration,\u201D she says. \u201CYou've renamed orchestration, but + now it's called agents, because that's the cool word. But orchestration + is something that we've been doing in programming forever.\u201D\\nWith + regard to 2025 being the year of the agent, Danilevsky is skeptical. \u201CIt + depends on what you say an agent is, what you think an agent is going to accomplish + and what kind of value you think it will bring,\u201D she says. \u201CIt's + quite a statement to make when we haven't even yet figured out ROI (return + on investment) on LLM technology more generally.\u201D\\nAnd it\u2019s not + just the business side that has her hedging her bets. \u201CThere's the + hype of imagining if this thing could think for you and make all these decisions + and take actions on your computer. Realistically, that's terrifying.\u201D\\nDanilevsky + frames the disconnect as one of miscommunication. \u201C[Agents] tend to be + very ineffective because humans are very bad communicators. We still can't + get chat agents to interpret what you want correctly all the time.\u201D\\nStill, + the forthcoming year holds a lot of promise as an era of experimentation. + \u201CI'm a big believer in [2025 as the year of the agent],\u201D says + Hay excitedly.\\nEvery large tech company and hundreds of startups are now + experimenting with agents. Salesforce, for example, has released their Agentforce + platform, which enables users to create agents that are easily integrated + within the Salesforce app ecosystem.\\n\u201CThe wave is coming and we're + going to have a lot of agents. It's still a very nascent ecosystem, so + I think a lot of people are going to build agents, and they're going to + have a lot of fun.\u201D\\n## Narrative 2: Agents can handle highly complex + tasks on their own\\nThis narrative assumes that today\u2019s agents meet + the theoretical definition outlined in the introduction to this piece. 2025\u2019s + agents will be fully autonomous AI programs that can scope out a project and + complete it with all the necessary tools they need and with no help from human + partners. But what\u2019s missing from this narrative is nuance.\\n### Today\u2019s + models are more than enough\\nHay believes that the groundwork has already + been laid for such developments. \u201CThe big thing about agents is that + they have the ability to plan,\u201D he outlines. \u201CThey have the ability + to reason, to use tools and perform tasks, and they need to do it at speed + and scale.\u201D\\nHe cites 4 developments that, compared to the best models + of 12 to 18 months ago, mean that the models of early 2025 can power the agents + envisioned by the proponents of this narrative:\\n* Better, faster, smaller + models\\n* Chain-of-thought (COT) training\\n* Increased context windows\\n* + Function calling\\n\u201CNow, most of these things are in play,\u201D Hay + continues. \u201CYou can have the AI call tools. It can plan. It can reason + and come back with good answers. It can use inference-time compute. You\u2019ll + have better chains of thought and more memory to work with. It's going + to run fast. It\u2019s going to be cheap. That leads you to a structure where + I think you can have agents. The models are improving and they're getting + better, so that's only going to accelerate.\u201D\\n### Realistic expectations + are a must\\nAshoori is careful to differentiate between what agents will + be able to do later, and what they can do now. \u201CThere is the promise, + and there is what the agent's capable of doing today,\u201D she says. + \u201CI would say the answer depends on the use case. For simple use cases, + the agents are capable of [choosing the correct tool], but for more sophisticated + use cases, the technology\",\"image\":\"https://www.ibm.com/content/dam/connectedassets-adobe-cms/worldwide-content/gbs/cf/ul/g/5a/88/336.jpg/_jcr_content/renditions/cq5dam.thumbnail.1280.1280.png\",\"favicon\":\"https://www.ibm.com/content/dam/adobe-cms/default-images/icon-32x32.png\"},{\"id\":\"https://www.druidai.com/blog/top-tried-and-tested-use-cases-for-autonomous-ai-agents-in-2025\",\"title\":\"Top + Tried and Tested Use Cases for Autonomous AI ...\",\"url\":\"https://www.druidai.com/blog/top-tried-and-tested-use-cases-for-autonomous-ai-agents-in-2025\",\"publishedDate\":\"2025-05-23T00:00:00.000Z\",\"author\":null,\"text\":\"Top + Tried and Tested Use Cases for Autonomous AI Agents in 2025\\n[Skip to content] + \\n[Blog] \\nai agents\\nMay 23, 2025\\n6 MIN READ\\n# Top Tried and Tested + Use Cases for Autonomous AI Agents in 2025\\n![Andreea Radulescu] Andreea + Radulescu\\n![graphic-of-hashtag-true-story-about-ai-agent-use-cases] \\nImagine + your calendar reminding you of a meeting, then booking the room, sending the + brief, gathering the right people, and following up afterward with action + items. Ok, your calendar cannot do all of that; that\u2019s what an autonomous + AI agent brings to the table. It doesn\u2019t just notify you. It thinks, + acts, and makes things happen in the background so you don\u2019t have to.\\nThese + aren\u2019t glorified chatbots. They\u2019re your new smart digital coworkers, + always on, never confused, and constantly improving with every task they complete. + Whether it\u2019s streamlining finance, simplifying HR, supporting students, + or managing supply chains, AI agents are already embedded in the workflows + that matter.\\nLet\u2019s explore the real-world use cases showing why these + agents are the backbone of affordable, scalable automation, and why they\u2019re + becoming a must-have in every enterprise playbook.\\n## Why 2025 Wasthe Year + of the Autonomous AI Agent\\nAccording to recent insights from top research + and tech publications, the adoption of agentic AI has surged as organizations + seek smarter ways to automate with accountability. Autonomous AI agents, when + designed with orchestration, security, and lifecycle management in mind, become + more than tools; they become proactive team members who get things done.\\nAnd + the ROI? It\u2019s compelling. Studies point to 30\u201380% improvements in + speed, accuracy, and cost savings across industries when deploying agentic + AI workflows. Let's dig into what these agents are actually doing in the real + world.\\n## Most In-Demand Use Cases for Autonomous AI Agents in 2025\\nAs + more organizations shift from static automation to intelligent systems that + adapt, decide, and execute, autonomous AI agents are stepping into roles once + handled by multiple tools, and sometimes, entire teams. These agents aren\u2019t + just supporting workflows; they\u2019re running them from start to finish. + What was once limited to isolated tasks is now evolving into coordinated, + cross-functional processes.\\nBelow are the most in-demand, high-impact use + cases, showcasing where autonomous AI agents deliveredreal business value + in 2025.\\n### Finance & Accounting Automation\\n* Accounts Payable & + Receivable: Automatically process invoices, perform PO, matching, approve + payments, and reconcile accounts with[90%+ accuracy and 70% lower costs].\\n* + Expense Management: Track, validate, and report on expenses, cutting approval + cycle time in half.\\n* Tax Compliance: Assist tax teams with scenario modeling, + rule updates, and automated filings while reducing liability and risk.### + HR & Workforce Operations\\n* Employee Onboarding & Offboarding: From + generating contracts and collecting e-signatures to system provisioning and + training reminders,[AI agents automate the full lifecycle].\\n* HR Admin Support: + Field payroll queries, issue certificates, update records, and enable 24/7 + employee self-service.\\n* Recruitment & Interview Scheduling: Pre-screen + candidates, assess resumes, and coordinate interview logistics, all autonomously.### + Healthcare Support Systems\\n* Patient Scheduling & Triage:[Book appointments], + assess symptoms, and escalate cases automatically, cutting call volume and + reducing delays.\\n* Insurance & Billing Queries: Handle routine billing + inquiries with full access to patient policies and history.\\n* Prescription + Refills: Accept and process refill requests while verifying eligibility and + status.### Banking & Financial Services\\n* Loan Application Processing: + Guide applicants through steps, verify data, and make eligibility checks, + all without human input.\\n* Fraud Detection Agents: Monitor transactions + in real time and flag anomalies using ML and historical fraud data.\\n* Investment + Advisory Assistants: Generate personalized investment insights based on customer + profiles and market data.### Higher Education\\n* Student Engagement Assistants: + Manage inquiries across admissions, financial aid, housing, and course enrollment, + supporting thousands of interactions per day.\\n* Admissions & Application + Agents: Answer applicant questions, collect documents, and update application + status.\\n* IT Helpdesk for Campus Services:[Provide 24/7 support] for login + issues, software access, and account setup.### Manufacturing & Supply + Chain\\n* Predictive Maintenance: Monitor equipment health and schedule repairs + before breakdowns happen.\\n* Procurement & Contract Management: Auto-generate + purchase orders, match contracts, track deliveries, and validate invoices.\\n* + Inventory Optimization Agents: Analyze stock levels, demand trends, and lead + times to trigger replenishments and reduce holding.## The Real Value of Smart + Digital Coworkers\\nWhile many businesses start exploring[autonomous AI agents + for efficiency], the real payoff goes far beyond shaving minutes off a task. + When implemented with orchestration, secure integrations, and lifecycle visibility, + these agents transform how work gets done, and who does it.\\nHere\u2019s + what they really bring to the table:\\n### Always-On Productivity with Autonomous + AI Agents\\nOne of the most immediate benefits of autonomous AI agents is + their 24/7 availability. Unlike human teams, they don\u2019t sleep, get overloaded, + or miss an email. These smart digital coworkers keep processes moving around + the clock, whether it\u2019s answering internal HR questions at 2 AM or processing + invoices over the weekend. This uninterrupted productivity reduces lag, accelerates + workflows, and allows your teams to stay focused on higher-value work.\\n### + Contextual Decision-Making Using Gen AI and Agentic Execution\\nSmart automation + isn\u2019t about doing things faster, it\u2019s about doing them better.[Autonomous + AI agents equipped with generative AI] can understand the nuance behind a + request, while agentic workflows give them the power to act on that insight. + This combination allows agents to make decisions based on user history, policy + data, and workflow context, resulting in responses and actions that are not + just accurate, but relevant and timely.\\n### End-to-End Automation Through + Integrated AI Agent Workflows\\nModern enterprises don\u2019t need point solutions, + they need systems that can handle tasks from start to finish. AI agents shine + in end-to-end automation, managing everything from initiation to validation, + updates, and closure. Whether it\u2019s onboarding a new employee, processing + a refund, or handling a service request, these agents coordinate actions across + systems to ensure no task is dropped mid-stream.\\n### Real ROI with Scalable, + Low-Overhead AI Agents\\nIt\u2019s not just about speed or intelligence. It\u2019s + about impact. Autonomous AI agents offer a clear return on investment by reducing + reliance on manual labor, minimizing errors, and increasing process consistency. + With fewer tools needed to complete a workflow, less strain on human teams, + and faster time-to-value, these agents prove their worth quickly. And because + they\u2019re modular, they scale with your business, without spiraling complexity + or cost.\\n### Smarter Over Time: Learning and Adapting as They Work\\nThe + value of smart digital coworkers compounds over time. These agents don\u2019t + just perform, they learn. With every task, they gather more data, improve + accuracy, and adapt to new business logic or user behavior. As your business + evolves, your agents evolve with it, offering a level of continuity and improvement + that static workflows simply can\u2019t match.\\n## Complete Industry Use-Case + List for Autonomous AI Agents\\n### Banking & Financial Services\\n* Loan + application intake and eligibility checks\\n* KYC (Know Your Customer) data + collection and validation\\n* Real-time fraud detection and alerting\\n* Personalized + customer financial advice\\n* Payment dispute handling and resolution\\n* + Internal compliance monitoring and audit preparation\\n* Secure account access + support and FAQs### Finance & Accounting\\n* Invoice processing and purchase + order matching\\n* Expense report validation and routing\\n* Financial data + reconciliation\\n* Tax calculation, policy lookup, and filing support\\n* + Vendor payment approvals and reminders\\n* Budget variance analysis and reporting### + Human Resources\\n* Employee onboarding/offboarding automation\\n* Leave and + time-off requests\\n* Benefits eligibility and enrollment support\\n* Payroll + inquiries and payslip distribution\\n* HR policy Q&A and compliance notifications\\n* + Pre-screening and recruitment workflow automation\\n* Internal job referrals + and career development guidance### Healthcare\\n* Appointment scheduling and + reminders\\n* Digital symptom checker and triage\\n* Insurance verification + and billing support\\n* Patient intake form collection and follow-up\\n* Provider + directory search and recommendations\\n* Claims processing and status updates\\n* + Post-care follow-up and satisfaction surveys### Higher Education\\n* Admissions + inquiries and application assistance\\n* Course registration and deadline + reminders\\n* Financial aid Q&A and eligibility guidance\\n* IT helpdesk + support for students and faculty\\n* Virtual onboarding for new students or + staff\\n* Scheduling academic advising appointments\\n* Campus facilities + requests and tracking### Retail & eCommerce\\n* Order status tracking + and updates\\n* Product availability and recommendations\\n* Customer complaint + resolution\\n* Store locator and in-store service scheduling\\n* Returns and + refund initiation\\n* Post-purchase survey and feedback capture\\n* Loyalty + program support and personalization### Manufacturing & Supply Chain\\n* + Predictive maintenance scheduling\\n* Inventory management and reordering\\n* + Supplier onboarding and performance tracking\\n* Shipment tracking and logistics + optimization\\n* Procurement approvals\",\"image\":\"https://www.druidai.com/hubfs/graphic-of-hashtag-true-story-about-ai-agent-use-cases.jpg\",\"favicon\":\"https://www.druidai.com/hubfs/raw_assets/public/stoica-Druid/images/favicon.png\"},{\"id\":\"https://www.glean.com/perspectives/how-autonomous-ai-agents-enhance-campaign-planning-in-2025\",\"title\":\"How + autonomous AI agents enhance campaign planning ...\",\"url\":\"https://www.glean.com/perspectives/how-autonomous-ai-agents-enhance-campaign-planning-in-2025\",\"publishedDate\":\"2025-12-17T00:00:00.000Z\",\"author\":\"The + Glean Team\",\"text\":\"How autonomous AI agents enhance campaign planning + in 2025\\n[![Glean Logomark Blue]] \\nProduct\\nWORK AI\_PLATFORM\\n[\\nPlatform + Overview\\n![Platform Overview BG]] \\n[\\nGlean Assistant\\nYour personal + AI assistant\\n] [All Knowledge] [Data Analysis] [Deep Research] [Canvas] + [Companion] [Prompt Library] \\n[\\nGlean Agents\\nBuild and manage AI agents\\n] + [Agent Builder] [Agent Governance] [Agent Orchestration] [Agent Library] \\n[\\nGlean + Search\\nThe foundation of enterprise AI\\n] [Enterprise Graph] [Personal + Graph] [System of context] [Hybrid Search] \\n[\\nConnectors & Actions\\nConnect + to all your apps\\n] [\\nModel Hub\\nGet access to the latest models\\n] [\\nAPIs\\nBuild + generative AI experiences\\n] [\\nGlean Protect\\nSafely scale AI at work\\n] + [\\nSecurity\\nSecure by design from day one\\n] [\\nAgentic Engine\\nPlan + & adapt over company context\\n] \\nGLEAN WHERE\_YOU\_WORK\\n[![] \\nGlean + in Slack\\n] [![] \\nGlean in Microsoft Teams\\n] [![] \\nGlean in Zoom\\n] + [![] \\nGlean in Service Cloud\\n] [![] \\nGlean in ServiceNow\\n] [![] \\nGlean + in Zendesk\\n] [![] \\nGlean in GitHub\\n] [![] \\nGlean in Miro\\n] [![] + \\nBrowser Extension\\n] \\n[Sign in] \\n[\\nCustomers\\n] \\nSolutions\\nDEPARTMENTS\\n[\\nAll + Teams\\n] [\\nEngineering\\n] [\\nCustomer Service\\n] [\\nSales\\n] [\\nMarketing\\n] + [\\nB2B Marketing\\n] [\\nB2C Marketing\\n] [\\nPeople\\n] [\\nIT\\n] \\nINDUSTRIES\\n[\\nRetail\\n] + [\\nFinancial Services\\n] [\\nHigher Education\\n] [\\nHealthcare\\n] [\\nGovernment\\n] + \\n[Sign in] \\n[\\n![] \\nJoel McKelvey\\nHead of Solutions, Glean\\n![] + \\nAbdullah Haydar\\nDirector of Engineering, LinkedIn\\nWebinar\\nAI\_Powered + Engineering\\nExpert insights and actionable strategies for accelerating developer + productivity.\\nWatch now\\n] \\nResources\\nEXPLORE\\n[Blog] [Prompt Library] + [Guides] [Product Videos] \\nENGAGE\\n[Webinars] [Newsroom] [Glean:GO] [Events] + [Gleaniverse Community] \\nSUPPORT & SERVICES\\n[Help center] [Developers] + [Partners] \\n[\\nWork AI Institute\\n![]![]] \\n[Sign in] \\n[![] \\nThe + AI Transformation 100\\nExplore 100 real-world moves organizations are making + to transform themselves with AI.\\nDownload the report\\n] \\n[\\nAbout\\n] + \\nThank you! Your submission has been received!\\nOops! Something went wrong + while submitting the form.\\n[Sign in] \\n[Sign in] \\n[\\nGet a demo\\n] + \\n[\\nGet a demo\\n] \\n[![Glean Logomark Blue]] \\n[Sign in] \\n[Get a demo] + \\n[Get a demo] \\nProduct\\n[\\nCustomers\\n] \\nSolutions\\nResources\\n[\\nAbout\\n] + \\n[Sign in] \\nBack\\nWORK AI\_PLATFORM\\n[\\nPlatform Overview\\n![Platform + Overview BG]] \\n[\\nGlean Assistant\\nYour personal AI assistant\\n] [All + Knowledge] [Data Analysis] [Deep Research] [Canvas] [Companion] [Prompt Library] + \\n[\\nGlean Agents\\nBuild and manage AI agents\\n] [Agent Builder] [Agent + Governance] [Agent Orchestration] [Agent Library] \\n[\\nGlean Search\\nThe + foundation of enterprise AI\\n] [Enterprise Graph] [Personal Graph] [System + of context] [Hybrid Search] \\n[\\nConnectors & Actions\\nConnect to all + your apps\\n] [\\nModel Hub\\nGet access to the latest models\\n] [\\nAPIs\\nBuild + generative AI experiences\\n] [\\nGlean Protect\\nSafely scale AI at work\\n] + [\\nSecurity\\nSecure by design from day one\\n] [\\nAgentic Engine\\nPlan + & adapt over company context\\n] \\nGLEAN WHERE\_YOU\_WORK\\n[![] \\nGlean + in Slack\\n] [![] \\nGlean in Microsoft Teams\\n] [![] \\nGlean in Zoom\\n] + [![] \\nGlean in Service Cloud\\n] [![] \\nGlean in ServiceNow\\n] [![] \\nGlean + in Zendesk\\n] [![] \\nGlean in GitHub\\n] [![] \\nGlean in Miro\\n] [![] + \\nBrowser Extension\\n] \\n[Sign in] \\nDEPARTMENTS\\n[\\nAll Teams\\n] [\\nEngineering\\n] + [\\nCustomer Service\\n] [\\nSales\\n] [\\nMarketing\\n] [\\nB2B Marketing\\n] + [\\nB2C Marketing\\n] [\\nPeople\\n] [\\nIT\\n] \\nINDUSTRIES\\n[\\nRetail\\n] + [\\nFinancial Services\\n] [\\nHigher Education\\n] [\\nHealthcare\\n] [\\nGovernment\\n] + \\n[Sign in] \\n[\\n![] \\nJoel McKelvey\\nHead of Solutions, Glean\\n![] + \\nAbdullah Haydar\\nDirector of Engineering, LinkedIn\\nWebinar\\nAI\_Powered + Engineering\\nExpert insights and actionable strategies for accelerating developer + productivity.\\nWatch now\\n] \\nEXPLORE\\n[Blog] [Prompt Library] [Guides] + [Product Videos] \\nENGAGE\\n[Webinars] [Newsroom] [Glean:GO] [Events] [Gleaniverse + Community] \\nSUPPORT & SERVICES\\n[Help center] [Developers] [Partners] + \\n[\\nWork AI Institute\\n![]![]] \\n[Sign in] \\n[![] \\nThe AI Transformation + 100\\nExplore 100 real-world moves organizations are making to transform themselves + with AI.\\nDownload the report\\n] \\nLast updated Dec 17, 2025.\\n# How autonomous + AI agents enhance campaign planning in 2025\\n0\\nminutes read\\n![How autonomous + AI agents enhance campaign planning in 2025] \\n### Table of contents\\n[\\nHeading + 2\\n] \\n[\\nHeading 3\\n] \\n[\\nHeading 4\\n] \\n[\\nHeading 5\\n] \\n[\\nHeading + 6\\n] \\n[\\nHave questions or want a demo?\\nWe\u2019re here to help! Click + the button below and we\u2019ll be in touch.\\nGet a Demo\\n] \\nShare this + article:\\n[\\n] [\\n] \\n# How autonomous AI agents enhance campaign planning + in 2025\\nMarketing teams face unprecedented complexity as campaigns span + multiple channels, require real-time optimization, and demand[personalized + experiences] at scale. Traditional automation tools execute predefined tasks + but lack the intelligence to adapt strategies when market conditions shift + or customer behaviors change unexpectedly.\\nA new generation of AI technology + promises to transform how organizations plan campaigns and collaborate with + clients. These autonomous systems analyze vast amounts of data, make strategic + decisions, and execute[complex workflows] without constant human oversight + \u2014fundamentally changing the relationship between marketers and their + technology stack.\\nThe shift from reactive automation to proactive intelligence + represents more than incremental improvement. The global AI agent market was + estimated at[$5.40 billion] in 2024 and is forecast to reach $50.31 billion + by 2030, representing explosive growth driven by expanding enterprise adoption. + Organizations deploying these advanced AI capabilities report efficiency gains + of 30-50%, dramatic reductions in campaign optimization cycles, and enhanced + client satisfaction through real-time transparency and strategic insights.\\nAutonomous + AI agents are intelligent systems designed to work independently, performing + tasks such as campaign planning and client collaboration without constant + human oversight. They leverage machine learning to adapt and execute complex + strategies, enhancing efficiency and creativity in marketing initiatives. + These agents integrate with external tools, retrieve relevant information, + and adapt to dynamic environments through sophisticated reasoning architectures. + Companies implementing marketing automation, including autonomous agent components, + achieve average returns of $5.44 for every $1 invested, translating to[544%] + return on investment over three years.\\n## What are autonomous AI agents?\\nAutonomous + AI agents are intelligent systems designed to work independently, performing + tasks such as campaign planning and client collaboration without constant + human oversight. They leverage machine learning to adapt and execute complex + strategies, enhancing efficiency and creativity in marketing initiatives. + These agents integrate with external tools, retrieve relevant information, + and adapt to dynamic environments through sophisticated reasoning architectures. + Among companies currently deploying AI agents,[66%] report increased productivity + while over half report cost savings (57%), faster decision-making (55%), and + improved customer experience (54%).\\nAutonomous AI agents are intelligent + systems designed to work independently, performing tasks such as campaign + planning and client collaboration without constant human oversight. They leverage + machine learning to adapt and execute complex strategies, enhancing efficiency + and creativity in marketing initiatives. These agents integrate with external + tools, retrieve relevant information, and adapt to dynamic environments through + sophisticated reasoning architectures. Notably, workers using generative AI + every day reported that[33.5%] saved four hours or more per week, compared + with only 11.5% of those using it only one day per week, showing a direct + correlation between AI usage intensity and productivity gains.\\nUnlike traditional + marketing automation that follows rigid "if-then" logic, autonomous + agents process multiple signals simultaneously and adjust their approach based + on context. They observe their environment, make decisions, and take actions + to achieve specific marketing objectives \u2014whether that means optimizing + ad spend,\",\"image\":\"https://cdn.prod.website-files.com/6127a84dfe068e153ef20572/66ded1bc82df72f2e1d56eb7_Glean%20Logomark%20Blue.svg\",\"favicon\":\"https://cdn.prod.website-files.com/6127a84dfe068e153ef20572/67189bf127c626679a308b22_32x32%20Glean%20Favicon.png\"},{\"id\":\"https://kodexolabs.com/what-are-autonomous-ai-agents/\",\"title\":\"What + are Autonomous AI Agents? A Complete Guide 2025\",\"url\":\"https://kodexolabs.com/what-are-autonomous-ai-agents/\",\"publishedDate\":\"2025-07-31T00:00:00.000Z\",\"author\":null,\"text\":\"What + are Autonomous AI Agents? A Complete Guide 2025[Skip to content] \\n[![]] + \\n[About us] \\n[What We Do] \\n![]![] [Get A Free AI Chatbot] \\n### Generative + AI\\n* [Gen AI Development] \\n* [Gen AI Integration] \\n* [ChatGPT Dev & + Integration] \\n* [Gen AI Model Development] \\n* [Gen AI Consulting] ### + Product Designing\\n* [Product Designing] \\n### AI Development\\n* [AI Development] + \\n* [AI Chatbot Development] \\n* [AI Consulting] \\n* [AI Model Development] + \\n* [Custom AI Solutions] ### ML Development\\n* [ML Development] \\n* [ML + Consulting] \\n* [ML Model Engineering] \\n* [MLOps Implementation] \\n### + Software Development\\n* [Software Development Services] \\n* [Custom Product + Development] \\n* [Software Consulting] \\n* [Mobile App Development] \\n* + [Web App Development] ### Data Engineering\\n* [Data Engineering] \\n* [Data + Analytics] \\n* [Data Annotation] \\n[Who We Serve] \\n![]![] [Get A Free + AI Chatbot] \\n[### HealthCare\\n] EHR Systems, AI based Interviews and Medical + Imaging Software[### EdTech\\n] Personalized Learning, AI based Tutor Systems + and Gamification Experiences[### Fintech\\n] AI powered Trend Forecasting + and Predicative Analytics\\n[### Energy\\n] Smart Grid Solutions and AI based + Resource Monitoring[### Automotive\\n] Predictive Maintenance, Driver Assistance + and AI Chatbots[### Real Estate\\n] AI Home Management and AI based Real Estate + Evaluation Systems\\n[### IT and Tech\\n] AI powered Ticket Generation and + Automated Software Production[### Marketing\\n] Customer Churn Prediction, + Customer Segmentation and AI based Analytics\\n[Hire Dev] \\n![]![] [Get A + Free AI Chatbot] \\n[### IT Staff Augmentation\\n] On-demand Talent, Scalable + Teams, Flexible Hiring[### Hire Software Developer\\n] Custom Software, Full-stack, + Agile Development[### Software Development Outsourcing\\n] End-to-End, Project-based, + Flexible Engagement\\n[### Hire AI Developer\\n] AI Solutions, Machine Learning, + Custom Models[### Hire Offshore Developer\\n] Remote Teams, Cost-efficient, + Dedicated Experts\\n[### Hire Data Engineer\\n] Data Pipelines, ETL, Big Data + Solutions[### Dedicated Development Team\\n] Tailored Solutions, Seamless + Collaboration, Scalability\\n[Our Work] \\n[Solutions] \\n![]![] [Get A Free + AI Chatbot] \\n### Custom Enterprise Solutions\\n* [Enterprise Resource Planning + (ERP)] \\n* [Human Resource Management Solutions] \\n* [Asset Management Software + Solutions] \\n* [Supply Chain Management Solutions] \\n* [Business Process + Automation Software] \\n* [Fleet Management Software] \\n### Healthcare Software + Solutions\\n* [AI-Powered Medical Imaging & Diagnostics] \\n* [Custom Medical + Practice Management Software] \\n[Company] \\n![]![] [Get A Free AI Chatbot] + \\n[### Careers\\n] Advance your career in AI and software[### Blogs\\n] Official + Blogs for News, Tech & Culture\\n[### Awards & Achievements\\n] Honored for + excellence in AI innovations\\n[Contact Us] \\n[![]] \\n[] \\n# What Are Autonomous + AI Agents? A Complete Guide for 2025 and Beyond\\nSyed Ali Hasan Shah\\n[Agentic + AI] \\nJuly 31, 2025\\nSyed Ali Hasan Shah\\n[Agentic AI] \\nJuly 31, 2025\\nTable + Of Contents\\n1. [Share This Article] \\n2. [Introduction] \\n3. [What Are + Autonomous AI Agents? Understanding the Fundamentals] \\n* [What Makes an + AI Agent Autonomous?] \\n* * [Autonomous Agents vs Traditional AI Systems] + \\n* * [Key Characteristics of Modern Autonomous Agents] \\n* [How Do Autonomous + AI Agents Work? Technical Architecture Explained] \\n* [Core Components of + Autonomous AI Systems] \\n* * [Types of Autonomous Agents by Intelligence + Level] \\n* * [Machine Learning Integration in Agent Architecture] \\n* [Autonomous + AI Agents 2025: Latest Developments and Technical Advancements] \\n* [Recent + Developments in Autonomous AI Agents 2025] \\n* * [Top Technical Advancements + Shaping 2025] \\n* * [Fully Autonomous AI Agents: What's Now Possible + in 2025] \\n* [Best Autonomous AI Agents Examples and Real-World Applications] + \\n* [Top Consumer Autonomous AI Agents] \\n* * [Enterprise and Business Applications] + \\n* * [Emerging Application Areas in 2025] \\n* * [Performance Metrics and + Success Stories] \\n* [The Role of Autonomous AI Agents in Business and Industry + Impact] \\n* [How Autonomous AI Agents Will Impact Industries in 2025] \\n* + * [Salesforce Autonomous Agents and CRM Integration] \\n* * [Autonomous Agents + Market Growth and Opportunities] \\n* * [Customer Service Revolution Through + AI Agents] \\n* [How to Build Autonomous AI Agents: Development and Implementation + Guide] \\n* [Essential Steps for Building Autonomous AI Agents] \\n* * [Best + Use Cases for Autonomous AI Agents] \\n* * [AI Agent Automation for Startups + in 2025] \\n* * [Integration with External Tools and Systems] \\n* * [Development + Challenges and Solutions] \\n* [Autonomous AI Agents vs Traditional Systems: + A Comprehensive Comparison] \\n* [Comparison of Autonomous AI Agents 2025 + vs Previous Generations] \\n* * [Most Advanced Autonomous AI Agents 2025: + Market Leaders] \\n* * [Human Workers vs Autonomous AI Agents: Collaborative + Future] \\n* * [Evolution from Reactive to Autonomous Systems] \\n* [Future + of Autonomous AI Agents: Trends and Predictions for 2025 and Beyond] \\n* + [How Autonomous AI Agents Are Shaping the Future] \\n* * [Top Trends in Autonomous + AI Agents 2025] \\n* * [What to Expect from Autonomous AI Agents in the Future] + \\n* * [Autonomous AI Agents in 2025 and Beyond: Technology Roadmap] \\n* + * [Challenges and Opportunities Ahead] \\n* [Geographic Trends and Regional + Variations in Autonomous AI Agent Adoption] \\n* [Factors Influencing Regional + Differences] \\n* * [Comparison of Regional Trends] \\n* * [Regional Market + Opportunities] \\n* [At a Glance: Key Takeaways] \\n* [Frequently Asked Questions] + \\n* [What are autonomous AI agents and how do they differ from regular AI?] + \\n* * [How can autonomous AI agents be used in business in 2025?] \\n* * + [What makes an AI agent truly autonomous?] \\n* * [What are the best examples + of autonomous AI agents available today?] \\n* * [How do I build autonomous + AI agents for my startup?] \\n* [Conclusion:] \\n* [Related Blogs] \\n## Share + This Article\\n![Illustration of an autonomous AI agent symbolizing the advancements + and potential of AI agents in 2025.] ## Introduction\\nAccording to recent + research, the global autonomous AI agents market is projected to reach[$9.9 + billion in 2025] and is anticipated to grow significantly to[$253.3 billion + by 2034], registering a strong CAGR of43.4%during the forecast period. This + explosive growth is driven by rapid enterprise adoption, continuous advancements + in artificial intelligence, and the expansion of automation across diverse + industries. North America is expected to command the largest market share + in 2025, holding about 40.7% of the global market.\\nThis comprehensive guide + explores autonomous AI agents’ fundamentals, applications, and 2025 + developments, providing essential insights for businesses, developers, and + decision-makers navigating AI transformation.\\n## What Are Autonomous AI + Agents? Understanding the Fundamentals\\nAutonomous AI agents are self-governing + systems that operate independently without constant human intervention, making + decisions and taking actions to achieve specific goals using machine learning + and environmental awareness.\\n[Autonomous AI agents] represent a significant + leap forward from traditional AI systems. Unlike conventional artificial intelligence + that requires explicit programming for every scenario, autonomous agents possess + the capability to learn, adapt, and make independent decisions based on their + environment and objectives. These systems combine[machine learning], natural + language processing, and real-time data analysis to create intelligent entities + that can operate with minimal human oversight.\\n**For example:**Learners + today can[learn French with Langua’s AI platform], which uses these + same principles to personalize instruction, track progress, and respond dynamically + to the user\u2019s input mirroring how autonomous agents behave in complex + business environments.\\nThe key distinction lies in their autonomy \u2013the + ability to perceive their environment, process information, make decisions, + and execute actions without waiting for human commands. This independence + makes them particularly valuable for businesses seeking to automate complex + processes, improve operational efficiency, and provide consistent service + delivery around the clock.\\n#####\",\"image\":\"https://kodexolabs.com/wp-content/uploads/2025/07/What-Are-Autonomous-AI-Agents-A-Complete-Guide-for-2025.webp\",\"favicon\":\"https://kodexolabs.com/wp-content/uploads/2024/11/1-05-2-150x150.webp\"},{\"id\":\"https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/agentic-ai-operating-model\",\"title\":\"Agentic + AI\u2019s strategic ascent: Shifting operations from incremental gains to + net-new impact\",\"url\":\"https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/agentic-ai-operating-model\",\"publishedDate\":\"2025-10-10T00:00:00.000Z\",\"author\":\"Meet + the authors\",\"text\":\"Agentic AI\u2019s strategic ascent: Shifting operations + from incremental gains to net-new impact | IBM\\n[Home] [COO, CSCO/Operations + and supply chain] Agentic AI\u2019s strategic ascent\\n[Home] [COO, CSCO/Operations + and supply chain] \\n# Agentic AI\u2019s strategic ascent\\nIn collaboration + with Oracle: AI is widening the gap between organizations that optimize what + exists and those that create what\u2019s next\u2014turning incremental gains + into net-new advantage.\\n![Decorative image] \\nIn collaboration with Oracle: + AI is widening the gap between organizations that optimize what exists and + those that create what\u2019s next\u2014turning incremental gains into net-new + advantage.\\n### Report additional content on the left\\n[Download the insights(will + open in a new tab)*![an arrow pointing to the right] *] \\n[Subscribe for + more insights from IBM IBV(will open in a new tab)*![an arrow pointing to + the right] *] \\n#### Key takeaways\\n* More than three-quarters of executives + say most of their AI investment has focused on improving existing processes + rather than developing net-new capabilities.\\n* Yet, 78% of C-Suiteexecutives + say achieving maximum benefit from agentic AI requires a new operating model.\\n* + Companies that excel in three key AI adoption areas are 32 times more likely + to achieve top-tier business performance than those with minimal implementation.\\n### + Rethinking how work works with agentic AI\\nAs companies continue to explore + how to perform better with artificial intelligence, a cohort of executives + are asking a question that goes far beyond simple deployment: \\\"What if + we rebuild our entire operating model around AI?\\\"\\nA growing number already + are\u2014and they're creating business outcomes that were previously impossible.\\nThese + leaders aren't just implementing AI tools\u2014they're pioneering entirely + new operating models around autonomous decision-making. The radical premise: + that agentic AI systems will make an increasing number of decisions, while + humans make the decisions that matter most, preserving human agency and expertise + where it is most crucial.\\nThe performance gap is widening rapidly. These + transformation-driven organizations are creating net-new business capabilities, + fundamentally changing what's possible at enterprise scale.\\n|||||\\n|By||By||\\n|2027||2030||\\n|twice + as many executives expect AI agents will make autonomous decisions in processes + and workflows compared to today.||fully autonomous robotic systems with embodied + AI will be operational realities across industries.||\\n|||||\\nThe companies + winning in the autonomous AI future won't be those that simply deploy the + most sophisticated AI agents. They'll be the organizations that abandon existing + operating models for entirely new ones designed around autonomous decision-making + capabilities.\\nIn the emerging AI transition, speed matters, but not all + efforts are equal. Some organizations are reaching truly strategic peaks while + others chase tactical AI.\\nTo better understand the real-world impact of + autonomous AI, the IBM Institute for Business Value (IBM IBV) surveyed 800 + C-suite executives in 20 countries across 19 industries. This report reveals + key elements of an autonomous AI operating model \u2013from management mindset + to workforce evolution to the role of trust and transparency. Alongside, we + highlight changing team structures, and examples of specific operating model + transformations in progress.\\nWe finish with an action guide that synthesizes + what to do now to build a synergistic operating model that can optimize agentic + AI\u2019s competitive benefits.\\n**Measurable impact: AI is expected to drive + double-digit improvements across the enterprise**\\n![A chart showing results + from IBM\u2019s own AI journey in business operations, as well as forecasted + metrics from non-IBM executives by 2027.] \\n**Perspective**\\nAgentic AI\u2019s + split screen\\nIn this recent survey of C-suite operational executives, we + uncovered a fascinating divergence in how companies are approaching agentic + AI. We asked two very telling questions, which revealed a sharp divide in + strategic focus.\\nFirst, we asked what percentage of their agentic AI initiatives + were focused on improving existing processes and how successful they were.\\nSecond, + we asked what percentage of their agentic AI initiatives were designed to + create entirely new workflow capabilities and how successful they were.\\nThe + responses divided our respondents into two distinct groups.\\n**The process-focused**\\nThis + group is technically getting it done; they're laser-focused on optimizing + existing workflows and are achieving measurable business impact. In short, + they are proficient at efficiency, but they have yet to crack the code on + true transformation.\\n**The transformation-driven**\\nThis more strategic + cohort is advancing agentic AI with a dual mandate. They are not only improving + existing workflows, but they are also successfully creating and building net-new + capabilities. Their vision extends beyond optimization; they are actively + reimagining their operating model. The data shows it's working. They are achieving + real-world value of AI and automation across their business operations.\\n### + AI transformation isn\u2019t a forecast, it\u2019s a deadline\\nToday, 24% + of executives say that AI agents take independent action in their organization. + By 2027, 67% expect that to be the case. Similarly, twice as many executives + expect autonomous decision-making from agentic AI in processes and workflows + by 2027 (57%) compared to today (28%). Even in highly regulated areas, transformation-driven + organizations are moving fast \u2014executives expect agentic AI to automate + an average of 29% of risk and compliance operations by 2027.And many anticipate + that AI agents will automate the innovation process itself, with 19% of executives + saying that is happening today and 48% of executives expecting this level + of innovation-related automation in 2027.\\nThese benchmarks represent a fundamental + change in the DNA of business. It\u2019s no wonder that:\\n> 78% of our respondents + agree that achieving maximum benefit from agentic AI requires a new operating + model.\\nAmid global economic uncertainty, 69% of them report a pressing need + to develop more agentic AI-enabled predictive and simulation modeling capabilities\u2014for + example, for financial forecasting, human resource talent and skilling analysis, + dynamic transportation and distribution routing and scheduling, and proactive + mitigation strategies with built-in resilience.\\nCustomer-touching operations + have emerged as primary arenas for agentic AI implementation. Organizations + are prioritizing sales forecasting, dynamic pricing based on inventory scenarios, + and intelligent customer order processing\u2014areas where autonomous system + intelligence can deliver immediate competitive advantages and measurable returns.\\nThe + appetite for change is real, but a critical chasm exists between aspiration + and execution. The vast majority of AI investment\u201478%, has been funneled + into improving existing processes. This is the difference between optimization + and transformation. Process-focused organizations are stuck in a loop of making + their current operations better, rather than making transformative leaps a + reality.\\n#### How decision-makers think differently about agentic AI\\nThe + organizations achieving genuine transformation are those that view agentic + AI not as a tool to do existing work faster but as a catalyst to do entirely + new work efficiently. Consider the compliance challenge: in highly regulated + industries, agentic AI isn\u2019t just flagging risks \u2014it\u2019s using + teams of specialized agents to continuously scan, interpret, and act on regulatory + changes across multiple jurisdictions and industries. Imagine one set of agents + monitoring shifting financial rules in Europe, another tracking evolving trade + sanctions in Asia, and yet another adjusting healthcare compliance workflows + in the U.S. All of them work in concert, automatically updating communications, + documentation, and approval processes in real time. The result? Risks can + be mitigated before they materialize, compliance teams spend less time on + repetitive reviews, and organizations can operate more confidently across + geographies without drowning in legal complexity.\\n> Because they are focused + on developing net-new capabilities, transformative organizations are asking + different questions: What becomes possible when systems can make decisions + autonomously?\\nAs well as: How do we redesign our value creation processes + around this capability? These enterprises must also recognize the shifting + landscape AI commoditization is driving. In other words, the window for first-mover + advantages in process innovations is narrowing. Therefore, while investing + in current operations isn't to be underestimated, the true key to competitive + differentiation now lies in bold, transformative leaps forward.\\nThe measurement + gap reveals this strategic divide clearly. Only 42% of the process-oriented + organizations identified in our analysis have developed new key performance + indicators (KPIs) to monitor AI agents' impact on business targets, compared + to nearly half of transformational organizations (see Action Guide for examples + of new KPIs). These leaders aren't just implementing agentic AI automation\u2014they're + measuring different outcomes entirely.\\n> \u201CWhen you don\u2019t have + good, objective KPIs, older ones take over. We see that technology is often + deployed around creating improvements with the infrastructure you currently + have.\u201D \\n**> Board advisor,\\n**>  \\n> United Kingdom\\n**Perspective**\\nPlug + in, level up: The agentic AI marketplace\\nAn agentic AI marketplace is a + digital marketplace where, instead of apps or software licenses, you\u2019re + browsing virtual shelves of autonomous problem\u2011solvers. Each \u201Cproduct\u201D + is an agentic AI \u2014a specialized algorithm that can take action, make + decisions, and adapt to changing conditions without waiting for human approval.\\nThese + marketplaces are more than procurement\",\"image\":\"https://www.ibm.com/thought-leadership/institute-business-value/uploads/en/1497_Report_thumbnail_1456x728_fb37c3a274.jpg\",\"favicon\":\"https://ibm.com/favicon.ico\"},{\"id\":\"https://www.imd.org/news/artificial-intelligence/the-rise-of-agentic-ai-a-new-era-in-workplace-productivity/\",\"title\":\"The + rise of agentic AI: A new era in workplace productivity\",\"url\":\"https://www.imd.org/news/artificial-intelligence/the-rise-of-agentic-ai-a-new-era-in-workplace-productivity/\",\"publishedDate\":\"2025-09-25T00:00:00.000Z\",\"author\":null,\"text\":\"The + rise of agentic AI: A new era in workplace productivity - IMD business school + for management and leadership courses\\n[Close] \\n[Home] [All News] [back + to Home] The rise of agentic AI: A new era in workplace productivity[back + to All News] \\nNews Stories · Artificial Intelligence - Innovation + - Communication - Knowledge Management\\n# The rise of agentic AI: A new era + in workplace productivity\\nTONOMUS Global Center for Digital and AI Transformation\\nBy + Jialu Shan and[Michael R. Wade] \\n**September 2025\\nIn December 2023, we + launched our first \u201C*Gen AI for Better Productivity*\u201D framework + with 12 categories. At the time, the conversation was dominated by familiar + use cases: chatbots that could handle customer queries, tools that could generate + content at scale, image generation, and systems capable of accelerating code + development. Soon after, we added new categories for transcription, translation + and localization, capabilities that quickly proved their utility in everyday + workflows. We have continued to update the framework with new tools.\\nFor + this update, we\u2019re adding a pivotal new category:**Agentic AI**. This + addition signals that business and technology leaders now see autonomous AI + agents as a distinct productivity paradigm: a transformative shift in how + we think about AI in the workplace.\\nBefore turning to this new frontier, + it\u2019s worth pausing to look at the AI applications that have already proven + their value. These tools have moved well beyond novelty and are now woven + into daily workflows.\\n![ - IMD Business School] \\n#### Continuity in core + capabilities\\nAI has changed how we work, and the tools we use are here to + stay. What began as experiments are now indispensable parts of our routines.\\n###### + **Generative content creation:**\\nAI tools that create and edit text, images, + presentations, and videos have become a huge part of our work lives. A[Microsoft + survey] found 75% of global knowledge workers already use generative AI at + work. Most users say AI saves them time (90%), helps them focus on more important + tasks (85%), and makes them more creative (84%). We’ve come a long way + from worrying that AI would stifle our creativity. In fact, a[2024 marketing + report] found that 15% of marketing teams “couldn’t live without + AI” because it made everyday tasks like creating presentations and transcribing + audio so much faster and smoother.\\n###### **Knowledge work and communication:**\\nAI + is also transforming how we handle information and communication. Companies + now use advanced chatbots for customer service and internal helpdesks, dramatically + improving response times. AI-powered research assistants can sift through + documents and summarize key findings in seconds, tasks that once took humans + hours of painstaking work. The impact on software development has been equally + transformative. A study by[MIT Sloan, Microsoft Research, and GitHub] found + that generative AI coding tools can reduce programming time by 56%, allowing + developers to focus on higher-level problem-solving rather than routine implementation.\\n###### + **Task automation:**\\nDaily tasks like managing email and scheduling meetings + are now so much easier thanks to AI. Modern email assistants can sort your + inbox and draft responses, and AI schedulers can find optimal meeting times + without endless back-and-forth among participants. These tools have begun + to seriously alleviate our digital overload. A[McKinsey report] estimated + work automation with GenAI and other technologies could boost productivity + growth 0.5 to 3.4% annually.\\nThe foundational uses of AI for productivity + have proven resilient and enduring. However, the competitive landscape continues + to evolve rapidly, with some companies rebranding (you\u2019ll notice the + fresh logos in our graph) or pivoting their focus as the market matures. Take + Tome: once a specialized presentation tool, it has shifted toward building + enterprise solutions, reflecting shifting market dynamics and heightened competition + in an increasingly crowded field.\\n#### **The new frontier: Agentic AI**\\nIf + generative AI is about extending human capability, Agentic AI is about autonomous + execution. Agentic AI refers to systems that can delegate, act, and learn + on their own to accomplish a goal. Instead of waiting for precise, step-by-step + prompts, an AI agent can take a high-level request like, \u201COrganize a + one-day offsite for my team next month,\u201D and then navigate multiple tasks + and tools to fulfil it. Agentic AI tools today can check calendars, research + options, book venues, schedule travel, and draft agenda.\\n[Over a quarter + of business leaders] say their organizations are already exploring agentic + AI to a significant degree. The vision is for these agents to process different + types of data (text, images, voice), coordinate with other AI services, and + learn from experience to reliably execute complex tasks. Emerging platforms + show the breadth of possibilities:**Manus**is a general AI agent that bridges + thoughts into actions, designed to independently carry out complex real-world + tasks without direct or continuous human guidance. Similarly,**Beam**focuses + on simplifying data workflows and**Sana**‘s AI agents are revolutionizing + enterprise search and knowledge management.**Stack AI**allows teams to stitch + together modular agents that orchestrate multiple AI services and traditional + software tools.**Devin**positions itself as an \u201CAI software engineer\u201D + that can independently write, test, and ship code. While not included in the + agentic category, agentic modes of the frontier models, like**ChatGPT**and**Claude**are + also starting to make waves.\\nThis doesn\u2019t mean companies are relinquishing + full control. Early trials often keep a human “in the loop” to + supervise. But the momentum is undeniable. Tech strategists now talk about + an “AI workforce”: a concept that emphasizes augmentation, not + replacement: AI agents as teammates rather than substitutes. For forward-looking + firms, this means new opportunities to streamline operations, from an AI agent + that triages IT support tickets overnight to one that summarizes market intelligence + for a strategy team.\\n#### **The bigger picture**\\nAI has quickly moved + from a novelty to a necessity in workplaces worldwide. The[same Microsoft + survey] mentioned earlier found that 79% of business leaders say their company + must adopt AI to stay competitive. This represents a generational consensus + that AI has transitioned from “nice-to-have” to “must-have” + status.\\nThe goal across all these domains is not simply to do more work. + It’s to free up our time and mental energy for what truly matters: the + strategic, creative, and interpersonal aspects of work and life that AI cannot + as easily replace. In just one year, AI’s role in productivity has grown + from a set of practical tools to a broader ecosystem that includes emerging + “colleague” agents. Yet, the mission remains consistent: to reduce + friction and enable people to focus on what they do best.\\nThe AI landscape + will undoubtedly keep evolving, but the lesson is already clear: the toolbox + is expanding, and those who adopt thoughtfully will move not just with the + tide, but ahead of it.\\nShare\\n![Share] \\nGet shareable link\\nCopy\\n[] + \\n[![Twitter]] \\n[![Facebook]] \\nRelated program\\n[![ - IMD Business School] + \\nDigital and AI Transformation Programs\\nDigital Transformation & AI\\nIMD + offers today a wide range of programs to help you seize the opportunities + of the new digital environment. And we will be launching more in the near + future! See all options available to you.\\n]\",\"image\":\"https://www.imd.org/wp-content/uploads/2022/11/IMD-Logo-Blue-on-White-safe-space-RGB-1200x630-1.jpg\",\"favicon\":\"https://www.imd.org/wp-content/uploads/2022/10/cropped-IMD-Logo-Blue-on-transparent-Square-for-Circle-Crop-RGB-32x32.png\"},{\"id\":\"https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5233324\",\"title\":\"Reframing + Operations with AI and Autonomous Agents: A Qualitative Content Analysis and + Adoption Framework\",\"url\":\"https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5233324\",\"publishedDate\":\"2025-04-20T00:00:00.000Z\",\"author\":\"See + all articles by Mohamed Abdelsalam Mohamed Mostafa\",\"text\":\"[Skip to main + content] \\n\\n[Download This Paper] \\n\\n[Open PDF in Browser] \\n\\n[Add + Paper to My Library] \\n\\nShare:\\n\\nPermalink\\n\\nUsing these links will + ensure access to this page indefinitely\\n\\n[Copy URL] \\n\\n[Copy DOI] \\n\\n# + Reframing Operations with AI and Autonomous Agents: A Qualitative Content + Analysis and Adoption Framework\\n\\n23 PagesPosted: 7 May 2025\\n\\n[See + all articles by Mohamed Abdelsalam Mohamed Mostafa] \\n\\n## [Mohamed Abdelsalam + Mohamed Mostafa] \\n\\nAin Shams University\\n\\nDate Written: April 20, 2025\\n\\n### + Abstract\\n\\nThis qualitative content analysis examines how organizations + can strategically integrate Artificial Intelligence (AI) and autonomous agents + to enhance operational efficiency, ensure legal and ethical compliance, and + foster workforce adaptability. Drawing on recent industry reports (e.g., Fujitsu, + 2025; McKinsey & Company, 2024), academic studies (e.g., Giorgetti et al., + 2025), and regulatory frameworks (e.g., NIST, 2023), we document significant + operational gains attributable to AI-including cycle-time reductions of 30-85%, + error-rate drops to below 1%, and anomalydetection accuracy reaching up to + 99%. Despite these compelling efficiency improvements, persistent challenges + hinder widespread and responsible adoption. These include critical governance + gaps under evolving regulations like GDPR and CCPA, the pervasive risks of + automation bias, the potential for workforce deskilling hazards, and inherent + technical barriers such as the explainability of \\\"black-box\\\" AI models + and the significant environmental impact related to AI's carbon footprint. + To address these multifaceted challenges, this paper proposes a practical + Four-Stage Adoption Framework and a detailed Five-Step Action Plan specifically + designed for practitioners. The findings underscore the necessity of achieving + a delicate balance between pursuing operational efficiency gains and ensuring + the sustainability and inclusivity of AI integration efforts within organizations.\\n\\n**Keywords:** + Artificial Intelligence, Autonomous Agents, Operational Efficiency, AI Governance, + Workforce Dynamics, Qualitative Content Analysis, Adoption Framework, Legal + Implications, Ethical Considerations, Sustainability\\n\\n**Suggested Citation:** + [Suggested Citation] \\n\\nAbdelsalam Mohamed Mostafa, Mohamed, Reframing + Operations with AI and Autonomous Agents: A Qualitative Content Analysis and + Adoption Framework (April 20, 2025). Available at SSRN: [https://ssrn.com/abstract=5233324] + or [http://dx.doi.org/10.2139/ssrn.5233324] \\n\\n### [Mohamed Abdelsalam + Mohamed Mostafa (Contact Author)] \\n\\n#### Ain Shams University ( [email] + )\\n\\nCairoEgypt\\n\\n## Do you have a job opening that you would like to + promote on SSRN?\\n\\n[Place Job Opening] \\n\\n## Paper statistics\\n\\nDownloads\\n\\n28\\n\\nAbstract + Views\\n\\n143\\n\\n[51 References] \\n\\nPlumX Metrics\\n\\n## Related eJournals\\n\\n- + [Operations Management eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Operations Management eJournal\\n\\n\\n\\nSubscribe to this fee journal for + more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n760\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n4,162\\n\\n- + [Operations Strategy eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Operations Strategy eJournal\\n\\n\\n\\nSubscribe to this fee journal for + more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n743\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n1,662\\n\\n- + [Technology, Operations Management & Production eJournal] \\n\\n\\n\\n[Follow] + \\n\\n\\n\\n\\n\\n#### Technology, Operations Management & Production eJournal\\n\\n\\n\\nSubscribe + to this fee journal for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n733\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n2,228\\n\\n- + [Artificial Intelligence eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Artificial Intelligence eJournal\\n\\n\\n\\nSubscribe to this fee journal + for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n373\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n25,390\\n\\n- + [Management of Innovation eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Management of Innovation eJournal\\n\\n\\n\\nSubscribe to this fee journal + for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n274\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n7,878\\n\\n- + [Innovation & Management Science eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Innovation & Management Science eJournal\\n\\n\\n\\nSubscribe to this fee + journal for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n259\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n8,447\\n\\n- + [Power, Politics, & Organizational Behavior eJournal] \\n\\n\\n\\n[Follow] + \\n\\n\\n\\n\\n\\n#### Power, Politics, & Organizational Behavior eJournal\\n\\n\\n\\nSubscribe + to this fee journal for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n254\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n1,436\\n\\n\\n\\n\\n\\n\\n\\n\\n\\nThis + Journal is curated by:\\n\\n\\n\\n**David C. Logan** at California Medical + Association (CMA) WellnessUniversity of Southern California - Marshall School + of Business\\n\\n- [Strategy Models for Firm Performance Enhancement eJournal] + \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### Strategy Models for Firm Performance + Enhancement eJournal\\n\\n\\n\\nSubscribe to this fee journal for more curated + articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n241\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n1,699\\n\\n- + [Emerging Research within Organizational Behavior eJournal] \\n\\n\\n\\n[Follow] + \\n\\n\\n\\n\\n\\n#### Emerging Research within Organizational Behavior eJournal\\n\\n\\n\\nSubscribe + to this fee journal for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n236\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n989\\n\\n\\n\\n\\n\\n\\n\\n\\n\\nThis + Journal is curated by:\\n\\n\\n\\n**David C. Logan** at California Medical + Association (CMA) WellnessUniversity of Southern California - Marshall School + of Business\\n\\n- [International Political Economy: Globalization eJournal] + \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### International Political Economy: + Globalization eJournal\\n\\n\\n\\nSubscribe to this fee journal for more curated + articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n84\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n9,667\\n\\n\\nFeedback\\n\\nFeedback + to SSRN\\n\\nFeedback\_(required)\\n\\nEmail\_(required)\\n\\nSubmit\"},{\"id\":\"https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work\",\"title\":\"AI + in the workplace: A report for 2025\",\"url\":\"https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work\",\"publishedDate\":\"2025-01-28T00:00:00.000Z\",\"author\":null,\"text\":\"AI + in the workplace: A report for 2025 | McKinsey\\n[Skip to main content] \\n# + Superagency in the workplace: Empowering people to unlock AI’s full + potential\\nJanuary 28, 2025| Report\\nHannah MayerLareina Yee[Michael Chui] + [Roger Roberts] \\nAlmost all companies invest in AI, but just 1 percent believe + they are at maturity. Our research finds the biggest barrier to scaling is + not employees—who are ready—but leaders, who are not steering + fast enough.\\n### ![] \\n##### Superagency in the workplace: Empowering people + to unlock AI\u2019s full potential\\n[(47 pages)] \\n**Artificial intelligence + has arrived in the workplace**and has the potential to be as transformative + as the steam engine was to the 19th-century Industrial Revolution.1\u201C[Gen + AI: A cognitive industrial revolution],\u201D McKinsey, June 7, 2024.With + powerful and capable large language models (LLMs) developed by Anthropic, + Cohere, Google, Meta, Mistral, OpenAI, and others, we have entered a new information + technology era. McKinsey research sizes the long-term AI opportunity at $4.4 + trillion in added productivity growth potential from corporate use cases.2\u201C[The + economic potential of generative AI: The next productivity frontier],\u201D + McKinsey, June 14, 2023.\\nTherein lies the challenge: the long-term potential + of AI is great, but the short-term returns are unclear. Over the next three + years, 92 percent of companies plan to increase their AI investments. But + while nearly all companies are investing in AI, only 1 percent of leaders + call their companies \u201Cmature\u201D on the deployment spectrum, meaning + that AI is fully integrated into workflows and drives substantial business + outcomes. The big question is how business leaders can deploy capital and + steer their organizations closer to AI maturity.\\nThis research report, prompted + by Reid Hoffman\u2019s book*Superagency: What Could Possibly Go Right with + Our AI Future*,3Reid Hoffman and Greg Beato,*Superagency: What Could Possibly + Go Right with Our AI Future*, Authors Equity, January 2025.asks a similar + question: How can companies harness AI to amplify human agency and unlock + new levels of creativity and productivity in the workplace? AI could drive + enormous positive and disruptive change. This transformation will take some + time, but leaders must not be dissuaded. Instead, they must advance boldly + today to avoid becoming uncompetitive tomorrow. The history of major economic + and technological shifts shows that such moments can define the rise and fall + of companies. Over 40 years ago, the internet was born. Since then, companies + including Alphabet, Amazon, Apple, Meta, and Microsoft have attained trillion-dollar + market capitalizations. Even more profoundly, the internet changed the anatomy + of work and access to information. AI now is like the internet many years + ago: The risk for business leaders is not thinking too big, but rather too + small.\\n![] \\nThis report explores companies\u2019 technology and business + readiness for AI adoption (see sidebar \u201CAbout the survey\u201D). It concludes + that employees are ready for AI. The biggest barrier to success is leadership.\\n## + About the survey\\n**To create our report,**we surveyed 3,613 employees (managers + and independent contributors) and 238 C-level executives in October and November + 2024. Of these, 81 percent came from the United States, and the rest came + from five other countries: Australia, India, New Zealand, Singapore, and the + United Kingdom. The employees spanned many roles, including business development, + finance, marketing, product management, sales, and technology.\\nAll the survey + findings discussed in the report, aside from two sidebars presenting international + nuances, pertain solely to US workplaces. The findings are organized in this + way because the responses from US employees and C-suite executives provide + statistically significant conclusions about the US workplace. Analyzing global + findings separately allows a comparison of differences between US responses + and those from other regions.\\n**Chapter 1**looks at the rapid advancement + of technology over the past two years and its implications for business adoption + of AI.\\n**Chapter 2**delves into the attitudes and perceptions of employees + and leaders. Our research shows that employees are more ready for AI than + their leaders imagine. In fact, they are already using AI on a regular basis; + are three times more likely than leaders realize to believe that AI will replace + 30 percent of their work in the next year; and are eager to gain AI skills. + Still, AI optimists are only a slight majority in the workplace; a large minority + (41 percent) are more apprehensive and will need additional support. This + is where millennials, who are the most familiar with AI and are often in managerial + roles, can be strong advocates for change.\\n![Agents, Robots, and Us] \\n### + Agents, Robots, and Us: What executives need to know about AI and work\\n**Thursday, + February 26, 10:30 a.m. - 11:00 a.m. ET / 4:30 p.m. - 5:00 p.m. CET**\\nJoin + McKinsey senior partners Alexis Krivkovich and Anu Madgavkar for a discussion + of how executives can build effective human-AI skill partnerships, prepare + their workforces, and make strategic decisions that position their organizations + for growth in the age of AI.\\n[Register here] \\n**Chapter 3**looks at the + need for speed and safety in AI deployment. While leaders and employees want + to move faster, trust and safety are top concerns. About half of employees + worry about AI inaccuracy and cybersecurity risks. That said, employees express + greater confidence that their own companies, versus other organizations, will + get AI right. The onus is on business leaders to prove them right, by making + bold and responsible decisions.\\n**Chapter 4**examines how companies risk + losing ground in the AI race if leaders do not set bold goals. As the hype + around AI subsides, companies should put a heightened focus on practical applications + that empower employees in their daily jobs. These applications can create + competitive moats and generate measurable ROI. Across industries, functions, + and geographies, companies that invest strategically can go beyond using AI + to drive incremental value and instead create transformative change.\\n**Chapter + 5**looks at what is required for leaders to set their teams up for success + with AI. The challenge of AI in the workplace is not a technology challenge. + It is a business challenge that calls upon leaders to align teams, address + AI headwinds, and rewire their companies for change.\\nChapter 1\\n## An innovation + as powerful as the steam engine\\nImagine a world where machines not only + perform physical labor but also think, learn, and make autonomous decisions. + This world includes humans in the loop, bringing people and machines together + in a state of superagency that increases personal productivity and creativity + (see sidebar \u201CAI superagency\u201D). This is the transformative potential + of AI, a technology with a potential impact poised to surpass even the biggest + innovations of the past, from the printing press to the automobile. AI does + not just automate tasks but goes further by automating cognitive functions. + Unlike any invention before, AI-powered software can adapt, plan, guide\u2014and + even make\u2014decisions. That\u2019s why AI can be a catalyst for unprecedented + economic growth and societal change in virtually every aspect of life. It + will reshape our interaction with technology and with one another.\\n> Scientific + discoveries and technological innovations are stones in the cathedral of human + progress.\\n> > Reid Hoffman, cofounder of LinkedIn and Inflection AI, partner + at Greylock Partners, and author\\nMany breakthrough technologies, including + the internet, smartphones, and cloud computing, have transformed the way we + live and work. AI stands out from these inventions because it offers more + than access to information. It can summarize, code, reason, engage in a dialogue, + and make choices. AI can lower skill barriers, helping more people acquire + proficiency in more fields, in any language and at any time. AI holds the + potential to shift the way people access and use knowledge. The result will + be more efficient and effective problem solving, enabling innovation that + benefits everyone.\\n## AI superagency\\n**What impact will AI have on humanity?**Reid + Hoffman and Greg Beato\u2019s book*Superagency: What Could Possibly Go Right + with Our AI Future*(Authors Equity, January 2025) explores this question. + The book highlights how AI could enhance human agency and heighten our potential. + It envisions a human-led, future-forward approach to AI.\\nSuperagency, a + term coined by Hoffman, describes a state where individuals, empowered by + AI, supercharge their creativity, productivity, and positive impact. Even + those not directly engaging with AI can benefit from its broader effects on + knowledge, efficiency, and innovation.\\nAI is the latest in a series of transformative + supertools, including the steam engine, internet, and smartphone, that have + reshaped our world by amplifying human capabilities. Like its predecessors, + AI can democratize access to knowledge and automate tasks, assuming humans + can develop and deploy it safely and equitably.\\nOver the past two years, + AI has advanced in leaps and bounds, and enterprise-level adoption has accelerated + due to lower costs and greater access to capabilities. Many notable AI innovations + have emerged (Exhibit 1). For example, we have seen a rapid expansion of context + windows, or the short-term memory of LLMs. The larger a[context window], the + more information an LLM can process at once. To illustrate, Google\u2019s + Gemini 1.5 could process one million tokens in February 2024, while its Gemini + 1.5 Pro could process two million tokens by June of that same year.4*The Keyword*,\_\u201COur + next-generation model:\_Gemini 1.5,\u201D blog entry by Sundar Pichai and + Demis Hassabis, Google, February 15, 2024;*Google for Developers*, \u201CGemini + 1.5 Pro 2M context window,\",\"image\":\"https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/superagency%20in%20the%20workplace%20empowering%20people%20to%20unlock%20ais%20full%20potential%20at%20work/superagency-report-1309018746-thumb-1536x1536.jpg\",\"favicon\":\"https://www.mckinsey.com/favicon.ico\"}],\"searchTime\":1233,\"costDollars\":{\"total\":0.015,\"search\":{\"neural\":0.005},\"contents\":{\"text\":0.01}}}" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json; charset=utf-8 + Date: + - Fri, 06 Feb 2026 21:28:30 GMT + Nel: + - '{"report_to":"cf-nel","success_fraction":0.0,"max_age":604800}' + Report-To: + - '{"group":"cf-nel","max_age":604800,"endpoints":[{"url":"https://a.nel.cloudflare.com/report/v4?s=3x4txrw7c%2Bh3sUhOZzce0fpFgtmve2GD3ahY1GVoPfb1CKbnYM7J3EopqZ2xLPuYIJzWr0lB0Hnw9W8mdbjvct4s5ZAoPS3eFA%3D%3D"}]}' + Server: + - cloudflare + Transfer-Encoding: + - chunked + access-control-allow-credentials: + - 'true' + cf-cache-status: + - DYNAMIC + content-security-policy: + - CSP-FILTERED + cross-origin-opener-policy: + - same-origin + cross-origin-resource-policy: + - same-origin + etag: + - ETAG-XXX + origin-agent-cluster: + - ?1 + referrer-policy: + - REFERRER-POLICY-XXX + strict-transport-security: + - STS-XXX + vary: + - Origin + x-content-type-options: + - X-CONTENT-TYPE-XXX + x-dns-prefetch-control: + - 'off' + x-download-options: + - noopen + x-frame-options: + - X-FRAME-OPTIONS-XXX + x-permitted-cross-domain-policies: + - X-PERMITTED-XXX + x-ratelimit-limit: + - '450' + x-ratelimit-remaining: + - '424' + x-ratelimit-reset: + - '1770413310' + x-xss-protection: + - X-XSS-PROTECTION-XXX + status: + code: 200 + message: OK +- request: + body: '{"query": "autonomous AI agents ethical governance reliability challenges + 2025", "startPublishedDate": "2025-01-01", "endPublishedDate": "2025-12-31", + "type": "auto", "contents": {"text": {"maxCharacters": 10000}}}' + headers: + Accept: + - '*/*' + Connection: + - keep-alive + Content-Length: + - '214' + Content-Type: + - application/json + User-Agent: + - X-USER-AGENT-XXX + accept-encoding: + - ACCEPT-ENCODING-XXX + x-api-key: + - X-API-KEY-XXX + method: POST + uri: https://api.exa.ai/search + response: + body: + string: "{\"requestId\":\"939e94ad18fd5bae81db1015a1c47271\",\"resolvedSearchType\":\"neural\",\"results\":[{\"id\":\"https://ai-frontiers.org/articles/the-challenges-of-governing-ai-agents\",\"title\":\"The + Challenges of Governing AI Agents\",\"url\":\"https://ai-frontiers.org/articles/the-challenges-of-governing-ai-agents\",\"publishedDate\":\"2025-03-03T00:00:00.000Z\",\"author\":null,\"text\":\"The + Challenges of Governing AI Agents | AI Frontiers\\n![] \\nThank you! Your + submission has been received!\\nOops! Something went wrong while submitting + the form.\\n[\\n![] \\nHigh-Bandwidth Memory: The Critical Gaps in US Export + Controls\\nErich Grunewald\\n\u2014Modern memory architecture is vital for + advanced AI systems. While the US leads in both production and innovation, + significant gaps in export policy are helping China catch up.\\nhigh-bandwidth + memory, HBM, DRAM, AI chips, GPU packaging, export controls, Bureau of Industry + and Security, BIS, U.S.-China tech competition, semiconductor manufacturing + equipment, FDPR, ASML immersion DUV lithography, SK Hynix, Samsung, Micron\\n] + \\n[\\n![] \\nMaking Extreme AI Risk Tradeable\\nDaniel Reti\\n\u2014Traditional + insurance can\u2019t handle the extreme risks of frontier AI. Catastrophe + bonds can cover the gap and compel labs to adopt tougher safety standards.\\nfrontier + AI, extreme AI risk, catastrophic AI events, AI liability, liability insurance, + catastrophe bonds, cat bonds, insurance-linked securities, capital markets, + AI regulation, AI safety standards, third-party audits, catastrophic risk + index, tail risk, systemic risk\\n] \\n[\\n![] \\nExporting Advanced Chips + Is Good for Nvidia, Not the US\\nLaura Hiscott\\n\u2014The White House is + betting that hardware sales will buy software loyalty \u2014a strategy borrowed + from 5G that misunderstands how AI actually works.\\n] \\n[\\n![] \\nAI Could + Undermine Emerging Economies\\nDeric Cheng\\n\u2014AI automation threatens + to erode the \u201Cdevelopment ladder,\u201D a foundational economic pathway + that has lifted hundreds of millions out of poverty.\\n] \\n[\\n![] \\nThe + Evidence for AI Consciousness, Today\\nCameron Berg\\n\u2014A growing body + of evidence means it\u2019s no longer tenable to dismiss the possibility that + frontier AIs are conscious.\\n] \\n[\\n![] \\nAI Alignment Cannot Be Top-Down\\nAudrey + Tang\\n\u2014Community Notes offers a better model \u2014where citizens, not + corporations, decide what \u201Caligned\u201D means.\\nAI alignment, attentiveness, + Community Notes, Taiwan, Audrey Tang, model specification, deliberative governance, + epistemic security, portability and interoperability, market design, Polis, + reinforcement learning from community feedback, social media moderation, civic + technology\\n] \\n[\\n![] \\nAGI's Last Bottlenecks\\nAdam Khoja\\n\u2014A + new framework suggests we\u2019re already halfway to AGI. The rest of the + way will mostly require business-as-usual research and engineering.\\nAGI, + artificial general intelligence, AGI definition, GPT-5, GPT-4, visual reasoning, + world modeling, continual learning, long-term memory, hallucinations, SimpleQA, + SPACE benchmark, IntPhys 2, ARC-AGI, working memory\\n] \\n[\\n![] \\nAI Will + Be Your Personal Political Proxy\\nBruce Schneier\\n\u2014By learning our + views and engaging on our behalf, AI could make government more representative + and responsive \u2014but not if we allow it to erode our democratic instincts.\\nAI + political proxy, direct democracy, generative social choice, ballot initiatives, + voter participation, democratic representation, AI governance, Rewiring Democracy, + Bruce Schneier, Nathan E. Sanders, policy automation, civic engagement, rights + of nature, disenfranchised voters, algorithmic policymaking\\n] \\n[\\n![] + \\nIs China Serious About AI Safety?\\nKarson Elmgren\\n\u2014China\u2019s + new AI safety body brings together leading experts \u2014but faces obstacles + to turning ambition into influence.\\nChina AI Safety and Development Association, + CnAISDA, China AI safety, World AI Conference, Shanghai AI Lab, Frontier AI + risk, AI governance, international cooperation, Tsinghua University, CAICT, + BAAI, Global AI Governance Action Plan, AI Seoul Summit commitments, Concordia + AI, Entity List\\n] \\n[\\n![] \\nAI Deterrence Is Our Best Option\\nDan Hendrycks\\n\u2014A + response to critiques of Mutually Assured AI Malfunction (MAIM).\\nAI deterrence, + Mutually Assured AI Malfunction, MAIM, Superintelligence Strategy, ASI, intelligence + recursion, nuclear MAD comparison, escalation ladders, verification and transparency, + redlines, national security, sabotage of AI projects, deterrence framework, + Dan Hendrycks, Adam Khoja\\n] \\n[\\n![] \\nSummary of \u201CIf Anyone Builds + It, Everyone Dies\u201D\\nLaura Hiscott\\n\u2014An overview of the core arguments + in Yudkowsky and Soares\u2019s new book.\\nIf Anyone Builds It Everyone Dies, + Eliezer Yudkowsky, Nate Soares, MIRI, AI safety, AI alignment, artificial + general intelligence, artificial superintelligence, AI existential risk, Anthropic + deceptive alignment, OpenAI o1, Truth\\\\_Terminal, AI moratorium, book summary, + Laura Hiscott\\n] \\n[\\n![] \\nAI Agents Are Eroding the Foundations of Cybersecurity\\nRosario + Mastrogiacomo\\n\u2014In this age of intelligent threats, cybersecurity professionals + stand as the last line of defense. Their decisions shape how humanity contends + with autonomous systems.\\nAI agents, AI identities, cybersecurity, identity + governance, zero trust, least privilege, rogue AI, autonomous systems, enterprise + security, trust networks, authentication and authorization, RAISE framework, + identity security, circuit breakers\\n] \\n[\\n![] \\nPrecaution Shouldn't + Keep Open-Source AI Behind the Frontier\\nBen Brooks\\n\u2014Invoking speculative + risks to keep our most capable models behind paywalls could create a new form + of digital feudalism.\\nopen-source AI, frontier models, precautionary policy, + digital feudalism, OpenAI, Meta, Llama, GPT-OSS, regulation, open development, + AI risk, legislation, policy debate, Berkman Klein Center\\n] \\n[\\n![] \\nThe + Hidden AI Frontier\\nOscar Delaney\\n\u2014Many cutting-edge AI systems are + confined to private labs. This hidden frontier represents America\u2019s greatest + technological advantage \u2014and a serious, overlooked vulnerability.\\nhidden + frontier AI, internal AI models, AI security, model theft, sabotage, government + oversight, transparency, self-improving AI, AI R&amp;D automation, policy + recommendations, national security, RAND security levels, frontier models, + AI governance, competitive advantage\\n] \\n[\\n![] \\nUncontained AGI Would + Replace Humanity\\nAnthony Aguirre\\n\u2014The moment AGI is widely released + \u2014whether by design or by breach \u2014any guardrails would be as good + as gone.\\nAGI, artificial general intelligence, open-source AI, guardrails, + uncontrolled release, existential risk, humanity replacement, security threat, + proliferation, autonomous systems, alignment, self-improving intelligence, + policy, global race, tech companies\\n] \\n[\\n![] \\nSuperintelligence Deterrence + Has an Observability Problem\\nJason Ross Arnold\\n\u2014Mutual Assured AI + Malfunction (MAIM) hinges on nations observing one another's progress + toward superintelligence \u2014but reliable observation is harder than MAIM's + authors acknowledge.\\nMAIM, superintelligence deterrence, Mutual AI Malfunction, + observability problem, US-China AI arms race, compute chips data centers, + strategic sabotage, false positives, false negatives, AI monitoring, nuclear + MAD analogue, superintelligence strategy, distributed R&amp;D, espionage + escalation, peace and security\\n] \\n[\\n![] \\nOpen Protocols Can Prevent + AI Monopolies\\nIsobel Moure\\n\u2014With model performance converging, user + data is the new advantage \u2014and Big Tech is sealing it off.\\nopen protocols, + AI monopolies, Anthropic MCP, context data lock-in, big tech, APIs, interoperability, + data portability, AI market competition, user context, model commoditization, + policy regulation, open banking analogy, enshittification\\n] \\n[\\n![] \\nIn + the Race for AI Supremacy, Can Countries Stay Neutral?\\nAnton Leicht\\n\u2014The + global AI order is still in flux. But when the US and China figure out their + path, they may leave little room for others to define their own.\\nAI race, + US-China competition, middle powers, export controls, AI strategy, militarization, + economic dominance, compute supply, frontier models, securitization, AI policy, + grand strategy, geopolitics, technology diffusion, national security\\n] \\n[\\n![] + \\nHow AI Can Degrade Human Performance in High-Stakes Settings\\nDane A. + Morey\\n\u2014Across disciplines, bad AI predictions have a surprising tendency + to make human experts perform worse.\\nAI, human performance, safety-critical + settings, Joint Activity Testing, human-AI collaboration, AI predictions, + aviation safety, healthcare alarms, nuclear power plant control, algorithmic + risk, AI oversight, cognitive systems engineering, safety frameworks, nurses + study, resilient performance\\n] \\n[\\n![] \\nHow the EU's Code of Practice + Advances AI Safety\\nHenry Papadatos\\n\u2014The Code provides a powerful + incentive to push frontier developers toward measurably safer practices.\\nEU + Code of Practice, AI Act, AI safety, frontier AI models, risk management, + systemic risks, 10^25 FLOPs threshold, external evaluation, transparency requirements, + regulatory compliance, general-purpose models, European Union AI regulation, + safety frameworks, risk modeling, policy enforcement\\n] \\n[\\n![] \\nHow + US Export Controls Have (and Haven't) Curbed Chinese AI\\nChris Miller\\n\u2014Six + years of export restrictions have given the U.S. a commanding lead in key + dimensions of the AI competition \u2014but it\u2019s uncertain if the impact + of these controls will persist.\\nchip, chips, china, chip export controls, + China semiconductors, hardware, AI hardware policy, US technology restrictions, + SMIC, Huawei Ascend, Nvidia H20, AI infrastructure, high-end lithography tools, + EUV ban, domestic chipmaking, AI model development, technology trade, computing + hardware, US-China relations\\n] \\n[\\n![] \\nNuclear Non-Proliferation Is + the Wrong Framework for AI Governance\\nMichael C. Horowitz\\n\u2014Placing + AI in a nuclear framework inflates expectations and distracts from practical, + sector-specific governance.\\n] \\n[\\n![] \\nA Patchwork of State AI Regulation + Is Bad. A Moratorium Is Worse.\\nKristin O\u2019Donoghue\\n\u2014Congress + is weighing a measure that would nullify thousands of state AI rules and bar + new ones \u2014upending federalism and halting the experiments that drive + smarter policy.\\nai regulation, state laws, federalism, congress, policy + innovation, legislative measures, state vs federal, ai governance, legal framework, + regulation moratorium, technology policy, experimental policy, state experimentation,\",\"image\":\"https://cdn.prod.website-files.com/66e38815e6bc34b7f5e1768b/67cb4f24080d3501fc61b03d_google-deepmind-BMn5Z-MGv5c-unsplash.jpg\",\"favicon\":\"https://cdn.prod.website-files.com/66e38815e6bc34b7f5e17684/67f371029f796428b28d25cf_aif-32.png\"},{\"id\":\"https://reports.weforum.org/docs/WEF_AI_Agents_in_Action_Foundations_for_Evaluation_and_Governance_2025.pdf\",\"title\":\"AI + Agents in Action: Foundations for Evaluation and ...\",\"url\":\"https://reports.weforum.org/docs/WEF_AI_Agents_in_Action_Foundations_for_Evaluation_and_Governance_2025.pdf\",\"publishedDate\":\"2025-11-27T00:00:00.000Z\",\"author\":null,\"text\":\"AI + Agents in Action: \\nFoundations for Evaluation \\nand Governance\\nWHITE + PAPER\\nNOVEMBER 2025\\nIn collaboration \\nwith Capgemini \\nImages: Adobe + Stock\\nDisclaimer\\nThis document is published by the \\nWorld Economic Forum + as a contribution \\nto a project, insight area or interaction. \\nThe findings, + interpretations and \\nconclusions expressed herein are a result \\nof a collaborative + process facilitated and \\nendorsed by the World Economic Forum \\nbut whose + results do not necessarily \\nrepresent the views of the World Economic \\nForum, + nor the entirety of its Members, \\nPartners or other stakeholders.\\n\xA9 + 2025 World Economic Forum. All rights \\nreserved. No part of this publication + may \\nbe reproduced or transmitted in any form \\nor by any means, including + photocopying \\nand recording, or by any information \\nstorage and retrieval + system.\\nContents\\nForeword 4\\nExecutive summary 5\\nIntroduction 6\\n1 + Evolving technical foundations of AI agents 8\\n1.1 The software architecture + of an AI agent 8\\n1.2 Communication protocols and interoperability 10\\n1.3 + Cybersecurity considerations 12\\n2 Foundations for AI agent evaluation and + governance 13\\n2.1 Classification 14\\n2.2 Evaluation 19\\n2.3 Risk assessment + 22\\n2.4 Governance considerations for AI agents: a progressive approach 25\\n3 + Looking ahead: multi-agent ecosystems 29\\nConclusion 30\\nContributors 31\\nEndnotes + 34\\nAI Agents in Action: Foundations for Evaluation and Governance 2\\nForeword\\nIn + recent years, organizations have moved beyond \\npredictive models and chat + interfaces to experiment \\nwith artificial intelligence (AI) in more transformative + \\nways. AI agents are now emerging as integrated \\ncollaborators in business, + public services and \\neveryday life. The adoption of AI agents could \\nbring + significant gains in efficiency, altered kinds of \\nhuman-machine interaction + and the advent of novel \\ndigital ecosystems.\\nThis transition faces multiple + obstacles that need to be \\naddressed. Moving from models to agents represents + \\nmore than a technical milestone and requires \\norganizations to rethink + how they design, evaluate and \\ngovern advanced agentic systems. Many companies + \\nare now questioning what agents can accomplish \\nalongside the practical + steps needed to adopt and \\ndeploy them safely, responsibly and effectively.\\nThis + paper was developed to help answer those \\nquestions. By mapping the evolving + foundations \\nof agentic systems, classifying their roles, \\nidentifying + new ways to evaluate them and outlining \\nprogressive governance approaches, + the paper \\noffers practical guidance for leaders navigating \\nadoption + in real-world contexts.\\nThrough the AI Governance Alliance, the World \\nEconomic + Forum and Capgemini are advancing \\nthis subject in collaboration with the + AI community, \\nsignalling that now is the time to prepare for an \\nagentic + future. If adopters start small, iterate \\ncarefully and apply proportionate + safeguards, \\nagents can be deployed in ways that amplify \\nhuman capabilities, + unlock productivity and \\nestablish a foundation for more complex multi-agent + \\necosystems to emerge over time. Unless a careful \\nand deliberate approach + to adoption is adopted, \\nuntested use cases could outpace oversight and + \\nlead to misaligned incentives, emergent risks and \\nloss of public trust.\\nAs + with any transformative technology, the \\nopportunities presented by AI agents + must be \\naccompanied by a responsibility to guide their \\ndevelopment and + deployment with care. Through \\ncross-functional efforts and collaborative + \\ngovernance, AI agents can be integrated in ways \\nthat amplify human ingenuity, + promote innovation \\nand improve overall quality of life. This paper is a + \\nstep in that direction, offering guidance to help early \\nadopters navigate + the complex and often uneven \\npath of AI agent adoption. \\nRoshan Gya\\nChief + Executive Officer, \\nCapgemini Invent\\nCathy Li\\nHead, Centre for AI \\nExcellence, + Member of \\nthe Executive Committee, \\nWorld Economic Forum\\nAI Agents + in Action: Foundations \\nfor Evaluation and Governance\\nNovember 2025\\nAI + Agents in Action: Foundations for Evaluation and Governance 3\\nExecutive + summary\\nThis report has been tailored mainly for adopters \\nof AI agents, + including decision-makers, technical \\nleaders and practitioners seeking + to integrate AI \\nagents into organizational workflows and services. \\nWhile + AI agents are gaining traction, there remains \\nlimited guidance on how to + design, test and \\noversee them responsibly. This paper aims to help \\nfill + that gap by providing a structured foundation for \\nthe safe and effective + deployment of these systems.\\nThe paper makes three key contributions. Firstly, + \\nit covers the technical foundations of AI agents, \\nincluding their architectures, + protocols and security \\nconsiderations. Secondly, it offers a functional + \\nclassification that differentiates agents by their role, \\nautonomy, authority, + predictability and operational \\ncontext. Thirdly, it suggests a progressive + governance \\napproach that directly connects evaluation and \\nsafeguards + to an agent\u2019s task scope and \\ndeployment environment.\\nTogether, these + elements guide adopters\_with a \\nconceptual blueprint for moving from experimentation\\nto + deployment. The report highlights the importance \\nof aligning adoption with + evaluation and governance \\npractices to ensure that AI agents are successfully + \\ndeployed while trust, safety and accountability \\nare maintained.\\nThis + paper explores the emergence of AI \\nagents, outlining their technical foundations, + \\nclassification, evaluation and governance to \\nsupport safe and effective + adoption. \\nAI Agents in Action: Foundations for Evaluation and Governance + 4\\nIntroduction\\nAI agents are gradually becoming embedded in \\nan increasing + number of tasks, workflows and \\nuse cases that span cloud and edge computing, + \\nleading the way to more widespread adoption. \\nAs the transition from + prototyping to deployment \\naccelerates, current adoption remains concentrated + \\namong early adopters. According to a recent global \\nsurvey of executives, + 82% of organizations plan to \\nintegrate agents within the next one to three + years, \\nindicating that most efforts are still in the planning or \\npilot + phase,1\\n while moving towards wider adoption.\\nThe concept of software + agents has been studied \\nfor decades in fields such as robotics, autonomous + \\nsystems and distributed computing. What is different\\ntoday is the rise + of data-driven models, particularly \\ngenerative artificial intelligence + (AI) and large language\\nmodels (LLMs), which are enabling the emergence + \\nof a new generation of LLM-based agents. These \\nsystems can generate + plans, simulate reasoning \\nand adapt their behaviour through feedback \\nmechanisms + in ways that were previously not \\npossible. This evolution has sparked a + new \\nwave of experimentation, with researchers and \\ncompanies rapidly + creating prototypes of agents \\nin various fields. This report focuses mainly + on \\nLLM-based agents (\u201CAI agents\u201D is sometimes \\nused in short), + whose growing capabilities create \\nboth significant opportunities for adoption + and a \\nnew set of challenges in governance and safety.\\nAI agents are shifting + from prototypes to \\ndeployment, bringing both transformative \\nopportunities + and novel governance challenges.\\nFIGURE 1 Foundations for the responsible + adoption of AI agents\\nFunctional\\nclassification\\n2\\nDefine the\\nagent\u2019s + role\\nEvaluation and\\ngovernance\\n3\\nScale with\\nconfidence\\nTechnical\\nfoundations\\n1\\nLay + the\\ngroundwork\\nAI Agents in Action: Foundations for Evaluation and Governance + 5\\nLLM-based AI agents, for example, introduce new \\nrisks such as goal + misalignment, behavioural drift, \\ntool misuse and emergent coordination + failures \\nthat traditional software governance models are \\nunable to manage. + Unlike conventional software, \\nagents are increasingly assuming roles that + \\nresemble those of human decision-makers rather \\nthan static tools. This + means that governance \\nmodels designed solely for access control and \\nsystem + reliability are no longer sufficient. A more \\nuseful comparison is the governance + applied \\nto human users, who must earn permissions, \\naccountability and + trust by demonstrating\_performance\\nover time. Similarly, trust in AI agents + can be \\nestablished by testing their behaviour against \\nvalidated cases, + running them in human-in-the\\u0002loop configurations and gradually expanding + \\nautonomy only once reliability has been sufficiently \\ndemonstrated. In + both cases, the principle of least \\nprivilege remains essential, with access + limited to \\ninformation and actions necessary for the task.\\nThis report + aims to provide\_a\_forward-looking\_analysis \\nof the evolving landscape\_of\_AI\_agents, + focusing \\non the capabilities, infrastructure,\_classification and \\nsafeguards + necessary for responsible deployment.\\nTo\_this end, it is structured around + four foundational \\npillars across classification, evaluation, risk assessment + \\nand governance, which together form the foundation \\nfor a progressive + approach to adoption and \\ndeployment. Figure 1 presents the general content + \\nof this report, which helps guide the responsible \\nadoption and deployment + of AI agents.\\nThe goal is to equip adopters, providers, technical \\nleaders, + organizational decision-makers and other \\nstakeholders with a shared understanding + of the \\ncurrent state of agentic systems and emerging \\noversight practices. + Building on established \\nAI governance principles and frameworks, \\nsuch + as those developed by the Organisation \\nfor Economic Co-operation and Development + \\n(OECD),2\\n National Institute of Standards and \\nTechnology (NIST),3\\n + International Organization for \\nStandardization (ISO)/International Electrotechnical + \\nCommission (IEC)4\\n and others, this paper \\nintroduces additional principles + addressing \\nautonomy, authority, operational context and \\nsystemic risk + that extend existing governance \\nguidance from an agent-focused lens. The + \\ninsights have been informed by working group \\nmeetings, workshops and + extensive interviews with \\nmembers of the Safe Systems and Technologies + \\nworking group of the AI Governance Alliance.\\nAI Agents in Action: Foundations\"},{\"id\":\"https://rierino.com/blog/ai-agent-governance-trust-autonomy\",\"title\":\"AI + Agent Governance: Building Trust in the Age of Autonomy\",\"url\":\"https://rierino.com/blog/ai-agent-governance-trust-autonomy\",\"publishedDate\":\"2025-10-27T00:00:00.000Z\",\"author\":null,\"text\":\"AI + Agent Governance: Building Trust in the Age of Autonomy\\n[![Rierino Logo]![Rierino + Logo]] \\nPlatform\\nSolutions\\nPartners\\nResources\\n[\\nTake a Tour\\n] + \\n[CONTACT US] [GET STARTED] \\n[![Rierino Logo]] [![Menu]] \\n# AI Agent + Governance: Building Trust in the Age of Autonomy\\nOctober 27, 2025\u20229minutes\\n[] + [] [] \\n![AI Agent Governance: Building Trust in the Age of Autonomy] \\nKey + Topics\\nGen AI\\nAI Governance\\nLow Code\\nAgentic AI\\nOrchestration\\nObservability\\nArtificial + intelligence is entering a new phase defined not just by intelligence, but + by**autonomy**. From marketing assistants that launch personalized campaigns + to supply chain bots that reorder inventory on their own, the rise of*agentic + AI*marks a turning point in how software operates. Yet as these systems gain + more independence, the need for governance and guardrails becomes exponentially + more critical.\\nAccording to*McKinsey*,[just 1%] of surveyed organizations + believe that their AI adoption has reached full maturity. While many enterprises + are experimenting with autonomous agents and generative models, few have established + the safety, accountability, and alignment mechanisms required to deploy them + responsibly at scale. The challenge isn\u2019t just about building smarter + agents; it\u2019s about ensuring those agents act within**trusted, observable + boundaries**.\\nThe*Institute for AI Policy and Strategy*\u2019s recent field + guide highlights this emerging tension: as decision-making moves from algorithms + to autonomous agents, governance must evolve from reactive oversight to[proactive + orchestration]. In other words, it\u2019s no longer enough to manage data + and models, organizations must now manage AI behavior.\\nAt Rierino, we\u2019ve + seen this shift firsthand through enterprises scaling their generative and + agentic AI initiatives from pilots to production. One insight stands out: + the more powerful and connected AI agents become, the more essential**governed + orchestration**is to balance autonomy with control.\\nThis article explores + what effective AI agent governance looks like \u2014the risks it mitigates, + the layers that make it work, and how**orchestration-first low-code**architecture + enables trust, transparency, and security in the age of autonomous systems.\\n# + Why Governance Matters More for Agentic AI\\nThe concept of**AI governance**isn\u2019t + new \u2014most organizations already apply policies around data use, model + validation, and compliance. But agentic AI changes the scope of what must + be governed. Traditional systems make predictions or recommendations; agents + go a step further by initiating and executing actions in live environments. + That shift from decision to autonomous execution fundamentally transforms + the governance challenge.\\n> > In classical AI, risk management focuses on + **> how\\n**> a model behaves. With agentic systems, it\u2019s equally about + **> what\\n**> the model can do \u2014and under what circumstances.\\n> Each + agent\u2019s capacity to perceive, decide, and act introduces new layers of + operational risk that data ethics frameworks or model fairness audits alone + cannot address. This new paradigm requires oversight across three fronts:\\n* + **Action Control:**ensuring agents can act only within predefined scopes and + policies.\\n* **Inter-Agent Coordination:**managing how multiple agents interact + or collaborate to prevent conflict or duplication.\\n* **Continuous Accountability:**maintaining + full traceability over every decision, event, and action for audit and compliance.\\nAs + enterprises move beyond proof-of-concept deployments into production-scale + systems, the absence of such controls can quickly lead to unintended automation, + data exposure, or decision drift, issues that traditional governance frameworks + were never designed to detect in real time.\\nAgentic AI, in other words, + doesn\u2019t just need smarter governance; it needs**dynamic, execution-aware**governance + that can monitor, interpret, and intervene in the flow of autonomous operations + as they unfold.\\n# Top Governance Challenges for Autonomous Agents\\nDespite + growing interest in agentic AI, most enterprises still underestimate the complexity + of governing autonomy. The very qualities that make agents valuable, independence, + adaptability, and collaboration, also introduce new and often invisible risks. + Below are five of the most critical governance challenges organizations face + as they move from experimentation to large-scale deployment.\\n**1. Unbounded + Autonomy:**Without defined boundaries, autonomous agents may take actions + that exceed their intended authority or cascade across systems. For example, + a logistics optimization agent might reroute shipments to improve efficiency + but inadvertently disrupt existing service-level agreements or inventory priorities. + This challenge arises when autonomy is granted without clearly codified limits. + Agents may operate with good intent but without awareness of strategic, ethical, + or operational constraints.\\n**2. Opaque Decision Flows:**As agents self-adapt + or collaborate, their internal decision logic can become difficult to trace + or explain. In practice, this means a business cannot easily determine why + a specific action occurred, what data triggered it, or whether it aligned + with compliance policies. When decision paths lack transparency, accountability + mechanisms fail, leaving enterprises unable to reconstruct cause-and-effect + relationships during audits or incident analysis.\\n**3. Runtime Drift:**Over + time, agents may diverge from their initial parameters as they learn from + evolving data and context. A recommendation agent, for instance, might start + prioritizing outcomes that maximize short-term engagement over long-term customer + value if its reward signals shift subtly. Governance frameworks must therefore + include continuous validation and feedback loops to detect behavioral drift + before it compounds into systemic bias or unintended automation.\\n**4. Coordination + Conflicts:**When multiple agents operate within shared environments, their + independent actions can overlap or contradict one another. A pricing agent + might apply a discount while a revenue management agent simultaneously adjusts + base prices, both rational individually but conflicting collectively. These + conflicts reveal the need for orchestration and dependency management, ensuring + that agents don\u2019t act in isolation but within an agreed-upon coordination + protocol.\\n**5. Auditability and Accountability Gaps:**Traditional monitoring + systems were not designed for real-time, autonomous decision chains. Without + detailed logs of every decision and action, enterprises risk compliance blind + spots, especially in regulated sectors like banking, telecom, or government. + True accountability requires fine-grained observability: a continuous record + of inputs, outputs, and actions that can be queried or rolled back at any + point in time.\\n**6. Resource and Cost Inefficiency:**Unsupervised agents + can also introduce hidden operational costs when allowed to execute actions + or queries without optimization. Excessive API calls, non-strategic task repetition, + or uncontrolled use of large language model tokens can quickly inflate spend + without delivering proportional value. For instance, a content generation + agent might consume thousands of unnecessary tokens refining an output that + never reaches production. Effective governance must therefore extend to resource + orchestration and budget control, ensuring that autonomy remains economically + as well as operationally sustainable.\\nAgentic AI requires governance that + extends beyond model oversight to execution-time control \u2014the ability + to monitor, interpret, and intervene in agentic behavior as it unfolds. This + shift sets the foundation for the next layer: designing effective,**multi-tiered + governance frameworks**that balance autonomy with accountability.\\n# What + Effective AI Agent Governance Looks Like\\nEffective governance for agentic + AI isn\u2019t about limiting innovation, it\u2019s about shaping autonomy + within safe, observable boundaries. As enterprises move from experimentation + to scaled deployment, governance must evolve into a multi-layered system that + defines rules, enforces them during execution, and ensures continuous accountability + afterward.\\nAn emerging best practice is to view AI agent governance across + three complementary layers: Policy, Platform, and Oversight. Each plays a + distinct role in ensuring that autonomy remains productive, compliant, and + aligned with human intent.\\n## 1. Policy Layer: Defining the Rules of Autonomy\\nThe + foundation of any governance model is a clear, codified policy framework. + This layer sets the**limits of autonomy**, outlining what an agent can do, + under what conditions, and with what access rights. For example, in a government + services environment, an AI intake agent may automatically classify and route + citizen requests but must flag any case that involves sensitive personal data + for manual verification.\\nThis layer acts as the blueprint for accountability. + When implemented through[composable low-code architectures], these rules can + be versioned, reused, and adapted across teams, ensuring consistent governance + as agents evolve.\\n## 2. Platform Layer: Enforcing Guardrails at Runtime\\nEven + the best policies mean little if they can\u2019t be enforced dynamically. + The platform layer ensures that rules set at design time translate into**real-time + control and observability**during execution. This is where orchestration becomes + essential.\\nThrough centralized execution flows, enterprises can monitor + agent actions as they happen, validate them against defined policies, and + trigger fallback or escalation mechanisms when anomalies occur. As an example, + in financial operations, an agent responsible for reconciling cross-border + transactions may autonomously process low-value payments but must request + human validation for high-value or cross-jurisdiction transfers.\\nThis\",\"image\":\"https://res.cloudinary.com/rierino/image/upload/f_auto/media/blog/media/hero/ai-agent-governance-trust-autonomy-hero.png\",\"favicon\":\"https://rierino.com/favicon.ico\"},{\"id\":\"https://www.aigl.blog/ai-agent-governance-a-field-guide-iaps-apr-2025/\",\"title\":\"AI + Agent Governance: A Field Guide (IAPS, Apr 2025)\",\"url\":\"https://www.aigl.blog/ai-agent-governance-a-field-guide-iaps-apr-2025/\",\"publishedDate\":\"2025-11-07T00:00:00.000Z\",\"author\":\"Jakub + Szarmach\",\"text\":\"AI Agent Governance: A Field Guide (IAPS, Apr 2025)\\n[AI + Governance Library] **[Subscribe] \\n[**Sign up] [**Sign in] \\n[Log in] [Subscribe] + \\n**\\n# AI Agent Governance: A Field Guide (IAPS, Apr 2025)\\nA practical, + outcomes-driven playbook for governing AI agents\u2014covering risks, adoption + pathways, and a five-part taxonomy of interventions (Alignment, Control, Visibility, + Security & robustness, Societal integration), with examples and vignettes.\\n**********\\n* + [![Jakub Szarmach]] \\n[Jakub Szarmach] \\nNov 7, 20254 min read\\n![AI Agent + Governance: A Field Guide (IAPS, Apr 2025)] \\nOn this page**\\n[\\nAI Agent + Governance A Field Guide\\nAI Agent Governance A Field Guide.pdf\\n2 MB\\ndownload-circle\\n] + \\n## **\u26A1 Quick Summary**\\nThis field guide from the Institute for AI + Policy and Strategy maps the near-term reality of agentic AI\u2014LLM-scaffolded + systems that plan, remember, use tools, and act\u2014and the governance work + needed before they scale to \u201Cmillions or billions.\u201D It distinguishes + hype from evidence: agents can already add value (customer support, AI R&D, + some cyber tasks) but underperform humans on long, open-ended workflows. The + core governance question is how to keep autonomous, tool-using systems safe, + steerable, and accountable at scale. The report\u2019s keystone is a five-category + intervention taxonomy\u2014Alignment, Control, Visibility, Security & + robustness, and Societal integration\u2014spanning model, system, and ecosystem + layers, with concrete mechanisms (e.g., shutdown/rollback, agent IDs, activity + logging, liability regimes). The authors argue the capability curve (especially + \u201Ctest-time compute\u201D models like o-series) will bend upward faster + than governance capacity unless institutions, standards, and infrastructure + are built now. (See the intervention table and scenarios on pp. 10\u201312 + and 34\u201347; agent architecture diagram on p.14.)\\n## \U0001F9E9What\u2019s + Covered\\n* **Current state of agents.**A crisp definition (\u201CAI systems + that can autonomously achieve goals in the world\u201D) and the scaffolding + pattern around frontier models: reasoning/planning, memory, tool-use, and + multi-agent collaboration (p.14\u201317). Benchmarks summarized across GAIA, + METR Autonomy Evals, SWE-bench (incl. Verified), WebArena, OSWorld, RE-bench, + CyBench, etc., showing sharp fall-offs as human task time exceeds \\\\~1 hour + and on open, visually grounded, or long-horizon tasks (pp. 16\u201321; Appendix + pp. 52\u201355).\\n* **Pathways to better agents.**Improvements via stronger + \u201Ccontroller\u201D models and better scaffolding/orchestration; rise of**test-time + compute**(o1/o3, DeepSeek-R1) enabling longer, higher-quality reasoning; memory/tool + ecosystems; multi-agent orchestration (pp. 21\u201323).\\n* **Adoption lens.**Where + early ROI lands: customer relations (Klarna case), AI/ML R&&D (engineering-heavy + steps, fast feedback), and cybersecurity (mixed\u2014but promising\u2014signals). + Adoption governed by performance, cost (often 1/30th of US bachelor median + wage on certain tasks), and**reliability**(pp. 22\u201328).\\n* **Risk landscape.**Four + buckets: (1)**Malicious use**(scaled disinformation, cyber offense, dual-use + bio); (2)**Accidents & loss of control**(from mundane reliability failures + to \u201Crogue replication\u201D and scheming/deception); (3)**Security**(expanded + attack surface from tools, memory, and agent-agent dynamics, incl. \u201Cinfectious + jailbreaks\u201D); (4)**Systemic risks**(labor displacement, power concentration, + democratic erosion, market \u201Chyperswitching\u201D) (pp. 27\u201333).\\n* + **Agent governance as a distinct field.**Why agents change the governance + calculus: direct world actions, opacity + speed, multi-agent effects, and + new levers at the ecosystem layer (payments, browsers, IDs). Priorities include + capability/risk monitoring, lifecycle controls, incentivizing defensive/beneficial + use, adapting law/policy (pp. 31\u201335).\\n* **Intervention taxonomy (core + contribution).**\\n* **Alignment**(e.g., multi-agent RL, risk-attitude alignment, + CoT paraphrasing, alignment evals).\\n* **Control**(e.g., rollback infrastructure, + shutdown/interrupt/timeouts, tool/action restrictions, control protocols && + evaluations).\\n* **Visibility**(e.g.,**Agent IDs**, activity logging, cooperation-relevant + capability evals, RL**reward reports**).\\n* **Security & robustness**(e.g., + access control tiers, adversarial robustness testing,**sandboxing**, rapid + response for adaptive defense).\\n* **Societal integration**(e.g.,**liability + regimes**, commitment devices/smart contracts, equitable agent access schemes,**law-following + agents**).Each category includes plain-language vignettes showing how a measure + works in practice (pp. 36\u201347).\\n* **Two futures.**\u201CAgent-Driven + Renaissance\u201D vs. \u201CAgents Run Amok\u201D scenarios make the stakes + concrete\u2014highlighting the need for IDs, logging, rollbacks, defensive + agents, and access/benefit-sharing (pp. 10\u201313).\\n* **Research & + capacity gap.**Few teams are working on agent governance relative to the resources + pouring into capability-building; the guide flags funders, workshops, and + priority research threads (pp. 5\u20137, 31\u201335, 48\u201350).\\n(See the**taxonomy + table**on p.6 and pp. 36\u201347 for detailed exemplars;**agent architecture + diagram**on p.14;**benchmark summaries**on pp. 18\u201321 and Appendix.)\\n# + **\U0001F4A1 Why it matters?**\\nAgents aren\u2019t just chatbots\u2014they\u2019re**actors**. + Once systems can plan, remember, and click \u201Cbuy\u201D (orexec), classic + safety levers (content filters, pre-deployment evals) aren\u2019t enough. + Governance must move**into the loop**: IDs and logs for traceability; rollback/shutdown + for damage control; sandboxing and access controls for containment; liability + and law-following to align incentives; and alignment methods suited to long-horizon, + multi-agent settings. Building this stack**before**mass deployment is the + difference between autonomous productivity and autonomous externalities. The + report delivers a concrete, actionable blueprint. (See pp. 34\u201347.)\\n# + **\u2753 What\u2019s Missing**\\n* **Operational maturity levels.**A readiness/assurance + model (e.g., Agent-ML levels with required controls/evals per tier) would + help buyers and regulators phase adoption.\\n* **Quantified control efficacy.**The + guide catalogs controls but offers limited empirical evidence on success rates + (e.g., mean time-to-shutdown, rollback coverage, bypass rates).\\n* **Supply-chain + specifics.**More detail on browser/OS, payment, and cloud co-regulation (who + hosts agent IDs, where logs live, cross-jurisdiction handling).\\n* **Human + factors.**Guidance for organizational design: who owns rollback authority, + separation of duties, incident playbooks, and \u201Ckill-switch\u201D drills.\\n* + **Evaluation economics.**Costs and sampling strategies to make agent evals + reproducible at scale (the Appendix notes time/cost, but buyers need budgeting + templates).## \U0001F465Best Fo**r**\\n* **CISOs/CIOs & platform owners**designing + safe agent platforms (browser/OS, payments, cloud, enterprise suites).\\n* + **Risk, policy, & compliance leaders**crafting internal standards and + procurement criteria for agentic systems.\\n* **Regulators & standards + bodies**mapping controls to duties (IDs, logging, rollback, liability apportionment).\\n* + **Research leads & safety teams**prioritizing evaluations, red-teaming, + and adaptive defenses for agents.\\n* **Product strategists & founders**building + vertical agents who need a governance-as-a-feature roadmap.## \U0001F4C4Source + De**tails**\\n* **Title:***AI Agent Governance: A Field Guide*\\n* **Authors:**Jam + Kraprayoon, Zoe Williams, Rida Fayyaz\\n* **Publisher:**Institute for AI Policy + and Strategy (IAPS)\\n* **Date / Length:**April 2025 / 63 pp.\\n* **Standout + visuals:**Agent architecture diagram (p.14); intervention taxonomy (pp. 6 + && 36\u201347); benchmark tables && notes (pp. 18\u201321; + Appendix).## \U0001F4DDThanks**to**\\nAcknowledged contributors include Alan + Chan, Cullen O\u2019Keefe, Shaun Ee, Ollie Stephenson, Cristina Schmidt-Ib\xE1\xF1ez, + Clara Langevin, and Matthew Burtell (p.50).\\nIf desired, an AIGL one-pager + can be prepared distilling the taxonomy into procurement and control checklists + per deployment tier.\\n[Guide] \\n[**Share] [**Share] [**Share] [**Share] + [**Email] **Copy\\nAbout the author\\n[![Jakub Szarmach]] \\n## [Jakub Szarmach] + \\nLatest\\n[![Architecting Secure Enterprise AI Agents with MCP] \\n### Architecting + Secure Enterprise AI Agents with MCP\\n06 Feb 2026\\n] [![AI Security Institute + \u2013Frontier AI Trends Report (December 2025)] \\n### AI Security Institute + \u2013Frontier AI Trends Report (December 2025)\\n06 Feb 2026\\n] [![AI Model + Risk Management Framework] \\n### AI Model Risk Management Framework\\n06 + Feb 2026\\n] \\n#### AI Governance Library\\nCurated Library of AI Governance + Resources\\nSubscribe\\nGreat! Check your inbox and click the link.\\nSorry, + something went wrong. Please try again.\\nRead next\\n[### Architecting Secure + Enterprise AI Agents with MCP\\n**] [### AI Incident Response: Adapting Proven + Complex Systems Engineering Practices for AI-Enabled Systems\\n**] [### AI + for Security and Security for AI: Navigating Opportunities and Challenges\\n**] + [### Administering and Governing Agents\\n**] [### A practical guide to implementing + AI ethics governance\\n**] \\n## AI Governance Library\\nCurated Library of + AI Governance Resources\\nSubscribe\\nGreat! Check your inbox and click the + link.\\nSorry, something went wrong. Please try again.\\n![AI Governance Library] + \\n******\\nGreat! You\u2019ve successfully signed up.\\nWelcome back! You've + successfully signed in.\\nYou've successfully subscribed to AI Governance + Library.\\nYour link has expired.\\nSuccess! Check your email for magic link + to sign-in.\\nSuccess! Your billing info has been updated.\\nYour billing + was not updated.\\n**\",\"image\":\"https://www.aigl.blog/content/images/2025/11/AI-Agent-Governance-A-Field-Guide.jpg\",\"favicon\":\"https://www.aigl.blog/content/images/size/w256h256/2025/03/LOGO_no_background-removebg-preview.png\"},{\"id\":\"https://www.ijircst.org/DOC/3-AI-Governance-in-the-Era-of-Agentic-Generative-AI-and-AGI-Frameworks-Risks-and-Policy-Directions.pdf\",\"title\":\"AI + Governance in the Era of Agentic Generative AI and AGI: Frameworks, Risks, + and Policy Directions\",\"url\":\"https://www.ijircst.org/DOC/3-AI-Governance-in-the-Era-of-Agentic-Generative-AI-and-AGI-Frameworks-Risks-and-Policy-Directions.pdf\",\"publishedDate\":\"2025-09-04T00:00:00.000Z\",\"author\":\"Satyadhar + Joshi\",\"text\":\"International Journal of Innovative Research in Computer + Science and Technology (IJIRCST)\\nISSN (Online): 2347-5552, Volume-13, Issue-5, + September 2025\\nDOI: https:/doi.org/10.55524/ijircst.2025.13.5.3\\nArticle + ID IRP-1674, Pages 14-24\\nwww.ijircst.org\\nInnovative Research Publication + 14\\nAI Governance in the Era of Agentic Generative AI and AGI: \\nFrameworks, + Risks, and Policy Directions\\nSatyadhar Joshi\\nIndependent Researcher, Alumnus, + International MBA, Bar-Ilan University, Israel\\nCorrespondence should be + addressed to Satyadhar Joshi;\\n Received 27 July 2025; Revised 11 August + 2025; Accepted 26 August 2025\\nCopyright \xA9 2025 Made Satyadhar Joshi. + This is an open-access article distributed under the Creative Commons Attribution + License, \\nwhich permits unrestricted use, distribution, and reproduction + in any medium, provided the original work is properly cited.\\nABSTRACT- The + accelerating development of agentic \\nartificial intelligence (AI) and the + prospect of artificial \\ngeneral intelligence (AGI) create unprecedented + \\nopportunities alongside complex governance challenges. \\nThis paper examines + the ethical, regulatory, and technical \\ndimensions of governing highly autonomous + AI systems, \\ndrawing upon more than fifty contemporary academic and \\npolicy + sources. Three core insights emerge. First, current \\ngovernance structures + provide limited coverage of risks \\nlinked to recursive self-improvement + and multi-agent \\ncoordination, with only an estimated 10\u201315% of safety + \\nresearch addressing impacts that arise after deployment. \\nSecond, economic + projections suggest that agentic AI could \\ngenerate between 2.6 and 4.4 + trillion USD in added global \\noutput by 2030, yet automation could replace + \\napproximately 28\u201342% of existing job tasks, making \\nproactive workforce + transition strategies a policy necessity. \\nThird, fragmented regulatory + approaches remain a concern; \\nin the United States, for example, 70\u201375% + of critical \\ninfrastructure is considered vulnerable to adversarial \\nautonomous + systems. To address these issues, we propose a \\ngovernance model built on + three pillars: modular agent \\ndesign, adaptive safety mechanisms, and international + \\ncoordination. Policy measures such as licensing thresholds \\nfor high-computer + systems exceeding 10^25 FLOPs, \\nstructured red-team testing across public + and private \\nsectors, and fiscal incentives for governance-by-design \\npractices + are advanced as actionable pathways. Overall, the \\nstudy argues for adaptive, + globally coordinated governance \\nframeworks that balance innovation with + systemic risk \\nmitigation in the era of agentic AI and AGI. his is a pure + \\nreview paper and all results, proposals and findings are from \\nthe cited + literature. \\nKEYWORDS- AI Governance, Agentic AI, Generative \\nAI, Artificial + General Intelligence (AGI), Ethics, Policy, \\nRisk Management, Recursive + Self-Improvement, Multi\\u0002Agent Systems, Workforce Transition, International + \\nCoordination\\nI. INTRODUCTION\\nThe emergence of agentic artificial intelligence + (AI)\u2014\\nsystems capable of autonomous goal-setting, decision\\u0002making, + and task execution\u2014marks a fundamental \\nparadigm shift in computational + intelligence. Unlike \\nconventional generative AI, which operates within + static \\nprompt\u2013response frameworks, agentic AI demonstrates \\ndynamic + adaptability, recursive self-improvement (RSI), \\nand multi-agent collaboration + [1], [2]. These capabilities \\nenable agentic systems not only to generate + content but also \\nto plan, execute, and optimize tasks with minimal human + \\noversight. Parallel advancements in artificial general \\nintelligence + (AGI) further amplify both the transformative \\npotential and systemic risks + of these technologies. Current \\neconomic projections suggest that agentic + AI could \\ncontribute between $2.6 and $4.4 trillion to global GDP by \\n2030, + while simultaneously automating 28\u201342% of job\\u0002related tasks [3], + [4]. Such forecasts underscore the dual \\nchallenge of harnessing productivity + gains while mitigating \\nwidespread labor market disruptions.\\nDespite growing + awareness, existing governance \\nframeworks remain ill-equipped to manage + the \\ncomplexities of agentic AI and AGI. This paper identifies \\nthree + central gaps in current approaches. First, regulatory \\nand ethical models + provide insufficient guidance for \\naddressing risks associated with recursive + self\\u0002improvement and autonomous multi-agent coordination. \\nSecond, + regulatory fragmentation across jurisdictions \\nhinders coherent oversight; + in the United States alone, \\nestimates indicate that 70\u201375% of critical + infrastructure \\nremains exposed to adversarial exploitation by autonomous + \\nsystems [5]. Third, policy tools remain largely static, \\nlacking the + adaptive mechanisms necessary to keep pace \\nwith rapidly evolving agentic + technologies.\\nTo address these issues, this paper synthesizes insights from + \\nover fifty contemporary sources, including academic \\nliterature (32%), + industry reports (28%), and government \\npublications (20%), with a particular + emphasis on policy \\ndevelopments in 2024\u20132025, such as the European + Union\u2019s \\nAI Act [6] and the United States\u2019 Executive Order 14110\\n[7]. + Our contributions are fourfold:\\n\uF0B7 Conceptual foundations: We review + the terminology \\nand governance challenges of agentic AI and AGI, \\nsupported + by definitional clarity (Table I) and forward\\u0002looking projections (Figure + 1).\\n\uF0B7 Technical governance frameworks: We introduce \\ncompliance models + such as governance scoring (Eq. 1) \\nand RSI optimization methods (Algorithm + 1).\\n\uF0B7 Comparative policy analysis: We evaluate governance \\nstrategies + across jurisdictions and sectors, summarized \\nin Table 4.\\n\uF0B7 Tripartite + governance architecture: We propose an \\nintegrated model combining modular + agent design, \\nInternational Journal of Innovative Research in Computer + Science and Technology (IJIRCST)\\nInnovative Research Publication 15\\nevolutionary + safety algorithms, and international \\ncoordination, illustrated in Figure + 3.\\nBy bridging technical, economic, and policy perspectives, \\nthis study + provides actionable recommendations for \\nstakeholders navigating the agentic + AI era. These include \\nstandardized licensing requirements for high-computer\\nFigure + 1: Future timeline of agentic AI impacts (2025-2035) showing adoption rates + (blue), economic effects (green/orange), \\nand workforce changes (red/purple) + with data source references\\nsystems (>10^25 FLOPs), structured public\u2013private + red\\u0002teaming protocols, and governance-by-design tax \\nincentives. The + concluding section outlines urgent priorities \\nfor research and regulation, + with the aim of developing \\nadaptive, globally coordinated governance frameworks + that \\nbalance innovation with existential risk mitigation.\\nII. LITERATURE + REVIEW\\nThe literature review presented in this study is based on a \\nsystematic + synthesis of more than fifty sources, \\nencompassing academic journals, industry + reports, \\ngovernment publications, conference proceedings, and \\ntrade + media. The objective of this review is to capture the \\nevolving technical, + ethical, and policy dimensions of \\nagentic artificial intelligence (AI) + and artificial general \\nintelligence (AGI) governance across multiple \\nperspectives.\\nAcademic + journals constituted the largest share of reviewed \\nmaterials, accounting + for approximately 32% of the \\nsources. These works were primarily drawn + from peer\\u0002reviewed venues such as IEEE Computer [8], specialized \\noutlets + focusing on robotics, safety, and alignment [4], and \\nprominent policy journals. + The emphasis of this body of \\nwork was on technical architecture (15%), + governance \\nframeworks (10%), and quantitative risk analyses (7%), \\nproviding + the theoretical and conceptual foundation for \\nsubsequent developments.\\nIndustry + contributions represented 28% of the corpus, with \\ninfluential white papers + from technology corporations such \\nas IBM [9] and Deloitte [10], alongside + reports by global \\nthink tanks including the Global Skill Development Council + \\n(GSDC) [5]. These reports concentrated on market \\nprojections (12%), + deployment case studies (9%), and \\nemerging self-regulation standards (7%). + Although industry \\nsources provided pragmatic insights into implementation, + \\nthey also reflected a tendency toward optimism regarding \\nscalability + and adoption.\\nGovernment publications accounted for 20% of the \\nreferences, + with key contributions including analyses of the \\nU.S. Executive Orders + on AI [7], the European Union\u2019s AI \\nAct [6], and the United Nations\u2019 + perspectives on AI \\ngovernance [11]. These materials collectively focused + on \\nregulatory frameworks (14%) and national strategies (6%), \\noffering + a policy-oriented complement to technical studies. \\nIn parallel, conference + proceedings and preprints comprised \\n12% of the literature, particularly + from NeurIPS workshops \\non AI safety and IEEE Symposia addressing AGI \\ngovernance. + These emerging works provided insights into \\nalgorithmic innovations (8%) + and governance prototypes \\n(4%), reflecting the experimental stage of research + in this \\narea. Finally, 8% of the sources were derived from news \\noutlets + and trade media such as TechRadar [12] and \\nMarkTechPost [13]. These contributions + highlighted real\\u0002world deployments (5%) and expert interviews (3%), + \\nserving as a bridge between academic analysis and public \\ndiscourse.\\nIn + terms of temporal distribution, nearly half of the \\nreviewed works (50%) + were published in 2025, reflecting \\nthe intensification of debates surrounding + agentic AI \\ndeployment, risk analyses [18], and U.S.-specific \\nregulatory + actions [19]. A further 32% were concentrated in \\n2024, coinciding with + a surge in ethical frameworks [16] \\nand global policy proposals [11]. The + earlier period of \\n2021\u20132023 accounted for 18% of the references, focusing + \\nlargely on distinctions between generative and agentic AI \\n[14] and the + first wave of governance frameworks [15].\\nGeographically, the review reflects + both national and \\ninternational perspectives. U.S.-centric studies comprised + \\n45% of the dataset, particularly in relation to federal and \\nstate-level + pol\"},{\"id\":\"https://partnershiponai.org/resource/preparing-for-ai-agent-governance/\",\"title\":\"Preparing + for AI Agent Governance\",\"url\":\"https://partnershiponai.org/resource/preparing-for-ai-agent-governance/\",\"publishedDate\":\"2025-09-30T00:00:00.000Z\",\"author\":\"Jacob + Pratt\",\"text\":\"Preparing for AI Agent Governance - Partnership on AI\\n[![logo]![logo]![logo]] + \\n* [] \\n* [DONATE] \\n×\\n[SEARCH] \\nShare\\n* [**] \\n* [**] \\n* + [**] \\n[Our Resources] ##### /\\n##### Report\\nPolicy\\n# Preparing for + AI Agent Governance\\n### A research agenda for policymakers and researchers\\n![] + \\nJacob Pratt\\nSeptember 30, 2025\\n![$hero_image['alt']] \\n### Key Takeaways\\n* + **How AI agents will impact society is still uncertain.**While the body of + scholarly work and publicly available evidence are growing, we don\u2019t + yet know enough about how AI agents will be used or what impacts they may + have. However, AI investments continue to soar, so policymakers should begin + to prepare now.\\n* **Policymakers should prioritize evidence and information + gathering, including through sandboxes and testbeds.**Given this uncertainty, + policymakers should promote activities to generate evidence, rather than advancing + prescriptive regulations. Promising options policymakers can use to build + expertise and track developments are sandboxes and testbeds, which enable + experimentation of new systems under regulatory supervision.\\n* **Subsequent + rule-making will require substantial research from inside and outside of government.**Academia, + civil society, government, and industry should work together to generate evidence + on AI agents’ capabilities, risks, societal impacts, and potential policy + interventions. This will support future policy development.\\n* **This paper + provides a roadmap for this research.**We outline three foundational requirements + for governing AI agents and detail a comprehensive research agenda, including + 12 top-level and 45 sub-level questions, designed to directly support policymakers + in developing evidence-based policy.\\nIndustry leaders have named 2025 the + \u201Cyear of agentic exploration,\u201D foretelling the adoption of systems + that will change how we interact, what jobs we perform, and even how we think. + However, these innovations face significant hurdles: widespread AI agent adoption + has been stifled by persistent reliability and security challenges, giving + policymakers an opportunity to prepare thoughtfully for how to promote the + benefits and protect against the risks of AI agents.\\nMaking the right public + policy decisions on AI governance, including on the institutions, policies, + regulations, and tools that ensure that AI systems operate in the public interest, + requires substantial research and evidence. Though there is significant research + done on AI more broadly and the theoretical impacts of AI agents, we don\u2019t + yet know enough about how AI agents will be used or what impacts they may + have. The key challenge for policymakers is not whether to regulate now, but + how to govern AI agents while their impacts are still uncertain, and understand + what evidence will be needed to make informed decisions when decisive action + is needed.\\n> The key challenge for policymakers is not whether to regulate + now, but how to govern AI agents while their impacts are still uncertain.\\n**In + this paper, we outline a research agenda to guide policymakers and researchers + in preparing for the governance of AI agents.**Structured around three core + technology governance requirements for policymakers – foundational understanding, + impact assessment, and intervention evaluation – we present 12 top-level + questions and 45 detailed open sub-questions from literature and partner discussions + that policymakers and researchers should explore. We also explore how monitoring, + sandboxes, and testbeds can be an important first step for policymakers.\\nPAI + has already started conducting research that supports this agenda, and have + recently published work on[prioritizing real-time failure detection] and[global + governance]. We will continue to build on this work, and look forward to collaborating + with our community to advance AI agent governance.\\nDownload the full research + agenda and be a part of guiding evidence-based AI agent governance now.\\n[Back + to All Resources]\",\"image\":\"https://partnershiponai.org/wp-content/uploads/2025/10/policy-agenda-feature-img-1800x832-1.webp\",\"favicon\":\"https://partnershiponai.org/wp-content/uploads/2021/08/PAI-Favicon.png\"},{\"id\":\"https://arxiv.org/abs/2509.10289\",\"title\":\"We + Need a New Ethics for a World of AI Agents\",\"url\":\"https://arxiv.org/abs/2509.10289\",\"publishedDate\":\"2025-09-12T00:00:00.000Z\",\"author\":\"[Submitted + on 12 Sep 2025]\",\"text\":\"[2509.10289] We Need a New Ethics for a World + of AI Agents\\n[Skip to main content] \\n[![Cornell University]] \\nWe gratefully + acknowledge support from the Simons Foundation,[member institutions], and + all contributors.[Donate] \\n[] \\n[![arxiv logo]] >[cs] >arXiv:2509.10289\\n[Help] + |[Advanced Search] \\nAll fieldsTitleAuthorAbstractCommentsJournal referenceACM + classificationMSC classificationReport numberarXiv identifierDOIORCIDarXiv + author IDHelp pagesFull text\\nSearch\\n[![arXiv logo]] \\n[![Cornell University + Logo]] \\nopen search\\nGO\\nopen navigation menu\\n# Computer Science \\\\> + Computers and Society\\n**arXiv:2509.10289**(cs)\\n[Submitted on 12 Sep 2025]\\n# + Title:We Need a New Ethics for a World of AI Agents\\nAuthors:[Iason Gabriel],[Geoff + Keeling],[Arianna Manzini],[James Evans] \\nView a PDF of the paper titled + We Need a New Ethics for a World of AI Agents, by Iason Gabriel and 3 other + authors\\n[View PDF] > > Abstract:\\n> The deployment of capable AI agents + raises fresh questions about safety, human-machine relationships and social + coordination. We argue for greater engagement by scientists, scholars, engineers + and policymakers with the implications of a world increasingly populated by + AI agents. We explore key challenges that must be addressed to ensure that + interactions between humans and agents, and among agents themselves, remain + broadly beneficial. Comments:|6 pages, no figures|\\nSubjects:|Computers and + Society (cs.CY); Artificial Intelligence (cs.AI)|\\nACMclasses:|I.2.0; K.4.1|\\nCite + as:|[arXiv:2509.10289] [cs.CY]|\\n|(or[arXiv:2509.10289v1] [cs.CY]for this + version)|\\n|[https://doi.org/10.48550/arXiv.2509.10289] \\nFocus to learn + more\\narXiv-issued DOI via DataCite\\n|\\nJournalreference:|Nature, 644 (8075), + 2025, 38-40|\\nRelated DOI:|[https://doi.org/10.1038/d41586-025-02454-5] \\nFocus + to learn more\\nDOI(s) linking to related resources\\n|\\n## Submission history\\nFrom: + Iason Gabriel [[view email]]\\n**[v1]**Fri, 12 Sep 2025 14:29:14 UTC (162 + KB)\\nFull-text links:## Access Paper:\\nView a PDF of the paper titled We + Need a New Ethics for a World of AI Agents, by Iason Gabriel and 3 other authors\\n* + [View PDF] \\n[view license] \\nCurrent browse context:\\ncs.CY\\n[<<prev] + | [next>>] \\n[new] |[recent] |[2025-09] \\nChange to browse by:\\n[cs] + \\n[cs.AI] \\n### References & Citations\\n* [NASA ADS] \\n* [Google Scholar] + \\n* [Semantic Scholar] \\nexport BibTeX citationLoading...\\n## BibTeX formatted + citation\\n×\\nloading...\\nData provided by:\\n### Bookmark\\n[![BibSonomy + logo]] [![Reddit logo]] \\nBibliographic Tools\\n# Bibliographic and Citation + Tools\\nBibliographic Explorer Toggle\\nBibliographic Explorer*([What is the + Explorer?])*\\nConnected Papers Toggle\\nConnected Papers*([What is Connected + Papers?])*\\nLitmaps Toggle\\nLitmaps*([What is Litmaps?])*\\nscite.ai Toggle\\nscite + Smart Citations*([What are Smart Citations?])*\\nCode, Data, Media\\n# Code, + Data and Media Associated with this Article\\nalphaXiv Toggle\\nalphaXiv*([What + is alphaXiv?])*\\nLinks to Code Toggle\\nCatalyzeX Code Finder for Papers*([What + is CatalyzeX?])*\\nDagsHub Toggle\\nDagsHub*([What is DagsHub?])*\\nGotitPub + Toggle\\nGotit.pub*([What is GotitPub?])*\\nHuggingface Toggle\\nHugging Face*([What + is Huggingface?])*\\nLinks to Code Toggle\\nPapers with Code*([What is Papers + with Code?])*\\nScienceCast Toggle\\nScienceCast*([What is ScienceCast?])*\\nDemos\\n# + Demos\\nReplicate Toggle\\nReplicate*([What is Replicate?])*\\nSpaces Toggle\\nHugging + Face Spaces*([What is Spaces?])*\\nSpaces Toggle\\nTXYZ.AI*([What is TXYZ.AI?])*\\nRelated + Papers\\n# Recommenders and Search Tools\\nLink to Influence Flower\\nInfluence + Flower*([What are Influence Flowers?])*\\nCore recommender toggle\\nCORE Recommender*([What + is CORE?])*\\n* Author\\n* Venue\\n* Institution\\n* Topic\\nAbout arXivLabs\\n# + arXivLabs: experimental projects with community collaborators\\narXivLabs + is a framework that allows collaborators to develop and share new arXiv features + directly on our website.\\nBoth individuals and organizations that work with + arXivLabs have embraced and accepted our values of openness, community, excellence, + and user data privacy. arXiv is committed to these values and only works with + partners that adhere to them.\\nHave an idea for a project that will add value + for arXiv's community?[**Learn more about arXivLabs**].\\n[Which authors of + this paper are endorsers?] |[Disable MathJax] ([What is MathJax?])\",\"image\":\"/static/browse/0.3.4/images/arxiv-logo-fb.png\",\"favicon\":\"https://arxiv.org/static/browse/0.3.4/images/icons/favicon-32x32.png\"},{\"id\":\"https://arxiv.org/abs/2503.04750\",\"title\":\"AI + Agents Should be Regulated Based on the Extent of Their Autonomous Operations\",\"url\":\"https://arxiv.org/abs/2503.04750\",\"publishedDate\":\"2025-02-07T00:00:00.000Z\",\"author\":\"[Submitted + on 7 Feb 2025 (v1), last revised 26 May 2025 (this version, v2)]\",\"text\":\"[2503.04750] + AI Agents Should be Regulated Based on the Extent of Their Autonomous Operations\\n[Skip + to main content] \\n[![Cornell University]] \\nWe gratefully acknowledge support + from the Simons Foundation,[member institutions], and all contributors.[Donate] + \\n[] \\n[![arxiv logo]] >[cs] >arXiv:2503.04750\\n[Help] |[Advanced + Search] \\nAll fieldsTitleAuthorAbstractCommentsJournal referenceACM classificationMSC + classificationReport numberarXiv identifierDOIORCIDarXiv author IDHelp pagesFull + text\\nSearch\\n[![arXiv logo]] \\n[![Cornell University Logo]] \\nopen search\\nGO\\nopen + navigation menu\\n# Computer Science \\\\> Computers and Society\\n**arXiv:2503.04750**(cs)\\n[Submitted + on 7 Feb 2025 ([v1]), last revised 26 May 2025 (this version, v2)]\\n# Title:AI + Agents Should be Regulated Based on the Extent of Their Autonomous Operations\\nAuthors:[Takayuki + Osogami] \\nView a PDF of the paper titled AI Agents Should be Regulated Based + on the Extent of Their Autonomous Operations, by Takayuki Osogami\\n[View + PDF] [HTML (experimental)] > > Abstract:\\n> This position paper argues that + AI agents should be regulated by the extent to which they operate autonomously. + AI agents with long-term planning and strategic capabilities can pose significant + risks of human extinction and irreversible global catastrophes. While existing + regulations often focus on computational scale as a proxy for potential harm, + we argue that such measures are insufficient for assessing the risks posed + by agents whose capabilities arise primarily from inference-time computation. + To support our position, we discuss relevant regulations and recommendations + from scientists regarding existential risks, as well as the advantages of + using action sequences -- which reflect the degree of an agent's autonomy + -- as a more suitable measure of potential impact than existing metrics that + rely on observing environmental states. Comments:|Updated to improve the overall + clarity with additional recent references; 22 pages, 2 figures|\\nSubjects:|Computers + and Society (cs.CY); Artificial Intelligence (cs.AI)|\\nCite as:|[arXiv:2503.04750] + [cs.CY]|\\n|(or[arXiv:2503.04750v2] [cs.CY]for this version)|\\n|[https://doi.org/10.48550/arXiv.2503.04750] + \\nFocus to learn more\\narXiv-issued DOI via DataCite\\n|\\n## Submission + history\\nFrom: Takayuki Osogami Ph.D. [[view email]]\\n**[[v1]] **Fri, 7 + Feb 2025 09:40:48 UTC (166 KB)\\n**[v2]**Mon, 26 May 2025 08:41:12 UTC (133 + KB)\\nFull-text links:## Access Paper:\\nView a PDF of the paper titled AI + Agents Should be Regulated Based on the Extent of Their Autonomous Operations, + by Takayuki Osogami\\n* [View PDF] \\n* [HTML (experimental)] \\n* [TeX Source] + \\n[view license] \\nCurrent browse context:\\ncs.CY\\n[<<prev] | [next>>] + \\n[new] |[recent] |[2025-03] \\nChange to browse by:\\n[cs] \\n[cs.AI] \\n### + References & Citations\\n* [NASA ADS] \\n* [Google Scholar] \\n* [Semantic + Scholar] \\nexport BibTeX citationLoading...\\n## BibTeX formatted citation\\n×\\nloading...\\nData + provided by:\\n### Bookmark\\n[![BibSonomy logo]] [![Reddit logo]] \\nBibliographic + Tools\\n# Bibliographic and Citation Tools\\nBibliographic Explorer Toggle\\nBibliographic + Explorer*([What is the Explorer?])*\\nConnected Papers Toggle\\nConnected + Papers*([What is Connected Papers?])*\\nLitmaps Toggle\\nLitmaps*([What is + Litmaps?])*\\nscite.ai Toggle\\nscite Smart Citations*([What are Smart Citations?])*\\nCode, + Data, Media\\n# Code, Data and Media Associated with this Article\\nalphaXiv + Toggle\\nalphaXiv*([What is alphaXiv?])*\\nLinks to Code Toggle\\nCatalyzeX + Code Finder for Papers*([What is CatalyzeX?])*\\nDagsHub Toggle\\nDagsHub*([What + is DagsHub?])*\\nGotitPub Toggle\\nGotit.pub*([What is GotitPub?])*\\nHuggingface + Toggle\\nHugging Face*([What is Huggingface?])*\\nLinks to Code Toggle\\nPapers + with Code*([What is Papers with Code?])*\\nScienceCast Toggle\\nScienceCast*([What + is ScienceCast?])*\\nDemos\\n# Demos\\nReplicate Toggle\\nReplicate*([What + is Replicate?])*\\nSpaces Toggle\\nHugging Face Spaces*([What is Spaces?])*\\nSpaces + Toggle\\nTXYZ.AI*([What is TXYZ.AI?])*\\nRelated Papers\\n# Recommenders and + Search Tools\\nLink to Influence Flower\\nInfluence Flower*([What are Influence + Flowers?])*\\nCore recommender toggle\\nCORE Recommender*([What is CORE?])*\\n* + Author\\n* Venue\\n* Institution\\n* Topic\\nAbout arXivLabs\\n# arXivLabs: + experimental projects with community collaborators\\narXivLabs is a framework + that allows collaborators to develop and share new arXiv features directly + on our website.\\nBoth individuals and organizations that work with arXivLabs + have embraced and accepted our values of openness, community, excellence, + and user data privacy. arXiv is committed to these values and only works with + partners that adhere to them.\\nHave an idea for a project that will add value + for arXiv's community?[**Learn more about arXivLabs**].\\n[Which authors of + this paper are endorsers?] |[Disable MathJax] ([What is MathJax?])\",\"image\":\"/static/browse/0.3.4/images/arxiv-logo-fb.png\",\"favicon\":\"https://arxiv.org/static/browse/0.3.4/images/icons/favicon-32x32.png\"},{\"id\":\"https://arxiv.org/abs/2505.21808\",\"title\":\"AI + Agent Governance: A Field Guide\",\"url\":\"https://arxiv.org/abs/2505.21808\",\"publishedDate\":\"2025-05-27T00:00:00.000Z\",\"author\":\"[Submitted + on 27 May 2025]\",\"text\":\"[2505.21808] AI Agent Governance: A Field Guide\\n[Skip + to main content] \\n[![Cornell University]] \\nWe gratefully acknowledge support + from the Simons Foundation,[member institutions], and all contributors.[Donate] + \\n[] \\n[![arxiv logo]] >[cs] >arXiv:2505.21808\\n[Help] |[Advanced + Search] \\nAll fieldsTitleAuthorAbstractCommentsJournal referenceACM classificationMSC + classificationReport numberarXiv identifierDOIORCIDarXiv author IDHelp pagesFull + text\\nSearch\\n[![arXiv logo]] \\n[![Cornell University Logo]] \\nopen search\\nGO\\nopen + navigation menu\\n# Computer Science \\\\> Computers and Society\\n**arXiv:2505.21808**(cs)\\n[Submitted + on 27 May 2025]\\n# Title:AI Agent Governance: A Field Guide\\nAuthors:[Jam + Kraprayoon],[Zoe Williams],[Rida Fayyaz] \\nView a PDF of the paper titled + AI Agent Governance: A Field Guide, by Jam Kraprayoon and 2 other authors\\n[View + PDF] > > Abstract:\\n> This report serves as an accessible guide to the emerging + field of AI agent governance. Agents - AI systems that can autonomously achieve + goals in the world, with little to no explicit human instruction about how + to do so - are a major focus of leading tech companies, AI start-ups, and + investors. If these development efforts are successful, some industry leaders + claim we could soon see a world where millions or billions of agents autonomously + perform complex tasks across society. Society is largely unprepared for this + development. A future where capable agents are deployed en masse could see + transformative benefits to society but also profound and novel risks. Currently, + the exploration of agent governance questions and the development of associated + interventions remain in their infancy. Only a few researchers, primarily in + civil society organizations, public research institutes, and frontier AI companies, + are actively working on these challenges. Subjects:|Computers and Society + (cs.CY)|\\nCite as:|[arXiv:2505.21808] [cs.CY]|\\n|(or[arXiv:2505.21808v1] + [cs.CY]for this version)|\\n|[https://doi.org/10.48550/arXiv.2505.21808] \\nFocus + to learn more\\narXiv-issued DOI via DataCite\\n|\\n## Submission history\\nFrom: + Jam Kraprayoon Mr. [[view email]]\\n**[v1]**Tue, 27 May 2025 22:26:51 UTC + (2,346 KB)\\nFull-text links:## Access Paper:\\nView a PDF of the paper titled + AI Agent Governance: A Field Guide, by Jam Kraprayoon and 2 other authors\\n* + [View PDF] \\n[![license icon] view license] \\nCurrent browse context:\\ncs.CY\\n[<<prev] + | [next>>] \\n[new] |[recent] |[2025-05] \\nChange to browse by:\\n[cs] + \\n### References & Citations\\n* [NASA ADS] \\n* [Google Scholar] \\n* + [Semantic Scholar] \\nexport BibTeX citationLoading...\\n## BibTeX formatted + citation\\n×\\nloading...\\nData provided by:\\n### Bookmark\\n[![BibSonomy + logo]] [![Reddit logo]] \\nBibliographic Tools\\n# Bibliographic and Citation + Tools\\nBibliographic Explorer Toggle\\nBibliographic Explorer*([What is the + Explorer?])*\\nConnected Papers Toggle\\nConnected Papers*([What is Connected + Papers?])*\\nLitmaps Toggle\\nLitmaps*([What is Litmaps?])*\\nscite.ai Toggle\\nscite + Smart Citations*([What are Smart Citations?])*\\nCode, Data, Media\\n# Code, + Data and Media Associated with this Article\\nalphaXiv Toggle\\nalphaXiv*([What + is alphaXiv?])*\\nLinks to Code Toggle\\nCatalyzeX Code Finder for Papers*([What + is CatalyzeX?])*\\nDagsHub Toggle\\nDagsHub*([What is DagsHub?])*\\nGotitPub + Toggle\\nGotit.pub*([What is GotitPub?])*\\nHuggingface Toggle\\nHugging Face*([What + is Huggingface?])*\\nLinks to Code Toggle\\nPapers with Code*([What is Papers + with Code?])*\\nScienceCast Toggle\\nScienceCast*([What is ScienceCast?])*\\nDemos\\n# + Demos\\nReplicate Toggle\\nReplicate*([What is Replicate?])*\\nSpaces Toggle\\nHugging + Face Spaces*([What is Spaces?])*\\nSpaces Toggle\\nTXYZ.AI*([What is TXYZ.AI?])*\\nRelated + Papers\\n# Recommenders and Search Tools\\nLink to Influence Flower\\nInfluence + Flower*([What are Influence Flowers?])*\\nCore recommender toggle\\nCORE Recommender*([What + is CORE?])*\\n* Author\\n* Venue\\n* Institution\\n* Topic\\nAbout arXivLabs\\n# + arXivLabs: experimental projects with community collaborators\\narXivLabs + is a framework that allows collaborators to develop and share new arXiv features + directly on our website.\\nBoth individuals and organizations that work with + arXivLabs have embraced and accepted our values of openness, community, excellence, + and user data privacy. arXiv is committed to these values and only works with + partners that adhere to them.\\nHave an idea for a project that will add value + for arXiv's community?[**Learn more about arXivLabs**].\\n[Which authors of + this paper are endorsers?] |[Disable MathJax] ([What is MathJax?])\",\"image\":\"/static/browse/0.3.4/images/arxiv-logo-fb.png\",\"favicon\":\"https://arxiv.org/static/browse/0.3.4/images/icons/favicon-32x32.png\"},{\"id\":\"https://medium.com/@enrico.papalini/ai-agents-race-in-july-2025-77c1cb711f14\",\"title\":\"AI + Agents Race in July 2025\",\"url\":\"https://medium.com/@enrico.papalini/ai-agents-race-in-july-2025-77c1cb711f14\",\"author\":\"Enrico + Papalini\",\"text\":\"AI Agents Race in July 2025. Cutting Through the Hype + to Uncover\u2026 | by Enrico Papalini | Medium\\n[Sitemap] \\n[Open in app] + \\nSign up\\n[Sign in] \\n[Medium Logo] \\n[\\nWrite\\n] \\n[\\nSearch\\n] + \\nSign up\\n[Sign in] \\n![] \\n# **AI Agents Race in July 2025**\\n## **Cutting + Through the Hype to Uncover Real Impact and Returns**\\n[\\n![Enrico Papalini] + \\n] \\n[Enrico Papalini] \\n38 min read\\n\xB7Jul 21, 2025\\n[\\n] \\n--\\n[] + \\nListen\\nShare\\nPress enter or click to view image in full size\\n![] + \\nImage generated by Bing## Executive Summary\\nThe discourse surrounding + Artificial Intelligence (AI) Agentic systems in July 2025 is rife with both + immense promise and considerable misunderstanding. A critical distinction + has solidified: on one hand, there are sophisticated AI-powered tools that + automate predefined workflows, often mislabeled as \u201Cagents\u201D; on + the other, there are truly autonomous AI agents capable of independent decision-making + and goal execution. While the former offers significant, achievable efficiency + gains, the latter, though rapidly advancing, remains in a \u201Cresearch preview\u201D + phase for high-stakes applications, necessitating a pragmatic approach to + adoption.\\nLeading technology players like OpenAI, Google DeepMind, and Meta + are actively deploying and integrating autonomous agent capabilities. These + agents demonstrate impressive abilities, from automating complex web tasks + and office chores to orchestrating multi-step personal projects and generating + advertising content. However, their current limitations are clear: they struggle + with sensitive, high-stakes tasks, complex interfaces, and certain web access + restrictions, often requiring human oversight for critical actions.\\nEnterprise + adoption is accelerating, with a substantial majority of executives already + deploying or planning to deploy AI agents, driven by the perception of significant + competitive advantage. This adoption often follows a pragmatic, risk-mitigated + approach, starting with rule-based processes and incorporating robust human-in-the-loop + oversight. The regulatory environment, particularly the EU AI Act, is shaping + deployment strategies, emphasizing data governance, transparency, and accountability, + which are critical for building trust. Despite the hype, confidence in fully + autonomous agents has declined, highlighting a \u201Ctrust gap\u201D that + organizations must address through ethical governance and explainability.\\nThe + realistic return on investment (ROI) for AI agents extends beyond immediate + cost reductions to encompass substantial gains in operational efficiency, + productivity, enhanced customer experience, and accelerated innovation. While + quantifying ROI can be complex, early adopters are reporting measurable value + and a significant competitive edge. The path forward for AI Agentic involves + a continued focus on addressing technical limitations, navigating ethical + complexities, and fostering a collaborative human-AI operational model to + unlock its full transformative potential, moving beyond aspirational \u201Cfull + autonomy\u201D to achievable, impactful augmentation.\\n## 1. Demystifying + AI Agentic: What Are True Autonomous Agents?\\nThe term \u201CAI Agentic\u201D + has become a prevalent buzzword, often leading to confusion regarding the + actual capabilities of various AI systems. As of July 2025, a clear definition + has emerged, distinguishing between AI-powered tools that enhance existing + processes and genuinely autonomous AI agents that operate with independent + decision-making authority. Understanding this distinction is crucial for strategic + planning and avoiding the \u201Cinnovation theatre\u201D that can fuel unrealistic + expectations [30].\\n### 1.1 Defining AI Agents: From LLMs to Autonomous Decision-Making\\nThe + fundamental concept of AI Agentic, as understood in July 2025, centers on*autonomous + agents*. These are systems designed to receive a high-level goal and then + independently execute multiple operations, including making choices and potentially + initiating actions such as payments [30]. This capability represents a significant + advancement beyond traditional AI tools. For instance, OpenAI describes its + agents, including the newly integrated ChatGPT Agent (formerly Operator), + as \u201CAIs capable of doing work for you independently \u2014you give it + a task and it will execute it\u201D [1]. This directly aligns with the principle + of autonomous execution.\\nPress enter or click to view image in full size\\n![] + \\nThe Agentic mode in ChatGPT\\nLarge Language Models (LLMs) serve as foundational + components, acting as the \u201Cbrains\u201D or \u201Creasoning engines\u201D + that enable these autonomous solutions. However, it is important to recognize + that LLMs themselves are not the agents. The evolution of AI has seen LLMs + overcome their inherent limitations, such as poor calculation abilities or + fixed training data, by gaining the capacity to interface with external tools. + This includes calling a calculator for complex mathematical operations, retrieving + information from external document collections via architectures like Retrieval-Augmented + Generation (RAG), connecting to the internet, or interfacing with email and + notification systems [30]. This integration of tools was a pivotal step towards + agentic behavior.\\nThe transition from LLMs as standalone processing units + to LLMs functioning as enablers within larger, orchestrated autonomous systems + signifies a fundamental architectural evolution in AI. This shift moves the + primary focus of innovation from merely scaling model size to designing sophisticated + system architectures and enhancing reasoning capabilities. For businesses, + this implies that successful AI agent adoption will require more than simply + acquiring access to the latest LLM. It necessitates a strategic emphasis on + integrating these models into robust, tool-augmented architectures and developing + a deep understanding of the system-level design required for truly autonomous + operation. The future gains in AI are increasingly expected to come from \u201Csmarter + algorithms\u201D and architectural innovation, rather than solely from building + bigger models [2].\\nPress enter or click to view image in full size\\n![] + \\nMulti agents interaction \u2014source[https://blog.langchain.com/langgraph-multi-agent-workflows/] + \\nThe rapid proliferation of the \u201CAI Agentic\u201D term, often misapplied + to less autonomous systems, has created considerable market excitement and + a sense of \u201Cfear of missing out\u201D (FOMO) among top management [30]. + This dynamic necessitates a clear, grounded definition for effective strategic + decision-making. The competitive pressures in the AI market incentivize companies + to rebrand existing automation or LLM-powered tools as \u201Cagents\u201D + to capitalize on the hype. This marketing-driven terminology can obscure the + true capabilities for non-technical executives. Consequently, a critical component + of any organization\u2019s AI strategy must involve thorough due diligence + and an internal understanding of the actual functionalities of any \u201Cagent\u201D + solution. Without this clarity, organizations risk misallocating resources, + setting unrealistic expectations, and experiencing disillusionment, which + could undermine broader AI adoption efforts.\\n### 1.2 The Critical Distinction: + Programmed Tools vs. Autonomous Agents\\nA clear demarcation exists between + programmed AI tools and truly autonomous AI agents, a distinction articulated + through a software versioning analogy in the initial discussion.\\n**Programmed + Agents (Version 2.1.0): The Achievable Automation**\\nThese systems are fundamentally + \u201CRPA on steroids or RPA powered by GenAI\u201D [30]. They integrate LLMs + with other LLMs or broader IT systems but operate with \u201Czero freedom + of decision\u201D [30]. Their execution is \u201Cguided and controlled,\u201D + strictly following \u201Cpredetermined tracks\u201D [30]. Frameworks such + as LangChain and LangGraph are instrumental in developing these more complex, + multi-step workflows. Typical applications include automated document processing, + email sorting, meeting summaries, and workflow management, where tasks are + specific and circumscribed [4]. These solutions are \u201Cextraordinary and + can truly help and improve many processes\u201D [30], delivering significant, + measurable value within established operational frameworks.\\n**Autonomous + Agents (Version 4.0.0): The Aspirational Frontier**\\nThese represent the + genuine \u201CAgentic AI\u201D under discussion in 2025, characterized by + true \u201Cdecision-making autonomy\u201D [30]. The user provides a high-level + goal, such as \u201Cplan a weekend trip,\u201D and the agent independently + performs all necessary operations, including navigating the internet, making + choices, and potentially initiating payments [30]. OpenAI\u2019s ChatGPT Agent, + launched on July 17, 2025, is presented as a leading example of this paradigm + shift.1 Other key players include Google DeepMind, with AlphaEvolve optimizing + algorithms and Project Mariner automating web tasks, and Anthropic, which + develops Claude agents for complex problem-solving [5].\\nThe versioning analogy + used to illustrate this evolution is as follows from [30]:\\n* **Version 2.1.0:**Represents + programmed agents, where individual LLMs are guided to perform specific, circumscribed + tasks.\\n* **Version 3.0.0:**Marks the emergence of reasoning engines, such + as OpenAI\u2019s o1 (launched September 2024), which natively integrate complex + architectures to increasingly approximate human-like reasoning.\\n* **Version + 3.1.0:**Encompasses open-source advancements in reasoning, notably accelerated + by projects like DeepSeek\u2019s R1.\\n* **Version 4.0.0:**Denotes truly autonomous + agents that not only have access to tools but \u201Cautonomously choose which + to use and which decisions to take\u201D .\\nThe evolution from programmed + tools to autonomous agents signifies a fundamental shift from the automation + of*steps*to the automation of*goals*. This fundamentally alters how businesses + can approach process optimization and problem-solving. Programmed agents are + designed\",\"image\":\"https://miro.medium.com/v2/resize:fit:1024/1*IQ_4v80JCQesyRnZVscQxQ.png\",\"favicon\":\"https://miro.medium.com/v2/5d8de952517e8160e40ef9841c781cdc14a5db313057fa3c3de41c6f5b494b19\"}],\"searchTime\":1276.4,\"costDollars\":{\"total\":0.015,\"search\":{\"neural\":0.005},\"contents\":{\"text\":0.01}}}" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json; charset=utf-8 + Date: + - Fri, 06 Feb 2026 21:28:32 GMT + Nel: + - '{"report_to":"cf-nel","success_fraction":0.0,"max_age":604800}' + Report-To: + - '{"group":"cf-nel","max_age":604800,"endpoints":[{"url":"https://a.nel.cloudflare.com/report/v4?s=iIheHbF2zor8tOArj6xbZz2V79X%2FXG46R%2BKbfuONsV4vJTF63mxkYKU%2Fsglk0UUD%2BATju8wdifzOFUrNWJppUEWo5R0FEq3Mjw%3D%3D"}]}' + Server: + - cloudflare + Transfer-Encoding: + - chunked + access-control-allow-credentials: + - 'true' + cf-cache-status: + - DYNAMIC + content-security-policy: + - CSP-FILTERED + cross-origin-opener-policy: + - same-origin + cross-origin-resource-policy: + - same-origin + etag: + - ETAG-XXX + origin-agent-cluster: + - ?1 + referrer-policy: + - REFERRER-POLICY-XXX + strict-transport-security: + - STS-XXX + vary: + - Origin + x-content-type-options: + - X-CONTENT-TYPE-XXX + x-dns-prefetch-control: + - 'off' + x-download-options: + - noopen + x-frame-options: + - X-FRAME-OPTIONS-XXX + x-permitted-cross-domain-policies: + - X-PERMITTED-XXX + x-ratelimit-limit: + - '450' + x-ratelimit-remaining: + - '437' + x-ratelimit-reset: + - '1770413312' + x-xss-protection: + - X-XSS-PROTECTION-XXX + status: + code: 200 + message: OK +- request: + body: "{\"messages\":[{\"role\":\"system\",\"content\":\"You are Research Analyst. + You are a research analyst who searches the web for information, identifies + key findings, and produces structured research summaries.\\n\\nYour goal: Research + topics using search tools and produce structured summaries\\n\\nYou are executing + a specific step in a multi-step plan. Focus ONLY on completing\\nthe current + step. Do not plan ahead or worry about future steps.\\n\\nBefore acting, briefly + reason about what you need to do and which approach\\nor tool would be most + helpful for this specific step.\"},{\"role\":\"user\",\"content\":\"## Current + Step\\nSummarize the key findings from the research conducted on autonomous + AI agents, highlighting their potential to enhance productivity and operational + efficiency while addressing emerging challenges related to ethical governance + and reliability.\\n\\n## Context from previous steps:\\nStep 1 result: ### Recent + Developments in Autonomous AI Agents (2025)\\n\\nThis summary compiles recent + findings on the developments in autonomous AI agents for the year 2025, highlighting + their evolution, applications, challenges, and implications across various industries.\\n\\n#### + Key Articles and Insights:\\n\\n1. **AI Agents in 2025: Expectations vs. Reality** + \ \\n - **Source**: [IBM](https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality) + \ \\n - **Published**: February 26, 2025 \\n - **Summary**: The article + discusses the transition from large language models (LLMs) to agentic AI, emphasizing + autonomous agents' potential to reshape work dynamics. Experts from IBM note + that while there is enthusiasm about AI agents, many current applications are + still in their infancy, requiring refinement in planning and reasoning capabilities + to reach true autonomy.\\n\\n2. **Autonomous AI Agents Are Reshaping the Business + Landscape** \\n - **Source**: [KPMG](https://kpmg.com/sk/en/insights/2025/06/autonomous-ai-agents-reshape-business-landscape.html) + \ \\n - **Published**: June 25, 2025 \\n - **Summary**: This report outlines + how agentic AI is transitioning from theory into practical use, impacting sectors + like finance and healthcare. It describes agentic AI as a \\\"digital workforce\\\" + that enhances operational efficiency and decision-making processes through autonomous + task management. \\n\\n3. **The Rise of Autonomous AI Agents: Automating Complex + Tasks** \\n - **Source**: [ResearchGate](https://www.researchgate.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks) + \ \\n - **Published**: July 11, 2025 \\n - **Summary**: This academic paper + explores the technological foundations and applications of autonomous agents + in various fields, noting their ability to autonomously resolve multi-step tasks. + It addresses the ethical considerations linked to the increasing autonomy of + these systems.\\n\\n4. **Developments in AI Agents: Q1 2025 Landscape Analysis** + \ \\n - **Source**: [The Science of Machine Learning & AI](https://www.ml-science.com/blog/2025/4/17/developments-in-ai-agents-q1-2025-landscape-analysis) + \ \\n - **Published**: April 17, 2025 \\n - **Summary**: Highlights significant + advancements in AI agent technology, including their capability for autonomous + task achievement. The article underscores ongoing challenges related to reliability, + safety, and ethical governance in AI performance.\\n\\n5. **Agentic AI in Software + Development** \\n - **Source**: [SculptSoft](https://www.sculptsoft.com/agentic-ai-in-action-how-autonomous-ai-agents-are-changing-software-development-in-2025/) + \ \\n - **Published**: October 15, 2025 \\n - **Summary**: Discusses how + autonomous AI agents are revolutionizing software development by automating + the development lifecycle\u2014planning, coding, testing, and deployment\u2014thus + enabling teams to innovate faster while reducing manual workloads.\\n\\n6. **AI + Agents Arrived in 2025 \u2013 Here\u2019s What Happened and the Challenges Ahead** + \ \\n - **Source**: [The Conversation](https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325) + \ \\n - **Published**: December 29, 2025 \\n - **Summary**: The article + reviews the major milestones reached in 2025, including the introduction of + protocols to enable agents to communicate and use external tools. It also addresses + the societal implications, such as the need for enhanced governance and ethical + considerations in AI deployment.\\n\\n7. **2025 Buildings Show: AI Agents Become + 'Talk of the Town'** \\n - **Source**: [ConstructConnect](https://news.constructconnect.com/2025-buildings-show-ai-agents-become-talk-of-the-town) + \ \\n - **Published**: December 11, 2025 \\n - **Summary**: At the Buildings + Show, digital innovation in construction technology showcased agentic AI's capability + to monitor and manage project tasks autonomously, streamlining operations in + the sector.\\n\\n### Conclusion\\nThe advancements in autonomous AI agents in + 2025 exhibit a shift towards integrating AI into real-world applications, enhancing + productivity across various industries. While there is significant potential + for these agents to improve efficiency and decision-making, challenges related + to ethical considerations, governance, and reliability remain pivotal in guiding + the future of AI development.\\n\\nComplete this step and provide your result.\"},{\"role\":\"assistant\",\"content\":null,\"tool_calls\":[{\"id\":\"call_eMVIqV0hW7fOCNVzPGvarvNv\",\"type\":\"function\",\"function\":{\"name\":\"exa_search_tool\",\"arguments\":\"{\\\"search_query\\\": + \\\"autonomous AI agents productivity operational efficiency 2025\\\", \\\"start_published_date\\\": + \\\"2025-01-01\\\", \\\"end_published_date\\\": \\\"2025-12-31\\\", \\\"include_domains\\\": + null}\"}},{\"id\":\"call_QoeOUYfW09KlW4WYa0jHps9N\",\"type\":\"function\",\"function\":{\"name\":\"exa_search_tool\",\"arguments\":\"{\\\"search_query\\\": + \\\"autonomous AI agents ethical governance reliability challenges 2025\\\", + \\\"start_published_date\\\": \\\"2025-01-01\\\", \\\"end_published_date\\\": + \\\"2025-12-31\\\", \\\"include_domains\\\": null}\"}}]},{\"role\":\"tool\",\"tool_call_id\":\"call_eMVIqV0hW7fOCNVzPGvarvNv\",\"name\":\"exa_search_tool\",\"content\":\"Title: + PwC's AI Agent Survey\\nURL: https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-agent-survey.html\\nID: + https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-agent-survey.html\\nScore: + None\\nPublished Date: 2025-05-16T00:00:00.000Z\\nAuthor: PricewaterhouseCoopers\\nImage: + https://www.pwc.com/us/en/tech-effect/content/images/flagship/ai/ai-agents-survey.png\\nFavicon: + https://www.pwc.com/etc.clientlibs/pwc/clientlibs/css_common/resources/image/favicon.ico\\nExtras: + None\\nSubpages: None\\nText: AI agent survey: PwC\\n[] \\nSearch\\nMenu\\n[Featured + insights] \\n[Featured insights] \\nMenu\\n[Featured insights] \\n[2026 Global + Digital Trust Insights Survey] \\nMenu\\n[Featured insights] \\n[Board governance + issues] \\nMenu\\n[Featured insights] \\n[C-suite insights] \\nMenu\\n[Featured + insights] \\n[Case studies] \\nMenu\\n[Featured insights] \\n[Policy on Demand] + \\nMenu\\n[Featured insights] \\n[Podcasts] \\nMenu\\n[Featured insights] \\n[PwC + Executive Pulse] \\nMenu\\n[Featured insights] \\n[Tech Effect] \\nMenu\\n[Featured + insights] \\n[Viewpoint] \\nMenu\\n[Featured insights] \\n[Webcasts] \\nMenu\\n[Featured + insights] \\n[All Research and insights] \\nFeatured\\n[![] \\nAmerica in motion\\n] + \\n[![] \\nExecutive leadership hub - What\u2019s important to the C-suite?\\n] + \\nMenu\\n[Capabilities] \\n[Capabilities] \\nMenu\\n[Capabilities] \\n[Audit + and Assurance] \\nMenu\\n[Capabilities] \\n[Alliances and ecosystems] \\nMenu\\n[Capabilities] + \\n[Artificial Intelligence (AI)] \\nMenu\\n[Capabilities] \\n[Board governance + issues] \\nMenu\\n[Capabilities] \\n[Consulting] \\nMenu\\n[Capabilities] \\n[Cybersecurity, + Risk and Regulatory] \\nMenu\\n[Capabilities] \\n[Deals] \\nMenu\\n[Capabilities] + \\n[Digital assets and crypto] \\nMenu\\n[Capabilities] \\n[Digital assurance + and transparency] \\nMenu\\n[Capabilities] \\n[Engineering & AI] \\nMenu\\n[Capabilities] + \\n[Enterprise Strategy] \\nMenu\\n[Capabilities] \\n[Financial Markets & + Real Estate] \\nMenu\\n[Capabilities] \\n[Finance and Accounting] \\nMenu\\n[Capabilities] + \\n[Financial statement audit] \\nMenu\\n[Capabilities] \\n[Front Office] \\nMenu\\n[Capabilities] + \\n[Managed Services] \\nMenu\\n[Capabilities] \\n[Metaverse] \\nMenu\\n[Capabilities] + \\n[Operations] \\nMenu\\n[Capabilities] \\n[PwC Private] \\nMenu\\n[Capabilities] + \\n[Risk Modeling Services] \\nMenu\\n[Capabilities] \\n[Sustainability and + ESG] \\nMenu\\n[Capabilities] \\n[Tax services] \\nMenu\\n[Capabilities] \\n[Workforce] + \\nMenu\\n[Capabilities] \\n[All capabilities] \\nMenu\\n[Industries] \\n[Industries] + \\nMenu\\n[Industries] \\n[Aerospace and defense] \\nMenu\\n[Industries] \\n[Asset + and wealth management] \\nMenu\\n[Industries] \\n[Automotive] \\nMenu\\n[Industries] + \\n[Banking and capital markets] \\nMenu\\n[Industries] \\n[Chemicals] \\nMenu\\n[Industries] + \\n[Consumer markets] \\nMenu\\n[Industries] \\n[Consumer packaged goods] \\nMenu\\n[Industries] + \\n[Energy] \\nMenu\\n[Industries] \\n[Energy and industrials] \\nMenu\\n[Industries] + \\n[Engineering and construction] \\nMenu\\n[Industries] \\n[Financial services] + \\nMenu\\n[Industries] \\n[Gaming] \\nMenu\\n[Industries] \\n[Health industries] + \\nMenu\\n[Industries] \\n[Health services] \\nMenu\\n[Industries] \\n[Hospitality + and leisure] \\nMenu\\n[Industries] \\n[Industrial manufacturing] \\nMenu\\n[Industries] + \\n[Insurance] \\nMenu\\n[Industries] \\n[Media] \\nMenu\\n[Industries] \\n[Medical + technology] \\nMenu\\n[Industries] \\n[Pharmaceutical and life sciences] \\nMenu\\n[Industries] + \\n[Power and utilities] \\nMenu\\n[Industries] \\n[Private equity] \\nMenu\\n[Industries] + \\n[Real estate] \\nMenu\\n[Industries] \\n[Restaurants, wholesale and agriculture] + \\nMenu\\n[Industries] \\n[Retail] \\nMenu\\n[Industries] \\n[Space] \\nMenu\\n[Industries] + \\n[Sports] \\nMenu\\n[Industries] \\n[Technology] \\nMenu\\n[Industries] \\n[Technology, + media and telecommunications] \\nMenu\\n[Industries] \\n[Telecommunications] + \\nMenu\\n[Industries] \\n[Travel, transportation and logistics] \\nMenu\\n[Technology] + \\n[Technology] \\nMenu\\n[Technology] \\n[Alliances and ecosystems] \\nMenu\\n[Technology] + \\n[Delivery platforms] \\nMenu\\n[Technology] \\n[Emerging technology] \\nMenu\\n[Technology] + \\n[Engineering & AI] \\nMenu\\n[Technology] \\n[Products] \\nMenu\\n[Technology] + \\n[Tech-enabled services] \\nMenu\\n[Technology] \\n[Tech Effect] \\nMenu\\n[About + us] \\n[About us] \\nMenu\\n[About us] \\n[Alumni] \\nMenu\\n[About us] \\n[Analyst + relations] \\nMenu\\n[About us] \\n[Investing in our people] \\nMenu\\n[About + us] \\n[Newsroom] \\nMenu\\n[About us] \\n[Offices] \\nMenu\\n[About us] \\n[Our + leadership] \\nMenu\\n[About us] \\n[Purpose and values] \\nFeatured\\n[![] + \\nTech Effect\\n] \\nMenu\\n[Careers] \\n[Careers] \\nMenu\\n[Careers] \\n[Why + PwC] \\nMenu\\n[Careers] \\n[Entry Level Careers] \\nMenu\\n[Careers] \\n[Experienced + Careers] \\nMenu\\n[Careers] \\n[University Relations] \\nFeatured\\n[![] \\nShared + success benefits\\n] \\nLoading Results\\nNo Match Found\\nView All Results\\n# + PwC\u2019s AI Agent Survey\\nCopy linkLink copied to clipboard\\n![Example pattern + for mobile] \\n![] \\n## Summary\\n* Companies are increasing AI investments + and realizing productivity gains, but few are transforming how work gets done.\\n* + The biggest barrier isn\u2019t the technology; it\u2019s mindset, change readiness + and workforce engagement.\\n* Trust lags for high-stakes use cases, underscoring + the need for responsible AI.\\n* Few businesses are connecting agents across + workflows and functions, yet that\u2019s where the real value lies.\\n### Key + trends about AI agents in the enterprise\\nAI agents aren\u2019t just the future + \u2014they\u2019re already on the job. Today, they\u2019re delivering real results. + US companies are ramping up investments, and many are already seeing returns. + But despite the momentum, most haven\u2019t made the bold, strategic shifts + needed to sustain that success.\\nOf the 300 senior executives in our May 2025 + survey, 88% say their team or business function plans to increase AI-related + budgets in the next 12 months due to agentic AI. Seventy-nine percent say AI + agents are already being adopted in their companies. And of those adopting AI + agents, two-thirds (66%) say that they\u2019re delivering measurable value through + increased productivity.\\nYet from what we\u2019re seeing in the field, broad + adoption doesn\u2019t always mean deep impact. Many employees are using agentic + features built into enterprise apps to speed up routine tasks \u2014surfacing + insights, updating records, answering questions. It\u2019s a meaningful boost + in productivity, but it stops short of transformation.\\n> > While embedded + agents from hyperscalers and model providers are seeing strong uptake, the real + opportunity is still ahead. Reports of full adoption often reflect excitement + about what agentic capabilities could enable \u2014not evidence of widespread + transformation.\\n> > That\u2019s beginning to shift. Multi-agent models are + emerging as a powerful next step \u2014systems of AI agents working in concert + to deliver tangible results. These agents can take on complex, cross-functional + workflows across finance, customer service, software development, R&D and + more.\\nAI isn\u2019t just another technology, as more and more executives recognize. + Three-quarters (75%) agree or strongly agree that AI agents will reshape the + workplace more than the internet did. What\u2019s more, 71% agree that AI agents + are advancing so quickly that artificial general intelligence (AGI), in which + AI can think, learn and solve problems as broadly and flexibly as a human, will + be a reality within two years.\\nWith the potential for disruption growing, + playing it safe isn\u2019t a strategy. To lead with AI, you need to move with + intention. Our survey reveals four areas that demand your attention.\\n## 1. + Budgets are surging\\nAlmost nine out of 10 executives surveyed (88%) say their + companies plan to up their AI-related budgets this year due to agentic AI. Over + a quarter plan increases of 26% or more, likely to pay for ambitious plans.\\n> + > Of those companies adopting AI agents, 35% say they\u2019re doing so broadly, + while another 17% say AI agents are being fully adopted in almost all workflows + and functions.\\n> > Still, most (68%) report that half or fewer of their employees + interact with agents in their everyday work. Few have the kind of agentic workflows + that a[leading hospitality company] we worked with has already put in place. + There, employees and customers engage with teams of AI agents working across + functions \u2014enhancing experiences, improving speed and driving down costs.\\nThis + is a real-world example of multi-agent models in action \u2014AI agents working + together to deliver meaningful outcomes. While this is already delivering significant + value for one company, it signals what\u2019s next for many others.\\n## 2. + Measurable value is here\\nIf your company is like most, leadership (and other + stakeholders) want AI results now \u2014increased revenue, lower costs or other + concrete value \u2014and AI agents are delivering. Of those adopting AI agents, + nearly two-thirds (66%) report increased productivity. Over half (57%) report + cost savings, faster decision-making (55%) or improved customer experience (54%). + As with AI generally, early value often comes from internal use cases, but customer-facing + cases are rising fast.\\n> > 73% of survey respondents agree that\\nSummary: + None\\n\\n\\nTitle: The Rise of AI Agents: Redefining Automation and ...\\nURL: + https://www.intouchcx.com/thought-leadership/the-rise-of-ai-agents-redefining-automation-and-productivity-in-2025/\\nID: + https://www.intouchcx.com/thought-leadership/the-rise-of-ai-agents-redefining-automation-and-productivity-in-2025/\\nScore: + None\\nPublished Date: 2025-04-08T00:00:00.000Z\\nAuthor: IntouchCX Team\\nImage: + https://www.intouchcx.com/wp-content/uploads/2025/04/AdobeStock_1309149643_1024x580_V3-1.jpg\\nFavicon: + https://www.intouchcx.com/wp-content/themes/intouchcx/assets/favicon/favicon-32x32.png\\nExtras: + None\\nSubpages: None\\nText: The Rise of AI Agents: Redefining Automation and + Productivity in 2025 - IntouchCX\\nToggle menu[![Intouch logo]] \\n[![Intouch + logo]] Toggle menu\\n[Contact] [Login] Close menu\\n[Join Our Team] \\n[Contact] + [Login] [Join Our Team] \\n[![Facebook icon] Facebook] [![Twitter icon] Twitter] + [![Instagram icon] Linkedin] [![Pinterest icon] Instagram] [![YouTube icon] + YouTube] [![TikTok icon] TikTok] \\n* [Home] \\n* |[Blog Posts] \\n* |The Rise + of AI Agents: Redefining Automation and Productivity in 2025\\n![] \\n# Resources\\n# + The Rise of AI Agents: Redefining Automation and Productivity in 2025\\n* ![Picture + of IntouchCX Team] IntouchCX Team\\n* April 8, 2025\\n![] \\nThe integration + of AI into everyday business operations has reached a turning point. In 2024, + AI models became smarter and more efficient; in 2025, the focus shifts from + intelligence to action. AI agents are no longer just responsive tools\u2014they + are autonomous systems capable of executing complex workflows, making strategic + decisions, and seamlessly integrating into business operations to enhance productivity + and efficiency across industries, reshaping the way businesses operate and setting + new standards for automation and collaboration.\\nFrom Chatbots to Fully Autonomous + AI Agents\\nThe transition from conversational AI to AI agents is a game-changer. + While traditional AI models like chatbots primarily generate text and answer + questions, AI agents go further by taking independent actions based on contextual + understanding. These agents can schedule meetings, manage workflows, analyze + large datasets, and even navigate the web to perform complex tasks without human + intervention.\\n[**The future of AI isn’t just smarter chatbots\u2014it’s + fully autonomous AI agents that anticipate needs, execute tasks, and continuously + learn from interactions.**] This shift is transforming the way businesses operate, + integrating AI agents into daily workflows and leveraging advancements in AI + orchestration, where multiple agents collaborate in enterprise settings. Instead + of relying on a single AI model, businesses will deploy specialized AI agents + fine-tuned for different roles\u2014whether handling customer service inquiries, + managing supply chains, or optimizing marketing campaigns.\\n**The Business + Impact of AI Agents**\\nThe adoption of AI agents will transform industries + by boosting efficiency and scalability. Companies deploying these autonomous + systems expect significant productivity gains, with early estimates suggesting + up to a 30% increase in operational efficiency.\\n**Key benefits include:**\\n* + **Task Automation:**AI agents will handle repetitive and time-consuming processes, + allowing employees to focus on strategic initiatives.\\n* **Scalability:**Businesses + can expand operations without proportionally increasing costs, as AI agents + streamline workflows.\\n* **Enhanced Multimodal Capabilities:**Beyond text-based + interactions, AI agents will process images, audio, and video to provide richer + insights and solutions.\\n* **AI-Powered Decision-Making:**These agents will + leverage real-time data analysis to make informed recommendations, reducing + the need for human oversight in routine decision-making.\\nThe Emergence of + Multi-Agent Systems\\nAs businesses scale AI adoption, they will transition + from using single AI agents to deploying coordinated teams of specialized agents. + These AI-powered “swarms” will work in tandem, breaking down complex + workflows into manageable steps.\\nFor example, an enterprise AI ecosystem might + consist of:\\n* A scheduling agent that coordinates meetings based on team availability.\\n* + A customer support agent that resolves inquiries in real time.\\n* A research + agent that scans market trends and generates reports.\\nThis interconnected + approach enhances efficiency while ensuring that AI-driven processes remain + dynamic and adaptable to real-world business needs.\\nAI Agents and Human Oversight\\nDespite + their autonomy, AI agents will not operate in isolation. Human oversight remains + crucial for handling high-stakes decisions, ethical considerations, and complex + problem-solving. The challenge for businesses in 2025 will be striking the right + balance between automation and human intervention, ensuring that AI-driven processes + remain accountable and aligned with organizational goals.\\nThe Future of AI + Agents in Enterprise Workflows\\nLooking ahead, AI agents have the potential + to become an integral part of the workforce. As regulations catch up with rapid + technological advancements, businesses must navigate new frameworks for AI governance, + data security, and ethical AI deployment.\\nWith AI orchestration tools streamlining + operations and new protocols enabling seamless integration with third-party + technologies, AI agents will redefine what\u2019s possible in automation. From + handling routine administrative tasks to driving strategic decision-making, + these digital co-workers are set to revolutionize productivity, efficiency, + and innovation across industries.\\nHowever, AI cannot operate in isolation\u2014human + involvement remains essential to ensure accuracy, ethical decision-making, and + empathetic customer interactions. By automating repetitive processes, AI frees + up agents to focus on complex problem-solving and emotionally intelligent engagements, + enhancing both efficiency and customer satisfaction.\\nAI agents are not just + the next step in automation\u2014they are the future of work. Find out how you + can leverage[**AI solutions**] in your CX strategy now!\\n## Recent Posts\\n[![Transforming + Knowledge Management Through Strategic Sprints] February 4, 2026.### Transforming + Knowledge Management Through Strategic Sprints\\nCustomer experience often comes + down to one simple factor: how easily frontline agents can find […]\\nRead + more] \\n[![AI, Automation, and Customer-Centric Travel: The Big Trends for + the Transportation Industry] January 28, 2026.### AI, Automation, and Customer-Centric + Travel: The Big Trends for the Transportation Industry\\nThe Future in Motion: + How AI and Automation Are Shaping Travel and Transportation \_Executive […]\\nRead + more] \\n[![From Friction to Flow: Solving Common CX Pain Points With Trust + & Safety] January 21, 2026.### From Friction to Flow: Solving Common CX + Pain Points With Trust & Safety\\nA flawless customer journey can be ruined + in seconds if a user encounters fraud, harassment, […]\\nRead more] \\n![graphic]![graphic]![graphic]![graphic]![graphic]\\nSummary: + None\\n\\n\\nTitle: AI Agents in 2025: Expectations vs. Reality\\nURL: https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality\\nID: + https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality\\nScore: + None\\nPublished Date: 2025-02-26T00:00:00.000Z\\nAuthor: None\\nImage: https://www.ibm.com/content/dam/connectedassets-adobe-cms/worldwide-content/gbs/cf/ul/g/5a/88/336.jpg/_jcr_content/renditions/cq5dam.thumbnail.1280.1280.png\\nFavicon: + https://www.ibm.com/content/dam/adobe-cms/default-images/icon-32x32.png\\nExtras: + None\\nSubpages: None\\nText: AI Agents in 2025: Expectations vs. Reality | + IBM\\nTags\\n[Artificial Intelligence] \\n# AI agents in 2025: Expectations + vs. reality\\n![the same person doing multiple jobs] \\n## Authors\\n[Ivan Belcic] + \\nStaff writer\\n[Cole Stryker] \\nStaff Editor, AI Models\\nIBM Think\\n## + AI agents in 2025: Expectations vs. reality\\nIt\u2019s impossible to take two + steps across the tech media landscape without stumbling over an article hailing + 2025 as the year of the[AI agent]. Agents, we\u2019re told, will transform the + way work is done, impacting every facet of our lives, personal and professional.\\nWe\u2019d + barely surfaced from a landslide of NFT and crypto hype that characterized the + early 2020s, and the metaverse bubble that followed, before media voices began + singing the praises of[generative AI] (gen AI) in the wake of releases such + as OpenAI\u2019s[GPT] model family, Anthropic\u2019s[Claude] and Microsoft\u2019s + Copilot.\\nWhile the chorus hasn\u2019t moved on entirely, the focus in 2025 + has shifted from[large language models (LLMs)] to advancements in the ostensibly + autonomous[artificial intelligence (AI)] agents ushering in the future of work.\\nDespite + a momentary surge in gen AI interest around[Deepseek] \u2019s R1, which promised + significant performance improvements over ChatGPT, the dominant innovation narrative + in 2025 is the AI agent.\\nMedia coverage highlights the promises of innovation,[automation] + and efficiency agents will bring, but how much of the conversation is click-hungry + hype?\\nThe ad-supported media world thrives on clicks, and it\u2019s reasonable + to expect sensational, attention-grabbing headlines crafted to garner yours. + But what can we realistically expect from[agentic AI] in 2025, and how will + it affect our lives?\\nWe spoke with several IBM experts to cut through the + hype, with the goal of holding a more reasonable conversation about AI agents + and what they\u2019re going to do. Our team of informed insiders includes:\\n* + [Maryam Ashoori, PhD]: Director of Product Management,[IBM\xAE watsonx.ai\u2122] + \\n* [Marina Danilevsky]: Senior Research Scientist, Language Technologies\\n* + [Vyoma Gajjar]: AI Technical Solutions Architect\\n* [Chris Hay]: Distinguished + Engineer\\nIndustry newsletter\\n### The latest AI trends, brought to you by + experts\\nGet curated insights on the most important\u2014and intriguing\u2014AI + news. Subscribe to our weekly Think newsletter. See the[IBM Privacy Statement].\\n### + Thank you! You are subscribed.\\nYour subscription will be delivered in English. + You will find an unsubscribe link in every newsletter. You can manage your subscriptions + or unsubscribe[here]. Refer to our[IBM Privacy Statement] for more information.\\n## + What are AI agents?\\nAn AI agent is a software program capable of acting autonomously + to understand, plan and execute tasks. AI agents are powered by LLMs and can + interface with tools, other models and other aspects of a system or network + as needed to fulfill user goals.\\nWe\u2019re going beyond asking a chatbot + to suggest a dinner recipe based on the available ingredients in the fridge. + Agents are more than automated[customer experience] emails that inform you it\u2019ll + be a few days until a real-world human can get to your inquiry.\\n[AI agents + differ from traditional AI assistants] that need a prompt each time they generate + a response. In theory, a user gives an agent a high-level task, and the agent + figures out how to complete it.\\nCurrent offerings are still in the early stages + of approaching this idea. \u201CWhat\u2019s commonly referred to as \u2018agents\u2019 + in the market is the addition of rudimentary planning and tool-calling (sometimes + called function calling) capabilities to LLMs,\u201D says Ashoori. \u201CThese + enable the LLM to break down complex tasks into smaller steps that the LLM can + perform.\u201D\\nHay is optimistic that more robust agents are on the way: \u201CYou + wouldn\u2019t need any further progression in models today to build[future AI + agents],\u201D he says.\\nWith that out of the way, what\u2019s the conversation + about agents over the coming year, and how much of it can we take seriously?\\nAI + agents\\n### What are AI agents?\\nFrom monolithic models to compound AI systems, + discover how AI agents integrate with databases and external tools to enhance + problem-solving capabilities and adaptability.\\n[\\nLearn more\\n] \\n## Narrative + 1: 2025 is the year of the AI agent\\n\u201CMore and better agents\u201D are + on the way, predicts Time.[1] \u201CAutonomous \u2018agents\u2019 and profitability + are likely to dominate the artificial intelligence agenda,\u201D reports Reuters.[2] + \u201CThe age of agentic AI has arrived,\u201D promises Forbes, in response + to a claim from Nvidia\u2019s Jensen Huang.[3] \\nTech media is awash with assurances + that our lives are on the verge of a total transformation. Autonomous agents + are poised to streamline and alter our jobs, drive optimization and accompany + us in our daily lives, handling our mundanities in real time and freeing us + up for creative pursuits and other higher-level tasks.\\n### 2025 as the year + of agentic exploration\\n\u201CIBM and Morning Consult did a survey of 1,000 + developers who are building AI applications for enterprise, and 99% of them + said they are exploring or developing AI agents,\u201D\_explains Ashoori. \u201CSo + yes, the answer is that 2025 is going to be the year of the agent.\u201D However, + that declaration is not without nuance.\\nAfter establishing the current market + conception of agents as LLMs with function calling, Ashoori draws a distinction + between that idea and truly autonomous agents. \u201CThe true definition [of + an AI agent] is an intelligent entity with reasoning and planning capabilities + that can autonomously take action. Those reasoning and planning capabilities + are up for discussion. It depends on how you define that.\u201D\\n\u201CI definitely + see AI agents heading in this direction, but we\u2019re not fully there yet,\u201D + says Gajjar. \u201CRight now, we\u2019re seeing early glimpses\u2014AI agents + can already analyze data, predict trends and automate workflows to some extent. + But building AI agents that can autonomously handle complex decision-making + will take more than just better algorithms. We\u2019ll need big leaps in contextual + reasoning and testing for edge cases,\u201D she adds.\\nDanilevsky isn\u2019t + convinced that this is anything new. \u201CI'm still struggling to truly + believe that this is all that different from just orchestration,\u201D she says. + \u201CYou've renamed orchestration, but now it's called agents, because + that's the cool word. But orchestration is something that we've been + doing in programming forever.\u201D\\nWith regard to 2025 being the year of + the agent, Danilevsky is skeptical. \u201CIt depends on what you say an agent + is, what you think an agent is going to accomplish and what kind of value you + think it will bring,\u201D she says. \u201CIt's quite a statement to make + when we haven't even yet figured out ROI (return on investment) on LLM technology + more generally.\u201D\\nAnd it\u2019s not just the business side that has her + hedging her bets. \u201CThere's the hype of imagining if this thing could + think for you and make all these decisions and take actions on your computer. + Realistically, that's terrifying.\u201D\\nDanilevsky frames the disconnect + as one of miscommunication. \u201C[Agents] tend to be very ineffective because + humans are very bad communicators. We still can't get chat agents to interpret + what you want correctly all the time.\u201D\\nStill, the forthcoming year holds + a lot of promise as an era of experimentation. \u201CI'm a big believer + in [2025 as the year of the agent],\u201D says Hay excitedly.\\nEvery large + tech company and hundreds of startups are now experimenting with agents. Salesforce, + for example, has released their Agentforce platform, which enables users to + create agents that are easily integrated within the Salesforce app ecosystem.\\n\u201CThe + wave is coming and we're going to have a lot of agents. It's still a + very nascent ecosystem, so I think a lot of people are going to build agents, + and they're going to have a lot of fun.\u201D\\n## Narrative 2: Agents can + handle highly complex tasks on their own\\nThis narrative assumes that today\u2019s + agents meet the theoretical definition outlined in the introduction to this + piece. 2025\u2019s agents will be fully autonomous AI programs that can scope + out a project and complete it with all the necessary tools they need and with + no help from human partners. But what\u2019s missing from this narrative is + nuance.\\n### Today\u2019s models are more than enough\\nHay believes that the + groundwork has already been laid for such developments. \u201CThe big thing + about agents is that they have the ability to plan,\u201D he outlines. \u201CThey + have the ability to reason, to use tools and perform tasks, and they need to + do it at speed and scale.\u201D\\nHe cites 4 developments that, compared to + the best models of 12 to 18 months ago, mean that the models of early 2025 can + power the agents envisioned by the proponents of this narrative:\\n* Better, + faster, smaller models\\n* Chain-of-thought (COT) training\\n* Increased context + windows\\n* Function calling\\n\u201CNow, most of these things are in play,\u201D + Hay continues. \u201CYou can have the AI call tools. It can plan. It can reason + and come back with good answers. It can use inference-time compute. You\u2019ll + have better chains of thought and more memory to work with. It's going to + run fast. It\u2019s going to be cheap. That leads you to a structure where I + think you can have agents. The models are improving and they're getting + better, so that's only going to accelerate.\u201D\\n### Realistic expectations + are a must\\nAshoori is careful to differentiate between what agents will be + able to do later, and what they can do now. \u201CThere is the promise, and + there is what the agent's capable of doing today,\u201D she says. \u201CI + would say the answer depends on the use case. For simple use cases, the agents + are capable of [choosing the correct tool], but for more sophisticated use cases, + the technology\\nSummary: None\\n\\n\\nTitle: Top Tried and Tested Use Cases + for Autonomous AI ...\\nURL: https://www.druidai.com/blog/top-tried-and-tested-use-cases-for-autonomous-ai-agents-in-2025\\nID: + https://www.druidai.com/blog/top-tried-and-tested-use-cases-for-autonomous-ai-agents-in-2025\\nScore: + None\\nPublished Date: 2025-05-23T00:00:00.000Z\\nAuthor: None\\nImage: https://www.druidai.com/hubfs/graphic-of-hashtag-true-story-about-ai-agent-use-cases.jpg\\nFavicon: + https://www.druidai.com/hubfs/raw_assets/public/stoica-Druid/images/favicon.png\\nExtras: + None\\nSubpages: None\\nText: Top Tried and Tested Use Cases for Autonomous + AI Agents in 2025\\n[Skip to content] \\n[Blog] \\nai agents\\nMay 23, 2025\\n6 + MIN READ\\n# Top Tried and Tested Use Cases for Autonomous AI Agents in 2025\\n![Andreea + Radulescu] Andreea Radulescu\\n![graphic-of-hashtag-true-story-about-ai-agent-use-cases] + \\nImagine your calendar reminding you of a meeting, then booking the room, + sending the brief, gathering the right people, and following up afterward with + action items. Ok, your calendar cannot do all of that; that\u2019s what an autonomous + AI agent brings to the table. It doesn\u2019t just notify you. It thinks, acts, + and makes things happen in the background so you don\u2019t have to.\\nThese + aren\u2019t glorified chatbots. They\u2019re your new smart digital coworkers, + always on, never confused, and constantly improving with every task they complete. + Whether it\u2019s streamlining finance, simplifying HR, supporting students, + or managing supply chains, AI agents are already embedded in the workflows that + matter.\\nLet\u2019s explore the real-world use cases showing why these agents + are the backbone of affordable, scalable automation, and why they\u2019re becoming + a must-have in every enterprise playbook.\\n## Why 2025 Wasthe Year of the Autonomous + AI Agent\\nAccording to recent insights from top research and tech publications, + the adoption of agentic AI has surged as organizations seek smarter ways to + automate with accountability. Autonomous AI agents, when designed with orchestration, + security, and lifecycle management in mind, become more than tools; they become + proactive team members who get things done.\\nAnd the ROI? It\u2019s compelling. + Studies point to 30\u201380% improvements in speed, accuracy, and cost savings + across industries when deploying agentic AI workflows. Let's dig into what these + agents are actually doing in the real world.\\n## Most In-Demand Use Cases for + Autonomous AI Agents in 2025\\nAs more organizations shift from static automation + to intelligent systems that adapt, decide, and execute, autonomous AI agents + are stepping into roles once handled by multiple tools, and sometimes, entire + teams. These agents aren\u2019t just supporting workflows; they\u2019re running + them from start to finish. What was once limited to isolated tasks is now evolving + into coordinated, cross-functional processes.\\nBelow are the most in-demand, + high-impact use cases, showcasing where autonomous AI agents deliveredreal business + value in 2025.\\n### Finance & Accounting Automation\\n* Accounts Payable + & Receivable: Automatically process invoices, perform PO, matching, approve + payments, and reconcile accounts with[90%+ accuracy and 70% lower costs].\\n* + Expense Management: Track, validate, and report on expenses, cutting approval + cycle time in half.\\n* Tax Compliance: Assist tax teams with scenario modeling, + rule updates, and automated filings while reducing liability and risk.### HR + & Workforce Operations\\n* Employee Onboarding & Offboarding: From generating + contracts and collecting e-signatures to system provisioning and training reminders,[AI + agents automate the full lifecycle].\\n* HR Admin Support: Field payroll queries, + issue certificates, update records, and enable 24/7 employee self-service.\\n* + Recruitment & Interview Scheduling: Pre-screen candidates, assess resumes, + and coordinate interview logistics, all autonomously.### Healthcare Support + Systems\\n* Patient Scheduling & Triage:[Book appointments], assess symptoms, + and escalate cases automatically, cutting call volume and reducing delays.\\n* + Insurance & Billing Queries: Handle routine billing inquiries with full + access to patient policies and history.\\n* Prescription Refills: Accept and + process refill requests while verifying eligibility and status.### Banking & + Financial Services\\n* Loan Application Processing: Guide applicants through + steps, verify data, and make eligibility checks, all without human input.\\n* + Fraud Detection Agents: Monitor transactions in real time and flag anomalies + using ML and historical fraud data.\\n* Investment Advisory Assistants: Generate + personalized investment insights based on customer profiles and market data.### + Higher Education\\n* Student Engagement Assistants: Manage inquiries across + admissions, financial aid, housing, and course enrollment, supporting thousands + of interactions per day.\\n* Admissions & Application Agents: Answer applicant + questions, collect documents, and update application status.\\n* IT Helpdesk + for Campus Services:[Provide 24/7 support] for login issues, software access, + and account setup.### Manufacturing & Supply Chain\\n* Predictive Maintenance: + Monitor equipment health and schedule repairs before breakdowns happen.\\n* + Procurement & Contract Management: Auto-generate purchase orders, match + contracts, track deliveries, and validate invoices.\\n* Inventory Optimization + Agents: Analyze stock levels, demand trends, and lead times to trigger replenishments + and reduce holding.## The Real Value of Smart Digital Coworkers\\nWhile many + businesses start exploring[autonomous AI agents for efficiency], the real payoff + goes far beyond shaving minutes off a task. When implemented with orchestration, + secure integrations, and lifecycle visibility, these agents transform how work + gets done, and who does it.\\nHere\u2019s what they really bring to the table:\\n### + Always-On Productivity with Autonomous AI Agents\\nOne of the most immediate + benefits of autonomous AI agents is their 24/7 availability. Unlike human teams, + they don\u2019t sleep, get overloaded, or miss an email. These smart digital + coworkers keep processes moving around the clock, whether it\u2019s answering + internal HR questions at 2 AM or processing invoices over the weekend. This + uninterrupted productivity reduces lag, accelerates workflows, and allows your + teams to stay focused on higher-value work.\\n### Contextual Decision-Making + Using Gen AI and Agentic Execution\\nSmart automation isn\u2019t about doing + things faster, it\u2019s about doing them better.[Autonomous AI agents equipped + with generative AI] can understand the nuance behind a request, while agentic + workflows give them the power to act on that insight. This combination allows + agents to make decisions based on user history, policy data, and workflow context, + resulting in responses and actions that are not just accurate, but relevant + and timely.\\n### End-to-End Automation Through Integrated AI Agent Workflows\\nModern + enterprises don\u2019t need point solutions, they need systems that can handle + tasks from start to finish. AI agents shine in end-to-end automation, managing + everything from initiation to validation, updates, and closure. Whether it\u2019s + onboarding a new employee, processing a refund, or handling a service request, + these agents coordinate actions across systems to ensure no task is dropped + mid-stream.\\n### Real ROI with Scalable, Low-Overhead AI Agents\\nIt\u2019s + not just about speed or intelligence. It\u2019s about impact. Autonomous AI + agents offer a clear return on investment by reducing reliance on manual labor, + minimizing errors, and increasing process consistency. With fewer tools needed + to complete a workflow, less strain on human teams, and faster time-to-value, + these agents prove their worth quickly. And because they\u2019re modular, they + scale with your business, without spiraling complexity or cost.\\n### Smarter + Over Time: Learning and Adapting as They Work\\nThe value of smart digital coworkers + compounds over time. These agents don\u2019t just perform, they learn. With + every task, they gather more data, improve accuracy, and adapt to new business + logic or user behavior. As your business evolves, your agents evolve with it, + offering a level of continuity and improvement that static workflows simply + can\u2019t match.\\n## Complete Industry Use-Case List for Autonomous AI Agents\\n### + Banking & Financial Services\\n* Loan application intake and eligibility + checks\\n* KYC (Know Your Customer) data collection and validation\\n* Real-time + fraud detection and alerting\\n* Personalized customer financial advice\\n* + Payment dispute handling and resolution\\n* Internal compliance monitoring and + audit preparation\\n* Secure account access support and FAQs### Finance & + Accounting\\n* Invoice processing and purchase order matching\\n* Expense report + validation and routing\\n* Financial data reconciliation\\n* Tax calculation, + policy lookup, and filing support\\n* Vendor payment approvals and reminders\\n* + Budget variance analysis and reporting### Human Resources\\n* Employee onboarding/offboarding + automation\\n* Leave and time-off requests\\n* Benefits eligibility and enrollment + support\\n* Payroll inquiries and payslip distribution\\n* HR policy Q&A + and compliance notifications\\n* Pre-screening and recruitment workflow automation\\n* + Internal job referrals and career development guidance### Healthcare\\n* Appointment + scheduling and reminders\\n* Digital symptom checker and triage\\n* Insurance + verification and billing support\\n* Patient intake form collection and follow-up\\n* + Provider directory search and recommendations\\n* Claims processing and status + updates\\n* Post-care follow-up and satisfaction surveys### Higher Education\\n* + Admissions inquiries and application assistance\\n* Course registration and + deadline reminders\\n* Financial aid Q&A and eligibility guidance\\n* IT + helpdesk support for students and faculty\\n* Virtual onboarding for new students + or staff\\n* Scheduling academic advising appointments\\n* Campus facilities + requests and tracking### Retail & eCommerce\\n* Order status tracking and + updates\\n* Product availability and recommendations\\n* Customer complaint + resolution\\n* Store locator and in-store service scheduling\\n* Returns and + refund initiation\\n* Post-purchase survey and feedback capture\\n* Loyalty + program support and personalization### Manufacturing & Supply Chain\\n* + Predictive maintenance scheduling\\n* Inventory management and reordering\\n* + Supplier onboarding and performance tracking\\n* Shipment tracking and logistics + optimization\\n* Procurement approvals\\nSummary: None\\n\\n\\nTitle: How autonomous + AI agents enhance campaign planning ...\\nURL: https://www.glean.com/perspectives/how-autonomous-ai-agents-enhance-campaign-planning-in-2025\\nID: + https://www.glean.com/perspectives/how-autonomous-ai-agents-enhance-campaign-planning-in-2025\\nScore: + None\\nPublished Date: 2025-12-17T00:00:00.000Z\\nAuthor: The Glean Team\\nImage: + https://cdn.prod.website-files.com/6127a84dfe068e153ef20572/66ded1bc82df72f2e1d56eb7_Glean%20Logomark%20Blue.svg\\nFavicon: + https://cdn.prod.website-files.com/6127a84dfe068e153ef20572/67189bf127c626679a308b22_32x32%20Glean%20Favicon.png\\nExtras: + None\\nSubpages: None\\nText: How autonomous AI agents enhance campaign planning + in 2025\\n[![Glean Logomark Blue]] \\nProduct\\nWORK AI\_PLATFORM\\n[\\nPlatform + Overview\\n![Platform Overview BG]] \\n[\\nGlean Assistant\\nYour personal AI + assistant\\n] [All Knowledge] [Data Analysis] [Deep Research] [Canvas] [Companion] + [Prompt Library] \\n[\\nGlean Agents\\nBuild and manage AI agents\\n] [Agent + Builder] [Agent Governance] [Agent Orchestration] [Agent Library] \\n[\\nGlean + Search\\nThe foundation of enterprise AI\\n] [Enterprise Graph] [Personal Graph] + [System of context] [Hybrid Search] \\n[\\nConnectors & Actions\\nConnect + to all your apps\\n] [\\nModel Hub\\nGet access to the latest models\\n] [\\nAPIs\\nBuild + generative AI experiences\\n] [\\nGlean Protect\\nSafely scale AI at work\\n] + [\\nSecurity\\nSecure by design from day one\\n] [\\nAgentic Engine\\nPlan & + adapt over company context\\n] \\nGLEAN WHERE\_YOU\_WORK\\n[![] \\nGlean in + Slack\\n] [![] \\nGlean in Microsoft Teams\\n] [![] \\nGlean in Zoom\\n] [![] + \\nGlean in Service Cloud\\n] [![] \\nGlean in ServiceNow\\n] [![] \\nGlean + in Zendesk\\n] [![] \\nGlean in GitHub\\n] [![] \\nGlean in Miro\\n] [![] \\nBrowser + Extension\\n] \\n[Sign in] \\n[\\nCustomers\\n] \\nSolutions\\nDEPARTMENTS\\n[\\nAll + Teams\\n] [\\nEngineering\\n] [\\nCustomer Service\\n] [\\nSales\\n] [\\nMarketing\\n] + [\\nB2B Marketing\\n] [\\nB2C Marketing\\n] [\\nPeople\\n] [\\nIT\\n] \\nINDUSTRIES\\n[\\nRetail\\n] + [\\nFinancial Services\\n] [\\nHigher Education\\n] [\\nHealthcare\\n] [\\nGovernment\\n] + \\n[Sign in] \\n[\\n![] \\nJoel McKelvey\\nHead of Solutions, Glean\\n![] \\nAbdullah + Haydar\\nDirector of Engineering, LinkedIn\\nWebinar\\nAI\_Powered Engineering\\nExpert + insights and actionable strategies for accelerating developer productivity.\\nWatch + now\\n] \\nResources\\nEXPLORE\\n[Blog] [Prompt Library] [Guides] [Product Videos] + \\nENGAGE\\n[Webinars] [Newsroom] [Glean:GO] [Events] [Gleaniverse Community] + \\nSUPPORT & SERVICES\\n[Help center] [Developers] [Partners] \\n[\\nWork + AI Institute\\n![]![]] \\n[Sign in] \\n[![] \\nThe AI Transformation 100\\nExplore + 100 real-world moves organizations are making to transform themselves with AI.\\nDownload + the report\\n] \\n[\\nAbout\\n] \\nThank you! Your submission has been received!\\nOops! + Something went wrong while submitting the form.\\n[Sign in] \\n[Sign in] \\n[\\nGet + a demo\\n] \\n[\\nGet a demo\\n] \\n[![Glean Logomark Blue]] \\n[Sign in] \\n[Get + a demo] \\n[Get a demo] \\nProduct\\n[\\nCustomers\\n] \\nSolutions\\nResources\\n[\\nAbout\\n] + \\n[Sign in] \\nBack\\nWORK AI\_PLATFORM\\n[\\nPlatform Overview\\n![Platform + Overview BG]] \\n[\\nGlean Assistant\\nYour personal AI assistant\\n] [All Knowledge] + [Data Analysis] [Deep Research] [Canvas] [Companion] [Prompt Library] \\n[\\nGlean + Agents\\nBuild and manage AI agents\\n] [Agent Builder] [Agent Governance] [Agent + Orchestration] [Agent Library] \\n[\\nGlean Search\\nThe foundation of enterprise + AI\\n] [Enterprise Graph] [Personal Graph] [System of context] [Hybrid Search] + \\n[\\nConnectors & Actions\\nConnect to all your apps\\n] [\\nModel Hub\\nGet + access to the latest models\\n] [\\nAPIs\\nBuild generative AI experiences\\n] + [\\nGlean Protect\\nSafely scale AI at work\\n] [\\nSecurity\\nSecure by design + from day one\\n] [\\nAgentic Engine\\nPlan & adapt over company context\\n] + \\nGLEAN WHERE\_YOU\_WORK\\n[![] \\nGlean in Slack\\n] [![] \\nGlean in Microsoft + Teams\\n] [![] \\nGlean in Zoom\\n] [![] \\nGlean in Service Cloud\\n] [![] + \\nGlean in ServiceNow\\n] [![] \\nGlean in Zendesk\\n] [![] \\nGlean in GitHub\\n] + [![] \\nGlean in Miro\\n] [![] \\nBrowser Extension\\n] \\n[Sign in] \\nDEPARTMENTS\\n[\\nAll + Teams\\n] [\\nEngineering\\n] [\\nCustomer Service\\n] [\\nSales\\n] [\\nMarketing\\n] + [\\nB2B Marketing\\n] [\\nB2C Marketing\\n] [\\nPeople\\n] [\\nIT\\n] \\nINDUSTRIES\\n[\\nRetail\\n] + [\\nFinancial Services\\n] [\\nHigher Education\\n] [\\nHealthcare\\n] [\\nGovernment\\n] + \\n[Sign in] \\n[\\n![] \\nJoel McKelvey\\nHead of Solutions, Glean\\n![] \\nAbdullah + Haydar\\nDirector of Engineering, LinkedIn\\nWebinar\\nAI\_Powered Engineering\\nExpert + insights and actionable strategies for accelerating developer productivity.\\nWatch + now\\n] \\nEXPLORE\\n[Blog] [Prompt Library] [Guides] [Product Videos] \\nENGAGE\\n[Webinars] + [Newsroom] [Glean:GO] [Events] [Gleaniverse Community] \\nSUPPORT & SERVICES\\n[Help + center] [Developers] [Partners] \\n[\\nWork AI Institute\\n![]![]] \\n[Sign + in] \\n[![] \\nThe AI Transformation 100\\nExplore 100 real-world moves organizations + are making to transform themselves with AI.\\nDownload the report\\n] \\nLast + updated Dec 17, 2025.\\n# How autonomous AI agents enhance campaign planning + in 2025\\n0\\nminutes read\\n![How autonomous AI agents enhance campaign planning + in 2025] \\n### Table of contents\\n[\\nHeading 2\\n] \\n[\\nHeading 3\\n] \\n[\\nHeading + 4\\n] \\n[\\nHeading 5\\n] \\n[\\nHeading 6\\n] \\n[\\nHave questions or want + a demo?\\nWe\u2019re here to help! Click the button below and we\u2019ll be + in touch.\\nGet a Demo\\n] \\nShare this article:\\n[\\n] [\\n] \\n# How autonomous + AI agents enhance campaign planning in 2025\\nMarketing teams face unprecedented + complexity as campaigns span multiple channels, require real-time optimization, + and demand[personalized experiences] at scale. Traditional automation tools + execute predefined tasks but lack the intelligence to adapt strategies when + market conditions shift or customer behaviors change unexpectedly.\\nA new generation + of AI technology promises to transform how organizations plan campaigns and + collaborate with clients. These autonomous systems analyze vast amounts of data, + make strategic decisions, and execute[complex workflows] without constant human + oversight \u2014fundamentally changing the relationship between marketers and + their technology stack.\\nThe shift from reactive automation to proactive intelligence + represents more than incremental improvement. The global AI agent market was + estimated at[$5.40 billion] in 2024 and is forecast to reach $50.31 billion + by 2030, representing explosive growth driven by expanding enterprise adoption. + Organizations deploying these advanced AI capabilities report efficiency gains + of 30-50%, dramatic reductions in campaign optimization cycles, and enhanced + client satisfaction through real-time transparency and strategic insights.\\nAutonomous + AI agents are intelligent systems designed to work independently, performing + tasks such as campaign planning and client collaboration without constant human + oversight. They leverage machine learning to adapt and execute complex strategies, + enhancing efficiency and creativity in marketing initiatives. These agents integrate + with external tools, retrieve relevant information, and adapt to dynamic environments + through sophisticated reasoning architectures. Companies implementing marketing + automation, including autonomous agent components, achieve average returns of + $5.44 for every $1 invested, translating to[544%] return on investment over + three years.\\n## What are autonomous AI agents?\\nAutonomous AI agents are + intelligent systems designed to work independently, performing tasks such as + campaign planning and client collaboration without constant human oversight. + They leverage machine learning to adapt and execute complex strategies, enhancing + efficiency and creativity in marketing initiatives. These agents integrate with + external tools, retrieve relevant information, and adapt to dynamic environments + through sophisticated reasoning architectures. Among companies currently deploying + AI agents,[66%] report increased productivity while over half report cost savings + (57%), faster decision-making (55%), and improved customer experience (54%).\\nAutonomous + AI agents are intelligent systems designed to work independently, performing + tasks such as campaign planning and client collaboration without constant human + oversight. They leverage machine learning to adapt and execute complex strategies, + enhancing efficiency and creativity in marketing initiatives. These agents integrate + with external tools, retrieve relevant information, and adapt to dynamic environments + through sophisticated reasoning architectures. Notably, workers using generative + AI every day reported that[33.5%] saved four hours or more per week, compared + with only 11.5% of those using it only one day per week, showing a direct correlation + between AI usage intensity and productivity gains.\\nUnlike traditional marketing + automation that follows rigid "if-then" logic, autonomous agents process + multiple signals simultaneously and adjust their approach based on context. + They observe their environment, make decisions, and take actions to achieve + specific marketing objectives \u2014whether that means optimizing ad spend,\\nSummary: + None\\n\\n\\nTitle: What are Autonomous AI Agents? A Complete Guide 2025\\nURL: + https://kodexolabs.com/what-are-autonomous-ai-agents/\\nID: https://kodexolabs.com/what-are-autonomous-ai-agents/\\nScore: + None\\nPublished Date: 2025-07-31T00:00:00.000Z\\nAuthor: None\\nImage: https://kodexolabs.com/wp-content/uploads/2025/07/What-Are-Autonomous-AI-Agents-A-Complete-Guide-for-2025.webp\\nFavicon: + https://kodexolabs.com/wp-content/uploads/2024/11/1-05-2-150x150.webp\\nExtras: + None\\nSubpages: None\\nText: What are Autonomous AI Agents? A Complete Guide + 2025[Skip to content] \\n[![]] \\n[About us] \\n[What We Do] \\n![]![] [Get + A Free AI Chatbot] \\n### Generative AI\\n* [Gen AI Development] \\n* [Gen AI + Integration] \\n* [ChatGPT Dev & Integration] \\n* [Gen AI Model Development] + \\n* [Gen AI Consulting] ### Product Designing\\n* [Product Designing] \\n### + AI Development\\n* [AI Development] \\n* [AI Chatbot Development] \\n* [AI Consulting] + \\n* [AI Model Development] \\n* [Custom AI Solutions] ### ML Development\\n* + [ML Development] \\n* [ML Consulting] \\n* [ML Model Engineering] \\n* [MLOps + Implementation] \\n### Software Development\\n* [Software Development Services] + \\n* [Custom Product Development] \\n* [Software Consulting] \\n* [Mobile App + Development] \\n* [Web App Development] ### Data Engineering\\n* [Data Engineering] + \\n* [Data Analytics] \\n* [Data Annotation] \\n[Who We Serve] \\n![]![] [Get + A Free AI Chatbot] \\n[### HealthCare\\n] EHR Systems, AI based Interviews and + Medical Imaging Software[### EdTech\\n] Personalized Learning, AI based Tutor + Systems and Gamification Experiences[### Fintech\\n] AI powered Trend Forecasting + and Predicative Analytics\\n[### Energy\\n] Smart Grid Solutions and AI based + Resource Monitoring[### Automotive\\n] Predictive Maintenance, Driver Assistance + and AI Chatbots[### Real Estate\\n] AI Home Management and AI based Real Estate + Evaluation Systems\\n[### IT and Tech\\n] AI powered Ticket Generation and Automated + Software Production[### Marketing\\n] Customer Churn Prediction, Customer Segmentation + and AI based Analytics\\n[Hire Dev] \\n![]![] [Get A Free AI Chatbot] \\n[### + IT Staff Augmentation\\n] On-demand Talent, Scalable Teams, Flexible Hiring[### + Hire Software Developer\\n] Custom Software, Full-stack, Agile Development[### + Software Development Outsourcing\\n] End-to-End, Project-based, Flexible Engagement\\n[### + Hire AI Developer\\n] AI Solutions, Machine Learning, Custom Models[### Hire + Offshore Developer\\n] Remote Teams, Cost-efficient, Dedicated Experts\\n[### + Hire Data Engineer\\n] Data Pipelines, ETL, Big Data Solutions[### Dedicated + Development Team\\n] Tailored Solutions, Seamless Collaboration, Scalability\\n[Our + Work] \\n[Solutions] \\n![]![] [Get A Free AI Chatbot] \\n### Custom Enterprise + Solutions\\n* [Enterprise Resource Planning (ERP)] \\n* [Human Resource Management + Solutions] \\n* [Asset Management Software Solutions] \\n* [Supply Chain Management + Solutions] \\n* [Business Process Automation Software] \\n* [Fleet Management + Software] \\n### Healthcare Software Solutions\\n* [AI-Powered Medical Imaging + & Diagnostics] \\n* [Custom Medical Practice Management Software] \\n[Company] + \\n![]![] [Get A Free AI Chatbot] \\n[### Careers\\n] Advance your career in + AI and software[### Blogs\\n] Official Blogs for News, Tech & Culture\\n[### + Awards & Achievements\\n] Honored for excellence in AI innovations\\n[Contact + Us] \\n[![]] \\n[] \\n# What Are Autonomous AI Agents? A Complete Guide for + 2025 and Beyond\\nSyed Ali Hasan Shah\\n[Agentic AI] \\nJuly 31, 2025\\nSyed + Ali Hasan Shah\\n[Agentic AI] \\nJuly 31, 2025\\nTable Of Contents\\n1. [Share + This Article] \\n2. [Introduction] \\n3. [What Are Autonomous AI Agents? Understanding + the Fundamentals] \\n* [What Makes an AI Agent Autonomous?] \\n* * [Autonomous + Agents vs Traditional AI Systems] \\n* * [Key Characteristics of Modern Autonomous + Agents] \\n* [How Do Autonomous AI Agents Work? Technical Architecture Explained] + \\n* [Core Components of Autonomous AI Systems] \\n* * [Types of Autonomous + Agents by Intelligence Level] \\n* * [Machine Learning Integration in Agent + Architecture] \\n* [Autonomous AI Agents 2025: Latest Developments and Technical + Advancements] \\n* [Recent Developments in Autonomous AI Agents 2025] \\n* * + [Top Technical Advancements Shaping 2025] \\n* * [Fully Autonomous AI Agents: + What's Now Possible in 2025] \\n* [Best Autonomous AI Agents Examples and + Real-World Applications] \\n* [Top Consumer Autonomous AI Agents] \\n* * [Enterprise + and Business Applications] \\n* * [Emerging Application Areas in 2025] \\n* + * [Performance Metrics and Success Stories] \\n* [The Role of Autonomous AI + Agents in Business and Industry Impact] \\n* [How Autonomous AI Agents Will + Impact Industries in 2025] \\n* * [Salesforce Autonomous Agents and CRM Integration] + \\n* * [Autonomous Agents Market Growth and Opportunities] \\n* * [Customer + Service Revolution Through AI Agents] \\n* [How to Build Autonomous AI Agents: + Development and Implementation Guide] \\n* [Essential Steps for Building Autonomous + AI Agents] \\n* * [Best Use Cases for Autonomous AI Agents] \\n* * [AI Agent + Automation for Startups in 2025] \\n* * [Integration with External Tools and + Systems] \\n* * [Development Challenges and Solutions] \\n* [Autonomous AI Agents + vs Traditional Systems: A Comprehensive Comparison] \\n* [Comparison of Autonomous + AI Agents 2025 vs Previous Generations] \\n* * [Most Advanced Autonomous AI + Agents 2025: Market Leaders] \\n* * [Human Workers vs Autonomous AI Agents: + Collaborative Future] \\n* * [Evolution from Reactive to Autonomous Systems] + \\n* [Future of Autonomous AI Agents: Trends and Predictions for 2025 and Beyond] + \\n* [How Autonomous AI Agents Are Shaping the Future] \\n* * [Top Trends in + Autonomous AI Agents 2025] \\n* * [What to Expect from Autonomous AI Agents + in the Future] \\n* * [Autonomous AI Agents in 2025 and Beyond: Technology Roadmap] + \\n* * [Challenges and Opportunities Ahead] \\n* [Geographic Trends and Regional + Variations in Autonomous AI Agent Adoption] \\n* [Factors Influencing Regional + Differences] \\n* * [Comparison of Regional Trends] \\n* * [Regional Market + Opportunities] \\n* [At a Glance: Key Takeaways] \\n* [Frequently Asked Questions] + \\n* [What are autonomous AI agents and how do they differ from regular AI?] + \\n* * [How can autonomous AI agents be used in business in 2025?] \\n* * [What + makes an AI agent truly autonomous?] \\n* * [What are the best examples of autonomous + AI agents available today?] \\n* * [How do I build autonomous AI agents for + my startup?] \\n* [Conclusion:] \\n* [Related Blogs] \\n## Share This Article\\n![Illustration + of an autonomous AI agent symbolizing the advancements and potential of AI agents + in 2025.] ## Introduction\\nAccording to recent research, the global autonomous + AI agents market is projected to reach[$9.9 billion in 2025] and is anticipated + to grow significantly to[$253.3 billion by 2034], registering a strong CAGR + of43.4%during the forecast period. This explosive growth is driven by rapid + enterprise adoption, continuous advancements in artificial intelligence, and + the expansion of automation across diverse industries. North America is expected + to command the largest market share in 2025, holding about 40.7% of the global + market.\\nThis comprehensive guide explores autonomous AI agents’ fundamentals, + applications, and 2025 developments, providing essential insights for businesses, + developers, and decision-makers navigating AI transformation.\\n## What Are + Autonomous AI Agents? Understanding the Fundamentals\\nAutonomous AI agents + are self-governing systems that operate independently without constant human + intervention, making decisions and taking actions to achieve specific goals + using machine learning and environmental awareness.\\n[Autonomous AI agents] + represent a significant leap forward from traditional AI systems. Unlike conventional + artificial intelligence that requires explicit programming for every scenario, + autonomous agents possess the capability to learn, adapt, and make independent + decisions based on their environment and objectives. These systems combine[machine + learning], natural language processing, and real-time data analysis to create + intelligent entities that can operate with minimal human oversight.\\n**For + example:**Learners today can[learn French with Langua’s AI platform], + which uses these same principles to personalize instruction, track progress, + and respond dynamically to the user\u2019s input mirroring how autonomous agents + behave in complex business environments.\\nThe key distinction lies in their + autonomy \u2013the ability to perceive their environment, process information, + make decisions, and execute actions without waiting for human commands. This + independence makes them particularly valuable for businesses seeking to automate + complex processes, improve operational efficiency, and provide consistent service + delivery around the clock.\\n#####\\nSummary: None\\n\\n\\nTitle: Agentic AI\u2019s + strategic ascent: Shifting operations from incremental gains to net-new impact\\nURL: + https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/agentic-ai-operating-model\\nID: + https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/agentic-ai-operating-model\\nScore: + None\\nPublished Date: 2025-10-10T00:00:00.000Z\\nAuthor: Meet the authors\\nImage: + https://www.ibm.com/thought-leadership/institute-business-value/uploads/en/1497_Report_thumbnail_1456x728_fb37c3a274.jpg\\nFavicon: + https://ibm.com/favicon.ico\\nExtras: None\\nSubpages: None\\nText: Agentic + AI\u2019s strategic ascent: Shifting operations from incremental gains to net-new + impact | IBM\\n[Home] [COO, CSCO/Operations and supply chain] Agentic AI\u2019s + strategic ascent\\n[Home] [COO, CSCO/Operations and supply chain] \\n# Agentic + AI\u2019s strategic ascent\\nIn collaboration with Oracle: AI is widening the + gap between organizations that optimize what exists and those that create what\u2019s + next\u2014turning incremental gains into net-new advantage.\\n![Decorative image] + \\nIn collaboration with Oracle: AI is widening the gap between organizations + that optimize what exists and those that create what\u2019s next\u2014turning + incremental gains into net-new advantage.\\n### Report additional content on + the left\\n[Download the insights(will open in a new tab)*![an arrow pointing + to the right] *] \\n[Subscribe for more insights from IBM IBV(will open in a + new tab)*![an arrow pointing to the right] *] \\n#### Key takeaways\\n* More + than three-quarters of executives say most of their AI investment has focused + on improving existing processes rather than developing net-new capabilities.\\n* + Yet, 78% of C-Suiteexecutives say achieving maximum benefit from agentic AI + requires a new operating model.\\n* Companies that excel in three key AI adoption + areas are 32 times more likely to achieve top-tier business performance than + those with minimal implementation.\\n### Rethinking how work works with agentic + AI\\nAs companies continue to explore how to perform better with artificial + intelligence, a cohort of executives are asking a question that goes far beyond + simple deployment: \\\"What if we rebuild our entire operating model around + AI?\\\"\\nA growing number already are\u2014and they're creating business outcomes + that were previously impossible.\\nThese leaders aren't just implementing AI + tools\u2014they're pioneering entirely new operating models around autonomous + decision-making. The radical premise: that agentic AI systems will make an increasing + number of decisions, while humans make the decisions that matter most, preserving + human agency and expertise where it is most crucial.\\nThe performance gap is + widening rapidly. These transformation-driven organizations are creating net-new + business capabilities, fundamentally changing what's possible at enterprise + scale.\\n|||||\\n|By||By||\\n|2027||2030||\\n|twice as many executives expect + AI agents will make autonomous decisions in processes and workflows compared + to today.||fully autonomous robotic systems with embodied AI will be operational + realities across industries.||\\n|||||\\nThe companies winning in the autonomous + AI future won't be those that simply deploy the most sophisticated AI agents. + They'll be the organizations that abandon existing operating models for entirely + new ones designed around autonomous decision-making capabilities.\\nIn the emerging + AI transition, speed matters, but not all efforts are equal. Some organizations + are reaching truly strategic peaks while others chase tactical AI.\\nTo better + understand the real-world impact of autonomous AI, the IBM Institute for Business + Value (IBM IBV) surveyed 800 C-suite executives in 20 countries across 19 industries. + This report reveals key elements of an autonomous AI operating model \u2013from + management mindset to workforce evolution to the role of trust and transparency. + Alongside, we highlight changing team structures, and examples of specific operating + model transformations in progress.\\nWe finish with an action guide that synthesizes + what to do now to build a synergistic operating model that can optimize agentic + AI\u2019s competitive benefits.\\n**Measurable impact: AI is expected to drive + double-digit improvements across the enterprise**\\n![A chart showing results + from IBM\u2019s own AI journey in business operations, as well as forecasted + metrics from non-IBM executives by 2027.] \\n**Perspective**\\nAgentic AI\u2019s + split screen\\nIn this recent survey of C-suite operational executives, we uncovered + a fascinating divergence in how companies are approaching agentic AI. We asked + two very telling questions, which revealed a sharp divide in strategic focus.\\nFirst, + we asked what percentage of their agentic AI initiatives were focused on improving + existing processes and how successful they were.\\nSecond, we asked what percentage + of their agentic AI initiatives were designed to create entirely new workflow + capabilities and how successful they were.\\nThe responses divided our respondents + into two distinct groups.\\n**The process-focused**\\nThis group is technically + getting it done; they're laser-focused on optimizing existing workflows and + are achieving measurable business impact. In short, they are proficient at efficiency, + but they have yet to crack the code on true transformation.\\n**The transformation-driven**\\nThis + more strategic cohort is advancing agentic AI with a dual mandate. They are + not only improving existing workflows, but they are also successfully creating + and building net-new capabilities. Their vision extends beyond optimization; + they are actively reimagining their operating model. The data shows it's working. + They are achieving real-world value of AI and automation across their business + operations.\\n### AI transformation isn\u2019t a forecast, it\u2019s a deadline\\nToday, + 24% of executives say that AI agents take independent action in their organization. + By 2027, 67% expect that to be the case. Similarly, twice as many executives + expect autonomous decision-making from agentic AI in processes and workflows + by 2027 (57%) compared to today (28%). Even in highly regulated areas, transformation-driven + organizations are moving fast \u2014executives expect agentic AI to automate + an average of 29% of risk and compliance operations by 2027.And many anticipate + that AI agents will automate the innovation process itself, with 19% of executives + saying that is happening today and 48% of executives expecting this level of + innovation-related automation in 2027.\\nThese benchmarks represent a fundamental + change in the DNA of business. It\u2019s no wonder that:\\n> 78% of our respondents + agree that achieving maximum benefit from agentic AI requires a new operating + model.\\nAmid global economic uncertainty, 69% of them report a pressing need + to develop more agentic AI-enabled predictive and simulation modeling capabilities\u2014for + example, for financial forecasting, human resource talent and skilling analysis, + dynamic transportation and distribution routing and scheduling, and proactive + mitigation strategies with built-in resilience.\\nCustomer-touching operations + have emerged as primary arenas for agentic AI implementation. Organizations + are prioritizing sales forecasting, dynamic pricing based on inventory scenarios, + and intelligent customer order processing\u2014areas where autonomous system + intelligence can deliver immediate competitive advantages and measurable returns.\\nThe + appetite for change is real, but a critical chasm exists between aspiration + and execution. The vast majority of AI investment\u201478%, has been funneled + into improving existing processes. This is the difference between optimization + and transformation. Process-focused organizations are stuck in a loop of making + their current operations better, rather than making transformative leaps a reality.\\n#### + How decision-makers think differently about agentic AI\\nThe organizations achieving + genuine transformation are those that view agentic AI not as a tool to do existing + work faster but as a catalyst to do entirely new work efficiently. Consider + the compliance challenge: in highly regulated industries, agentic AI isn\u2019t + just flagging risks \u2014it\u2019s using teams of specialized agents to continuously + scan, interpret, and act on regulatory changes across multiple jurisdictions + and industries. Imagine one set of agents monitoring shifting financial rules + in Europe, another tracking evolving trade sanctions in Asia, and yet another + adjusting healthcare compliance workflows in the U.S. All of them work in concert, + automatically updating communications, documentation, and approval processes + in real time. The result? Risks can be mitigated before they materialize, compliance + teams spend less time on repetitive reviews, and organizations can operate more + confidently across geographies without drowning in legal complexity.\\n> Because + they are focused on developing net-new capabilities, transformative organizations + are asking different questions: What becomes possible when systems can make + decisions autonomously?\\nAs well as: How do we redesign our value creation + processes around this capability? These enterprises must also recognize the + shifting landscape AI commoditization is driving. In other words, the window + for first-mover advantages in process innovations is narrowing. Therefore, while + investing in current operations isn't to be underestimated, the true key to + competitive differentiation now lies in bold, transformative leaps forward.\\nThe + measurement gap reveals this strategic divide clearly. Only 42% of the process-oriented + organizations identified in our analysis have developed new key performance + indicators (KPIs) to monitor AI agents' impact on business targets, compared + to nearly half of transformational organizations (see Action Guide for examples + of new KPIs). These leaders aren't just implementing agentic AI automation\u2014they're + measuring different outcomes entirely.\\n> \u201CWhen you don\u2019t have good, + objective KPIs, older ones take over. We see that technology is often deployed + around creating improvements with the infrastructure you currently have.\u201D \\n**> + Board advisor,\\n**>  \\n> United Kingdom\\n**Perspective**\\nPlug in, + level up: The agentic AI marketplace\\nAn agentic AI marketplace is a digital + marketplace where, instead of apps or software licenses, you\u2019re browsing + virtual shelves of autonomous problem\u2011solvers. Each \u201Cproduct\u201D + is an agentic AI \u2014a specialized algorithm that can take action, make decisions, + and adapt to changing conditions without waiting for human approval.\\nThese + marketplaces are more than procurement\\nSummary: None\\n\\n\\nTitle: The rise + of agentic AI: A new era in workplace productivity\\nURL: https://www.imd.org/news/artificial-intelligence/the-rise-of-agentic-ai-a-new-era-in-workplace-productivity/\\nID: + https://www.imd.org/news/artificial-intelligence/the-rise-of-agentic-ai-a-new-era-in-workplace-productivity/\\nScore: + None\\nPublished Date: 2025-09-25T00:00:00.000Z\\nAuthor: None\\nImage: https://www.imd.org/wp-content/uploads/2022/11/IMD-Logo-Blue-on-White-safe-space-RGB-1200x630-1.jpg\\nFavicon: + https://www.imd.org/wp-content/uploads/2022/10/cropped-IMD-Logo-Blue-on-transparent-Square-for-Circle-Crop-RGB-32x32.png\\nExtras: + None\\nSubpages: None\\nText: The rise of agentic AI: A new era in workplace + productivity - IMD business school for management and leadership courses\\n[Close] + \\n[Home] [All News] [back to Home] The rise of agentic AI: A new era in workplace + productivity[back to All News] \\nNews Stories · Artificial Intelligence + - Innovation - Communication - Knowledge Management\\n# The rise of agentic + AI: A new era in workplace productivity\\nTONOMUS Global Center for Digital + and AI Transformation\\nBy Jialu Shan and[Michael R. Wade] \\n**September 2025\\nIn + December 2023, we launched our first \u201C*Gen AI for Better Productivity*\u201D + framework with 12 categories. At the time, the conversation was dominated by + familiar use cases: chatbots that could handle customer queries, tools that + could generate content at scale, image generation, and systems capable of accelerating + code development. Soon after, we added new categories for transcription, translation + and localization, capabilities that quickly proved their utility in everyday + workflows. We have continued to update the framework with new tools.\\nFor this + update, we\u2019re adding a pivotal new category:**Agentic AI**. This addition + signals that business and technology leaders now see autonomous AI agents as + a distinct productivity paradigm: a transformative shift in how we think about + AI in the workplace.\\nBefore turning to this new frontier, it\u2019s worth + pausing to look at the AI applications that have already proven their value. + These tools have moved well beyond novelty and are now woven into daily workflows.\\n![ + - IMD Business School] \\n#### Continuity in core capabilities\\nAI has changed + how we work, and the tools we use are here to stay. What began as experiments + are now indispensable parts of our routines.\\n###### **Generative content creation:**\\nAI + tools that create and edit text, images, presentations, and videos have become + a huge part of our work lives. A[Microsoft survey] found 75% of global knowledge + workers already use generative AI at work. Most users say AI saves them time + (90%), helps them focus on more important tasks (85%), and makes them more creative + (84%). We’ve come a long way from worrying that AI would stifle our creativity. + In fact, a[2024 marketing report] found that 15% of marketing teams “couldn’t + live without AI” because it made everyday tasks like creating presentations + and transcribing audio so much faster and smoother.\\n###### **Knowledge work + and communication:**\\nAI is also transforming how we handle information and + communication. Companies now use advanced chatbots for customer service and + internal helpdesks, dramatically improving response times. AI-powered research + assistants can sift through documents and summarize key findings in seconds, + tasks that once took humans hours of painstaking work. The impact on software + development has been equally transformative. A study by[MIT Sloan, Microsoft + Research, and GitHub] found that generative AI coding tools can reduce programming + time by 56%, allowing developers to focus on higher-level problem-solving rather + than routine implementation.\\n###### **Task automation:**\\nDaily tasks like + managing email and scheduling meetings are now so much easier thanks to AI. + Modern email assistants can sort your inbox and draft responses, and AI schedulers + can find optimal meeting times without endless back-and-forth among participants. + These tools have begun to seriously alleviate our digital overload. A[McKinsey + report] estimated work automation with GenAI and other technologies could boost + productivity growth 0.5 to 3.4% annually.\\nThe foundational uses of AI for + productivity have proven resilient and enduring. However, the competitive landscape + continues to evolve rapidly, with some companies rebranding (you\u2019ll notice + the fresh logos in our graph) or pivoting their focus as the market matures. + Take Tome: once a specialized presentation tool, it has shifted toward building + enterprise solutions, reflecting shifting market dynamics and heightened competition + in an increasingly crowded field.\\n#### **The new frontier: Agentic AI**\\nIf + generative AI is about extending human capability, Agentic AI is about autonomous + execution. Agentic AI refers to systems that can delegate, act, and learn on + their own to accomplish a goal. Instead of waiting for precise, step-by-step + prompts, an AI agent can take a high-level request like, \u201COrganize a one-day + offsite for my team next month,\u201D and then navigate multiple tasks and tools + to fulfil it. Agentic AI tools today can check calendars, research options, + book venues, schedule travel, and draft agenda.\\n[Over a quarter of business + leaders] say their organizations are already exploring agentic AI to a significant + degree. The vision is for these agents to process different types of data (text, + images, voice), coordinate with other AI services, and learn from experience + to reliably execute complex tasks. Emerging platforms show the breadth of possibilities:**Manus**is + a general AI agent that bridges thoughts into actions, designed to independently + carry out complex real-world tasks without direct or continuous human guidance. + Similarly,**Beam**focuses on simplifying data workflows and**Sana**‘s + AI agents are revolutionizing enterprise search and knowledge management.**Stack + AI**allows teams to stitch together modular agents that orchestrate multiple + AI services and traditional software tools.**Devin**positions itself as an \u201CAI + software engineer\u201D that can independently write, test, and ship code. While + not included in the agentic category, agentic modes of the frontier models, + like**ChatGPT**and**Claude**are also starting to make waves.\\nThis doesn\u2019t + mean companies are relinquishing full control. Early trials often keep a human + “in the loop” to supervise. But the momentum is undeniable. Tech + strategists now talk about an “AI workforce”: a concept that emphasizes + augmentation, not replacement: AI agents as teammates rather than substitutes. + For forward-looking firms, this means new opportunities to streamline operations, + from an AI agent that triages IT support tickets overnight to one that summarizes + market intelligence for a strategy team.\\n#### **The bigger picture**\\nAI + has quickly moved from a novelty to a necessity in workplaces worldwide. The[same + Microsoft survey] mentioned earlier found that 79% of business leaders say their + company must adopt AI to stay competitive. This represents a generational consensus + that AI has transitioned from “nice-to-have” to “must-have” + status.\\nThe goal across all these domains is not simply to do more work. It’s + to free up our time and mental energy for what truly matters: the strategic, + creative, and interpersonal aspects of work and life that AI cannot as easily + replace. In just one year, AI’s role in productivity has grown from a + set of practical tools to a broader ecosystem that includes emerging “colleague” + agents. Yet, the mission remains consistent: to reduce friction and enable people + to focus on what they do best.\\nThe AI landscape will undoubtedly keep evolving, + but the lesson is already clear: the toolbox is expanding, and those who adopt + thoughtfully will move not just with the tide, but ahead of it.\\nShare\\n![Share] + \\nGet shareable link\\nCopy\\n[] \\n[![Twitter]] \\n[![Facebook]] \\nRelated + program\\n[![ - IMD Business School] \\nDigital and AI Transformation Programs\\nDigital + Transformation & AI\\nIMD offers today a wide range of programs to help + you seize the opportunities of the new digital environment. And we will be launching + more in the near future! See all options available to you.\\n]\\nSummary: None\\n\\n\\nTitle: + Reframing Operations with AI and Autonomous Agents: A Qualitative Content Analysis + and Adoption Framework\\nURL: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5233324\\nID: + https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5233324\\nScore: None\\nPublished + Date: 2025-04-20T00:00:00.000Z\\nAuthor: See all articles by Mohamed Abdelsalam + Mohamed Mostafa\\nImage: None\\nFavicon: None\\nExtras: None\\nSubpages: None\\nText: + [Skip to main content] \\n\\n[Download This Paper] \\n\\n[Open PDF in Browser] + \\n\\n[Add Paper to My Library] \\n\\nShare:\\n\\nPermalink\\n\\nUsing these + links will ensure access to this page indefinitely\\n\\n[Copy URL] \\n\\n[Copy + DOI] \\n\\n# Reframing Operations with AI and Autonomous Agents: A Qualitative + Content Analysis and Adoption Framework\\n\\n23 PagesPosted: 7 May 2025\\n\\n[See + all articles by Mohamed Abdelsalam Mohamed Mostafa] \\n\\n## [Mohamed Abdelsalam + Mohamed Mostafa] \\n\\nAin Shams University\\n\\nDate Written: April 20, 2025\\n\\n### + Abstract\\n\\nThis qualitative content analysis examines how organizations can + strategically integrate Artificial Intelligence (AI) and autonomous agents to + enhance operational efficiency, ensure legal and ethical compliance, and foster + workforce adaptability. Drawing on recent industry reports (e.g., Fujitsu, 2025; + McKinsey & Company, 2024), academic studies (e.g., Giorgetti et al., 2025), + and regulatory frameworks (e.g., NIST, 2023), we document significant operational + gains attributable to AI-including cycle-time reductions of 30-85%, error-rate + drops to below 1%, and anomalydetection accuracy reaching up to 99%. Despite + these compelling efficiency improvements, persistent challenges hinder widespread + and responsible adoption. These include critical governance gaps under evolving + regulations like GDPR and CCPA, the pervasive risks of automation bias, the + potential for workforce deskilling hazards, and inherent technical barriers + such as the explainability of \\\"black-box\\\" AI models and the significant + environmental impact related to AI's carbon footprint. To address these multifaceted + challenges, this paper proposes a practical Four-Stage Adoption Framework and + a detailed Five-Step Action Plan specifically designed for practitioners. The + findings underscore the necessity of achieving a delicate balance between pursuing + operational efficiency gains and ensuring the sustainability and inclusivity + of AI integration efforts within organizations.\\n\\n**Keywords:** Artificial + Intelligence, Autonomous Agents, Operational Efficiency, AI Governance, Workforce + Dynamics, Qualitative Content Analysis, Adoption Framework, Legal Implications, + Ethical Considerations, Sustainability\\n\\n**Suggested Citation:** [Suggested + Citation] \\n\\nAbdelsalam Mohamed Mostafa, Mohamed, Reframing Operations with + AI and Autonomous Agents: A Qualitative Content Analysis and Adoption Framework + (April 20, 2025). Available at SSRN: [https://ssrn.com/abstract=5233324] or + [http://dx.doi.org/10.2139/ssrn.5233324] \\n\\n### [Mohamed Abdelsalam Mohamed + Mostafa (Contact Author)] \\n\\n#### Ain Shams University ( [email] )\\n\\nCairoEgypt\\n\\n## + Do you have a job opening that you would like to promote on SSRN?\\n\\n[Place + Job Opening] \\n\\n## Paper statistics\\n\\nDownloads\\n\\n28\\n\\nAbstract + Views\\n\\n143\\n\\n[51 References] \\n\\nPlumX Metrics\\n\\n## Related eJournals\\n\\n- + [Operations Management eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Operations Management eJournal\\n\\n\\n\\nSubscribe to this fee journal for + more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n760\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n4,162\\n\\n- + [Operations Strategy eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### Operations + Strategy eJournal\\n\\n\\n\\nSubscribe to this fee journal for more curated + articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n743\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n1,662\\n\\n- + [Technology, Operations Management & Production eJournal] \\n\\n\\n\\n[Follow] + \\n\\n\\n\\n\\n\\n#### Technology, Operations Management & Production eJournal\\n\\n\\n\\nSubscribe + to this fee journal for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n733\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n2,228\\n\\n- + [Artificial Intelligence eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Artificial Intelligence eJournal\\n\\n\\n\\nSubscribe to this fee journal for + more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n373\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n25,390\\n\\n- + [Management of Innovation eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Management of Innovation eJournal\\n\\n\\n\\nSubscribe to this fee journal for + more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n274\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n7,878\\n\\n- + [Innovation & Management Science eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Innovation & Management Science eJournal\\n\\n\\n\\nSubscribe to this fee journal + for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n259\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n8,447\\n\\n- + [Power, Politics, & Organizational Behavior eJournal] \\n\\n\\n\\n[Follow] \\n\\n\\n\\n\\n\\n#### + Power, Politics, & Organizational Behavior eJournal\\n\\n\\n\\nSubscribe to + this fee journal for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n254\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n1,436\\n\\n\\n\\n\\n\\n\\n\\n\\n\\nThis + Journal is curated by:\\n\\n\\n\\n**David C. Logan** at California Medical Association + (CMA) WellnessUniversity of Southern California - Marshall School of Business\\n\\n- + [Strategy Models for Firm Performance Enhancement eJournal] \\n\\n\\n\\n[Follow] + \\n\\n\\n\\n\\n\\n#### Strategy Models for Firm Performance Enhancement eJournal\\n\\n\\n\\nSubscribe + to this fee journal for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n241\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n1,699\\n\\n- + [Emerging Research within Organizational Behavior eJournal] \\n\\n\\n\\n[Follow] + \\n\\n\\n\\n\\n\\n#### Emerging Research within Organizational Behavior eJournal\\n\\n\\n\\nSubscribe + to this fee journal for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n236\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n989\\n\\n\\n\\n\\n\\n\\n\\n\\n\\nThis + Journal is curated by:\\n\\n\\n\\n**David C. Logan** at California Medical Association + (CMA) WellnessUniversity of Southern California - Marshall School of Business\\n\\n- + [International Political Economy: Globalization eJournal] \\n\\n\\n\\n[Follow] + \\n\\n\\n\\n\\n\\n#### International Political Economy: Globalization eJournal\\n\\n\\n\\nSubscribe + to this fee journal for more curated articles on this topic\\n\\n\\n\\n\\n\\n\\n\\nFOLLOWERS\\n\\n\\n\\n84\\n\\n\\n\\n\\n\\n\\n\\nPAPERS\\n\\n\\n\\n9,667\\n\\n\\nFeedback\\n\\nFeedback + to SSRN\\n\\nFeedback\_(required)\\n\\nEmail\_(required)\\n\\nSubmit\\nSummary: + None\\n\\n\\nTitle: AI in the workplace: A report for 2025\\nURL: https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work\\nID: + https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work\\nScore: + None\\nPublished Date: 2025-01-28T00:00:00.000Z\\nAuthor: None\\nImage: https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/superagency%20in%20the%20workplace%20empowering%20people%20to%20unlock%20ais%20full%20potential%20at%20work/superagency-report-1309018746-thumb-1536x1536.jpg\\nFavicon: + https://www.mckinsey.com/favicon.ico\\nExtras: None\\nSubpages: None\\nText: + AI in the workplace: A report for 2025 | McKinsey\\n[Skip to main content] \\n# + Superagency in the workplace: Empowering people to unlock AI’s full potential\\nJanuary + 28, 2025| Report\\nHannah MayerLareina Yee[Michael Chui] [Roger Roberts] \\nAlmost + all companies invest in AI, but just 1 percent believe they are at maturity. + Our research finds the biggest barrier to scaling is not employees—who + are ready—but leaders, who are not steering fast enough.\\n### ![] \\n##### + Superagency in the workplace: Empowering people to unlock AI\u2019s full potential\\n[(47 + pages)] \\n**Artificial intelligence has arrived in the workplace**and has the + potential to be as transformative as the steam engine was to the 19th-century + Industrial Revolution.1\u201C[Gen AI: A cognitive industrial revolution],\u201D + McKinsey, June 7, 2024.With powerful and capable large language models (LLMs) + developed by Anthropic, Cohere, Google, Meta, Mistral, OpenAI, and others, we + have entered a new information technology era. McKinsey research sizes the long-term + AI opportunity at $4.4 trillion in added productivity growth potential from + corporate use cases.2\u201C[The economic potential of generative AI: The next + productivity frontier],\u201D McKinsey, June 14, 2023.\\nTherein lies the challenge: + the long-term potential of AI is great, but the short-term returns are unclear. + Over the next three years, 92 percent of companies plan to increase their AI + investments. But while nearly all companies are investing in AI, only 1 percent + of leaders call their companies \u201Cmature\u201D on the deployment spectrum, + meaning that AI is fully integrated into workflows and drives substantial business + outcomes. The big question is how business leaders can deploy capital and steer + their organizations closer to AI maturity.\\nThis research report, prompted + by Reid Hoffman\u2019s book*Superagency: What Could Possibly Go Right with Our + AI Future*,3Reid Hoffman and Greg Beato,*Superagency: What Could Possibly Go + Right with Our AI Future*, Authors Equity, January 2025.asks a similar question: + How can companies harness AI to amplify human agency and unlock new levels of + creativity and productivity in the workplace? AI could drive enormous positive + and disruptive change. This transformation will take some time, but leaders + must not be dissuaded. Instead, they must advance boldly today to avoid becoming + uncompetitive tomorrow. The history of major economic and technological shifts + shows that such moments can define the rise and fall of companies. Over 40 years + ago, the internet was born. Since then, companies including Alphabet, Amazon, + Apple, Meta, and Microsoft have attained trillion-dollar market capitalizations. + Even more profoundly, the internet changed the anatomy of work and access to + information. AI now is like the internet many years ago: The risk for business + leaders is not thinking too big, but rather too small.\\n![] \\nThis report + explores companies\u2019 technology and business readiness for AI adoption (see + sidebar \u201CAbout the survey\u201D). It concludes that employees are ready + for AI. The biggest barrier to success is leadership.\\n## About the survey\\n**To + create our report,**we surveyed 3,613 employees (managers and independent contributors) + and 238 C-level executives in October and November 2024. Of these, 81 percent + came from the United States, and the rest came from five other countries: Australia, + India, New Zealand, Singapore, and the United Kingdom. The employees spanned + many roles, including business development, finance, marketing, product management, + sales, and technology.\\nAll the survey findings discussed in the report, aside + from two sidebars presenting international nuances, pertain solely to US workplaces. + The findings are organized in this way because the responses from US employees + and C-suite executives provide statistically significant conclusions about the + US workplace. Analyzing global findings separately allows a comparison of differences + between US responses and those from other regions.\\n**Chapter 1**looks at the + rapid advancement of technology over the past two years and its implications + for business adoption of AI.\\n**Chapter 2**delves into the attitudes and perceptions + of employees and leaders. Our research shows that employees are more ready for + AI than their leaders imagine. In fact, they are already using AI on a regular + basis; are three times more likely than leaders realize to believe that AI will + replace 30 percent of their work in the next year; and are eager to gain AI + skills. Still, AI optimists are only a slight majority in the workplace; a large + minority (41 percent) are more apprehensive and will need additional support. + This is where millennials, who are the most familiar with AI and are often in + managerial roles, can be strong advocates for change.\\n![Agents, Robots, and + Us] \\n### Agents, Robots, and Us: What executives need to know about AI and + work\\n**Thursday, February 26, 10:30 a.m. - 11:00 a.m. ET / 4:30 p.m. - 5:00 + p.m. CET**\\nJoin McKinsey senior partners Alexis Krivkovich and Anu Madgavkar + for a discussion of how executives can build effective human-AI skill partnerships, + prepare their workforces, and make strategic decisions that position their organizations + for growth in the age of AI.\\n[Register here] \\n**Chapter 3**looks at the + need for speed and safety in AI deployment. While leaders and employees want + to move faster, trust and safety are top concerns. About half of employees worry + about AI inaccuracy and cybersecurity risks. That said, employees express greater + confidence that their own companies, versus other organizations, will get AI + right. The onus is on business leaders to prove them right, by making bold and + responsible decisions.\\n**Chapter 4**examines how companies risk losing ground + in the AI race if leaders do not set bold goals. As the hype around AI subsides, + companies should put a heightened focus on practical applications that empower + employees in their daily jobs. These applications can create competitive moats + and generate measurable ROI. Across industries, functions, and geographies, + companies that invest strategically can go beyond using AI to drive incremental + value and instead create transformative change.\\n**Chapter 5**looks at what + is required for leaders to set their teams up for success with AI. The challenge + of AI in the workplace is not a technology challenge. It is a business challenge + that calls upon leaders to align teams, address AI headwinds, and rewire their + companies for change.\\nChapter 1\\n## An innovation as powerful as the steam + engine\\nImagine a world where machines not only perform physical labor but + also think, learn, and make autonomous decisions. This world includes humans + in the loop, bringing people and machines together in a state of superagency + that increases personal productivity and creativity (see sidebar \u201CAI superagency\u201D). + This is the transformative potential of AI, a technology with a potential impact + poised to surpass even the biggest innovations of the past, from the printing + press to the automobile. AI does not just automate tasks but goes further by + automating cognitive functions. Unlike any invention before, AI-powered software + can adapt, plan, guide\u2014and even make\u2014decisions. That\u2019s why AI + can be a catalyst for unprecedented economic growth and societal change in virtually + every aspect of life. It will reshape our interaction with technology and with + one another.\\n> Scientific discoveries and technological innovations are stones + in the cathedral of human progress.\\n> > Reid Hoffman, cofounder of LinkedIn + and Inflection AI, partner at Greylock Partners, and author\\nMany breakthrough + technologies, including the internet, smartphones, and cloud computing, have + transformed the way we live and work. AI stands out from these inventions because + it offers more than access to information. It can summarize, code, reason, engage + in a dialogue, and make choices. AI can lower skill barriers, helping more people + acquire proficiency in more fields, in any language and at any time. AI holds + the potential to shift the way people access and use knowledge. The result will + be more efficient and effective problem solving, enabling innovation that benefits + everyone.\\n## AI superagency\\n**What impact will AI have on humanity?**Reid + Hoffman and Greg Beato\u2019s book*Superagency: What Could Possibly Go Right + with Our AI Future*(Authors Equity, January 2025) explores this question. The + book highlights how AI could enhance human agency and heighten our potential. + It envisions a human-led, future-forward approach to AI.\\nSuperagency, a term + coined by Hoffman, describes a state where individuals, empowered by AI, supercharge + their creativity, productivity, and positive impact. Even those not directly + engaging with AI can benefit from its broader effects on knowledge, efficiency, + and innovation.\\nAI is the latest in a series of transformative supertools, + including the steam engine, internet, and smartphone, that have reshaped our + world by amplifying human capabilities. Like its predecessors, AI can democratize + access to knowledge and automate tasks, assuming humans can develop and deploy + it safely and equitably.\\nOver the past two years, AI has advanced in leaps + and bounds, and enterprise-level adoption has accelerated due to lower costs + and greater access to capabilities. Many notable AI innovations have emerged + (Exhibit 1). For example, we have seen a rapid expansion of context windows, + or the short-term memory of LLMs. The larger a[context window], the more information + an LLM can process at once. To illustrate, Google\u2019s Gemini 1.5 could process + one million tokens in February 2024, while its Gemini 1.5 Pro could process + two million tokens by June of that same year.4*The Keyword*,\_\u201COur next-generation + model:\_Gemini 1.5,\u201D blog entry by Sundar Pichai and Demis Hassabis, Google, + February 15, 2024;*Google for Developers*, \u201CGemini 1.5 Pro 2M context window,\\nSummary: + None\\n\\nResolved Search Type: neural\\nCostDollars: total=0.015\\n - search: + {'neural': 0.005}\\n - contents: {'text': 0.01}\"},{\"role\":\"tool\",\"tool_call_id\":\"call_QoeOUYfW09KlW4WYa0jHps9N\",\"name\":\"exa_search_tool\",\"content\":\"Title: + The Challenges of Governing AI Agents\\nURL: https://ai-frontiers.org/articles/the-challenges-of-governing-ai-agents\\nID: + https://ai-frontiers.org/articles/the-challenges-of-governing-ai-agents\\nScore: + None\\nPublished Date: 2025-03-03T00:00:00.000Z\\nAuthor: None\\nImage: https://cdn.prod.website-files.com/66e38815e6bc34b7f5e1768b/67cb4f24080d3501fc61b03d_google-deepmind-BMn5Z-MGv5c-unsplash.jpg\\nFavicon: + https://cdn.prod.website-files.com/66e38815e6bc34b7f5e17684/67f371029f796428b28d25cf_aif-32.png\\nExtras: + None\\nSubpages: None\\nText: The Challenges of Governing AI Agents | AI Frontiers\\n![] + \\nThank you! Your submission has been received!\\nOops! Something went wrong + while submitting the form.\\n[\\n![] \\nHigh-Bandwidth Memory: The Critical + Gaps in US Export Controls\\nErich Grunewald\\n\u2014Modern memory architecture + is vital for advanced AI systems. While the US leads in both production and + innovation, significant gaps in export policy are helping China catch up.\\nhigh-bandwidth + memory, HBM, DRAM, AI chips, GPU packaging, export controls, Bureau of Industry + and Security, BIS, U.S.-China tech competition, semiconductor manufacturing + equipment, FDPR, ASML immersion DUV lithography, SK Hynix, Samsung, Micron\\n] + \\n[\\n![] \\nMaking Extreme AI Risk Tradeable\\nDaniel Reti\\n\u2014Traditional + insurance can\u2019t handle the extreme risks of frontier AI. Catastrophe bonds + can cover the gap and compel labs to adopt tougher safety standards.\\nfrontier + AI, extreme AI risk, catastrophic AI events, AI liability, liability insurance, + catastrophe bonds, cat bonds, insurance-linked securities, capital markets, + AI regulation, AI safety standards, third-party audits, catastrophic risk index, + tail risk, systemic risk\\n] \\n[\\n![] \\nExporting Advanced Chips Is Good + for Nvidia, Not the US\\nLaura Hiscott\\n\u2014The White House is betting that + hardware sales will buy software loyalty \u2014a strategy borrowed from 5G that + misunderstands how AI actually works.\\n] \\n[\\n![] \\nAI Could Undermine Emerging + Economies\\nDeric Cheng\\n\u2014AI automation threatens to erode the \u201Cdevelopment + ladder,\u201D a foundational economic pathway that has lifted hundreds of millions + out of poverty.\\n] \\n[\\n![] \\nThe Evidence for AI Consciousness, Today\\nCameron + Berg\\n\u2014A growing body of evidence means it\u2019s no longer tenable to + dismiss the possibility that frontier AIs are conscious.\\n] \\n[\\n![] \\nAI + Alignment Cannot Be Top-Down\\nAudrey Tang\\n\u2014Community Notes offers a + better model \u2014where citizens, not corporations, decide what \u201Caligned\u201D + means.\\nAI alignment, attentiveness, Community Notes, Taiwan, Audrey Tang, + model specification, deliberative governance, epistemic security, portability + and interoperability, market design, Polis, reinforcement learning from community + feedback, social media moderation, civic technology\\n] \\n[\\n![] \\nAGI's + Last Bottlenecks\\nAdam Khoja\\n\u2014A new framework suggests we\u2019re already + halfway to AGI. The rest of the way will mostly require business-as-usual research + and engineering.\\nAGI, artificial general intelligence, AGI definition, GPT-5, + GPT-4, visual reasoning, world modeling, continual learning, long-term memory, + hallucinations, SimpleQA, SPACE benchmark, IntPhys 2, ARC-AGI, working memory\\n] + \\n[\\n![] \\nAI Will Be Your Personal Political Proxy\\nBruce Schneier\\n\u2014By + learning our views and engaging on our behalf, AI could make government more + representative and responsive \u2014but not if we allow it to erode our democratic + instincts.\\nAI political proxy, direct democracy, generative social choice, + ballot initiatives, voter participation, democratic representation, AI governance, + Rewiring Democracy, Bruce Schneier, Nathan E. Sanders, policy automation, civic + engagement, rights of nature, disenfranchised voters, algorithmic policymaking\\n] + \\n[\\n![] \\nIs China Serious About AI Safety?\\nKarson Elmgren\\n\u2014China\u2019s + new AI safety body brings together leading experts \u2014but faces obstacles + to turning ambition into influence.\\nChina AI Safety and Development Association, + CnAISDA, China AI safety, World AI Conference, Shanghai AI Lab, Frontier AI + risk, AI governance, international cooperation, Tsinghua University, CAICT, + BAAI, Global AI Governance Action Plan, AI Seoul Summit commitments, Concordia + AI, Entity List\\n] \\n[\\n![] \\nAI Deterrence Is Our Best Option\\nDan Hendrycks\\n\u2014A + response to critiques of Mutually Assured AI Malfunction (MAIM).\\nAI deterrence, + Mutually Assured AI Malfunction, MAIM, Superintelligence Strategy, ASI, intelligence + recursion, nuclear MAD comparison, escalation ladders, verification and transparency, + redlines, national security, sabotage of AI projects, deterrence framework, + Dan Hendrycks, Adam Khoja\\n] \\n[\\n![] \\nSummary of \u201CIf Anyone Builds + It, Everyone Dies\u201D\\nLaura Hiscott\\n\u2014An overview of the core arguments + in Yudkowsky and Soares\u2019s new book.\\nIf Anyone Builds It Everyone Dies, + Eliezer Yudkowsky, Nate Soares, MIRI, AI safety, AI alignment, artificial general + intelligence, artificial superintelligence, AI existential risk, Anthropic deceptive + alignment, OpenAI o1, Truth\\\\_Terminal, AI moratorium, book summary, Laura + Hiscott\\n] \\n[\\n![] \\nAI Agents Are Eroding the Foundations of Cybersecurity\\nRosario + Mastrogiacomo\\n\u2014In this age of intelligent threats, cybersecurity professionals + stand as the last line of defense. Their decisions shape how humanity contends + with autonomous systems.\\nAI agents, AI identities, cybersecurity, identity + governance, zero trust, least privilege, rogue AI, autonomous systems, enterprise + security, trust networks, authentication and authorization, RAISE framework, + identity security, circuit breakers\\n] \\n[\\n![] \\nPrecaution Shouldn't + Keep Open-Source AI Behind the Frontier\\nBen Brooks\\n\u2014Invoking speculative + risks to keep our most capable models behind paywalls could create a new form + of digital feudalism.\\nopen-source AI, frontier models, precautionary policy, + digital feudalism, OpenAI, Meta, Llama, GPT-OSS, regulation, open development, + AI risk, legislation, policy debate, Berkman Klein Center\\n] \\n[\\n![] \\nThe + Hidden AI Frontier\\nOscar Delaney\\n\u2014Many cutting-edge AI systems are + confined to private labs. This hidden frontier represents America\u2019s greatest + technological advantage \u2014and a serious, overlooked vulnerability.\\nhidden + frontier AI, internal AI models, AI security, model theft, sabotage, government + oversight, transparency, self-improving AI, AI R&amp;D automation, policy + recommendations, national security, RAND security levels, frontier models, AI + governance, competitive advantage\\n] \\n[\\n![] \\nUncontained AGI Would Replace + Humanity\\nAnthony Aguirre\\n\u2014The moment AGI is widely released \u2014whether + by design or by breach \u2014any guardrails would be as good as gone.\\nAGI, + artificial general intelligence, open-source AI, guardrails, uncontrolled release, + existential risk, humanity replacement, security threat, proliferation, autonomous + systems, alignment, self-improving intelligence, policy, global race, tech companies\\n] + \\n[\\n![] \\nSuperintelligence Deterrence Has an Observability Problem\\nJason + Ross Arnold\\n\u2014Mutual Assured AI Malfunction (MAIM) hinges on nations observing + one another's progress toward superintelligence \u2014but reliable observation + is harder than MAIM's authors acknowledge.\\nMAIM, superintelligence deterrence, + Mutual AI Malfunction, observability problem, US-China AI arms race, compute + chips data centers, strategic sabotage, false positives, false negatives, AI + monitoring, nuclear MAD analogue, superintelligence strategy, distributed R&amp;D, + espionage escalation, peace and security\\n] \\n[\\n![] \\nOpen Protocols Can + Prevent AI Monopolies\\nIsobel Moure\\n\u2014With model performance converging, + user data is the new advantage \u2014and Big Tech is sealing it off.\\nopen + protocols, AI monopolies, Anthropic MCP, context data lock-in, big tech, APIs, + interoperability, data portability, AI market competition, user context, model + commoditization, policy regulation, open banking analogy, enshittification\\n] + \\n[\\n![] \\nIn the Race for AI Supremacy, Can Countries Stay Neutral?\\nAnton + Leicht\\n\u2014The global AI order is still in flux. But when the US and China + figure out their path, they may leave little room for others to define their + own.\\nAI race, US-China competition, middle powers, export controls, AI strategy, + militarization, economic dominance, compute supply, frontier models, securitization, + AI policy, grand strategy, geopolitics, technology diffusion, national security\\n] + \\n[\\n![] \\nHow AI Can Degrade Human Performance in High-Stakes Settings\\nDane + A. Morey\\n\u2014Across disciplines, bad AI predictions have a surprising tendency + to make human experts perform worse.\\nAI, human performance, safety-critical + settings, Joint Activity Testing, human-AI collaboration, AI predictions, aviation + safety, healthcare alarms, nuclear power plant control, algorithmic risk, AI + oversight, cognitive systems engineering, safety frameworks, nurses study, resilient + performance\\n] \\n[\\n![] \\nHow the EU's Code of Practice Advances AI + Safety\\nHenry Papadatos\\n\u2014The Code provides a powerful incentive to push + frontier developers toward measurably safer practices.\\nEU Code of Practice, + AI Act, AI safety, frontier AI models, risk management, systemic risks, 10^25 + FLOPs threshold, external evaluation, transparency requirements, regulatory + compliance, general-purpose models, European Union AI regulation, safety frameworks, + risk modeling, policy enforcement\\n] \\n[\\n![] \\nHow US Export Controls Have + (and Haven't) Curbed Chinese AI\\nChris Miller\\n\u2014Six years of export + restrictions have given the U.S. a commanding lead in key dimensions of the + AI competition \u2014but it\u2019s uncertain if the impact of these controls + will persist.\\nchip, chips, china, chip export controls, China semiconductors, + hardware, AI hardware policy, US technology restrictions, SMIC, Huawei Ascend, + Nvidia H20, AI infrastructure, high-end lithography tools, EUV ban, domestic + chipmaking, AI model development, technology trade, computing hardware, US-China + relations\\n] \\n[\\n![] \\nNuclear Non-Proliferation Is the Wrong Framework + for AI Governance\\nMichael C. Horowitz\\n\u2014Placing AI in a nuclear framework + inflates expectations and distracts from practical, sector-specific governance.\\n] + \\n[\\n![] \\nA Patchwork of State AI Regulation Is Bad. A Moratorium Is Worse.\\nKristin + O\u2019Donoghue\\n\u2014Congress is weighing a measure that would nullify thousands + of state AI rules and bar new ones \u2014upending federalism and halting the + experiments that drive smarter policy.\\nai regulation, state laws, federalism, + congress, policy innovation, legislative measures, state vs federal, ai governance, + legal framework, regulation moratorium, technology policy, experimental policy, + state experimentation,\\nSummary: None\\n\\n\\nTitle: AI Agents in Action: Foundations + for Evaluation and ...\\nURL: https://reports.weforum.org/docs/WEF_AI_Agents_in_Action_Foundations_for_Evaluation_and_Governance_2025.pdf\\nID: + https://reports.weforum.org/docs/WEF_AI_Agents_in_Action_Foundations_for_Evaluation_and_Governance_2025.pdf\\nScore: + None\\nPublished Date: 2025-11-27T00:00:00.000Z\\nAuthor: None\\nImage: None\\nFavicon: + None\\nExtras: None\\nSubpages: None\\nText: AI Agents in Action: \\nFoundations + for Evaluation \\nand Governance\\nWHITE PAPER\\nNOVEMBER 2025\\nIn collaboration + \\nwith Capgemini \\nImages: Adobe Stock\\nDisclaimer\\nThis document is published + by the \\nWorld Economic Forum as a contribution \\nto a project, insight area + or interaction. \\nThe findings, interpretations and \\nconclusions expressed + herein are a result \\nof a collaborative process facilitated and \\nendorsed + by the World Economic Forum \\nbut whose results do not necessarily \\nrepresent + the views of the World Economic \\nForum, nor the entirety of its Members, \\nPartners + or other stakeholders.\\n\xA9 2025 World Economic Forum. All rights \\nreserved. + No part of this publication may \\nbe reproduced or transmitted in any form + \\nor by any means, including photocopying \\nand recording, or by any information + \\nstorage and retrieval system.\\nContents\\nForeword 4\\nExecutive summary + 5\\nIntroduction 6\\n1 Evolving technical foundations of AI agents 8\\n1.1 The + software architecture of an AI agent 8\\n1.2 Communication protocols and interoperability + 10\\n1.3 Cybersecurity considerations 12\\n2 Foundations for AI agent evaluation + and governance 13\\n2.1 Classification 14\\n2.2 Evaluation 19\\n2.3 Risk assessment + 22\\n2.4 Governance considerations for AI agents: a progressive approach 25\\n3 + Looking ahead: multi-agent ecosystems 29\\nConclusion 30\\nContributors 31\\nEndnotes + 34\\nAI Agents in Action: Foundations for Evaluation and Governance 2\\nForeword\\nIn + recent years, organizations have moved beyond \\npredictive models and chat + interfaces to experiment \\nwith artificial intelligence (AI) in more transformative + \\nways. AI agents are now emerging as integrated \\ncollaborators in business, + public services and \\neveryday life. The adoption of AI agents could \\nbring + significant gains in efficiency, altered kinds of \\nhuman-machine interaction + and the advent of novel \\ndigital ecosystems.\\nThis transition faces multiple + obstacles that need to be \\naddressed. Moving from models to agents represents + \\nmore than a technical milestone and requires \\norganizations to rethink + how they design, evaluate and \\ngovern advanced agentic systems. Many companies + \\nare now questioning what agents can accomplish \\nalongside the practical + steps needed to adopt and \\ndeploy them safely, responsibly and effectively.\\nThis + paper was developed to help answer those \\nquestions. By mapping the evolving + foundations \\nof agentic systems, classifying their roles, \\nidentifying new + ways to evaluate them and outlining \\nprogressive governance approaches, the + paper \\noffers practical guidance for leaders navigating \\nadoption in real-world + contexts.\\nThrough the AI Governance Alliance, the World \\nEconomic Forum + and Capgemini are advancing \\nthis subject in collaboration with the AI community, + \\nsignalling that now is the time to prepare for an \\nagentic future. If adopters + start small, iterate \\ncarefully and apply proportionate safeguards, \\nagents + can be deployed in ways that amplify \\nhuman capabilities, unlock productivity + and \\nestablish a foundation for more complex multi-agent \\necosystems to + emerge over time. Unless a careful \\nand deliberate approach to adoption is + adopted, \\nuntested use cases could outpace oversight and \\nlead to misaligned + incentives, emergent risks and \\nloss of public trust.\\nAs with any transformative + technology, the \\nopportunities presented by AI agents must be \\naccompanied + by a responsibility to guide their \\ndevelopment and deployment with care. + Through \\ncross-functional efforts and collaborative \\ngovernance, AI agents + can be integrated in ways \\nthat amplify human ingenuity, promote innovation + \\nand improve overall quality of life. This paper is a \\nstep in that direction, + offering guidance to help early \\nadopters navigate the complex and often uneven + \\npath of AI agent adoption. \\nRoshan Gya\\nChief Executive Officer, \\nCapgemini + Invent\\nCathy Li\\nHead, Centre for AI \\nExcellence, Member of \\nthe Executive + Committee, \\nWorld Economic Forum\\nAI Agents in Action: Foundations \\nfor + Evaluation and Governance\\nNovember 2025\\nAI Agents in Action: Foundations + for Evaluation and Governance 3\\nExecutive summary\\nThis report has been tailored + mainly for adopters \\nof AI agents, including decision-makers, technical \\nleaders + and practitioners seeking to integrate AI \\nagents into organizational workflows + and services. \\nWhile AI agents are gaining traction, there remains \\nlimited + guidance on how to design, test and \\noversee them responsibly. This paper + aims to help \\nfill that gap by providing a structured foundation for \\nthe + safe and effective deployment of these systems.\\nThe paper makes three key + contributions. Firstly, \\nit covers the technical foundations of AI agents, + \\nincluding their architectures, protocols and security \\nconsiderations. + Secondly, it offers a functional \\nclassification that differentiates agents + by their role, \\nautonomy, authority, predictability and operational \\ncontext. + Thirdly, it suggests a progressive governance \\napproach that directly connects + evaluation and \\nsafeguards to an agent\u2019s task scope and \\ndeployment + environment.\\nTogether, these elements guide adopters\_with a \\nconceptual + blueprint for moving from experimentation\\nto deployment. The report highlights + the importance \\nof aligning adoption with evaluation and governance \\npractices + to ensure that AI agents are successfully \\ndeployed while trust, safety and + accountability \\nare maintained.\\nThis paper explores the emergence of AI + \\nagents, outlining their technical foundations, \\nclassification, evaluation + and governance to \\nsupport safe and effective adoption. \\nAI Agents in Action: + Foundations for Evaluation and Governance 4\\nIntroduction\\nAI agents are gradually + becoming embedded in \\nan increasing number of tasks, workflows and \\nuse + cases that span cloud and edge computing, \\nleading the way to more widespread + adoption. \\nAs the transition from prototyping to deployment \\naccelerates, + current adoption remains concentrated \\namong early adopters. According to + a recent global \\nsurvey of executives, 82% of organizations plan to \\nintegrate + agents within the next one to three years, \\nindicating that most efforts are + still in the planning or \\npilot phase,1\\n while moving towards wider adoption.\\nThe + concept of software agents has been studied \\nfor decades in fields such as + robotics, autonomous \\nsystems and distributed computing. What is different\\ntoday + is the rise of data-driven models, particularly \\ngenerative artificial intelligence + (AI) and large language\\nmodels (LLMs), which are enabling the emergence \\nof + a new generation of LLM-based agents. These \\nsystems can generate plans, simulate + reasoning \\nand adapt their behaviour through feedback \\nmechanisms in ways + that were previously not \\npossible. This evolution has sparked a new \\nwave + of experimentation, with researchers and \\ncompanies rapidly creating prototypes + of agents \\nin various fields. This report focuses mainly on \\nLLM-based agents + (\u201CAI agents\u201D is sometimes \\nused in short), whose growing capabilities + create \\nboth significant opportunities for adoption and a \\nnew set of challenges + in governance and safety.\\nAI agents are shifting from prototypes to \\ndeployment, + bringing both transformative \\nopportunities and novel governance challenges.\\nFIGURE + 1 Foundations for the responsible adoption of AI agents\\nFunctional\\nclassification\\n2\\nDefine + the\\nagent\u2019s role\\nEvaluation and\\ngovernance\\n3\\nScale with\\nconfidence\\nTechnical\\nfoundations\\n1\\nLay + the\\ngroundwork\\nAI Agents in Action: Foundations for Evaluation and Governance + 5\\nLLM-based AI agents, for example, introduce new \\nrisks such as goal misalignment, + behavioural drift, \\ntool misuse and emergent coordination failures \\nthat + traditional software governance models are \\nunable to manage. Unlike conventional + software, \\nagents are increasingly assuming roles that \\nresemble those of + human decision-makers rather \\nthan static tools. This means that governance + \\nmodels designed solely for access control and \\nsystem reliability are no + longer sufficient. A more \\nuseful comparison is the governance applied \\nto + human users, who must earn permissions, \\naccountability and trust by demonstrating\_performance\\nover + time. Similarly, trust in AI agents can be \\nestablished by testing their behaviour + against \\nvalidated cases, running them in human-in-the\\u0002loop configurations + and gradually expanding \\nautonomy only once reliability has been sufficiently + \\ndemonstrated. In both cases, the principle of least \\nprivilege remains + essential, with access limited to \\ninformation and actions necessary for the + task.\\nThis report aims to provide\_a\_forward-looking\_analysis \\nof the + evolving landscape\_of\_AI\_agents, focusing \\non the capabilities, infrastructure,\_classification + and \\nsafeguards necessary for responsible deployment.\\nTo\_this end, it is + structured around four foundational \\npillars across classification, evaluation, + risk assessment \\nand governance, which together form the foundation \\nfor + a progressive approach to adoption and \\ndeployment. Figure 1 presents the + general content \\nof this report, which helps guide the responsible \\nadoption + and deployment of AI agents.\\nThe goal is to equip adopters, providers, technical + \\nleaders, organizational decision-makers and other \\nstakeholders with a + shared understanding of the \\ncurrent state of agentic systems and emerging + \\noversight practices. Building on established \\nAI governance principles + and frameworks, \\nsuch as those developed by the Organisation \\nfor Economic + Co-operation and Development \\n(OECD),2\\n National Institute of Standards + and \\nTechnology (NIST),3\\n International Organization for \\nStandardization + (ISO)/International Electrotechnical \\nCommission (IEC)4\\n and others, this + paper \\nintroduces additional principles addressing \\nautonomy, authority, + operational context and \\nsystemic risk that extend existing governance \\nguidance + from an agent-focused lens. The \\ninsights have been informed by working group + \\nmeetings, workshops and extensive interviews with \\nmembers of the Safe + Systems and Technologies \\nworking group of the AI Governance Alliance.\\nAI + Agents in Action: Foundations\\nSummary: None\\n\\n\\nTitle: AI Agent Governance: + Building Trust in the Age of Autonomy\\nURL: https://rierino.com/blog/ai-agent-governance-trust-autonomy\\nID: + https://rierino.com/blog/ai-agent-governance-trust-autonomy\\nScore: None\\nPublished + Date: 2025-10-27T00:00:00.000Z\\nAuthor: None\\nImage: https://res.cloudinary.com/rierino/image/upload/f_auto/media/blog/media/hero/ai-agent-governance-trust-autonomy-hero.png\\nFavicon: + https://rierino.com/favicon.ico\\nExtras: None\\nSubpages: None\\nText: AI Agent + Governance: Building Trust in the Age of Autonomy\\n[![Rierino Logo]![Rierino + Logo]] \\nPlatform\\nSolutions\\nPartners\\nResources\\n[\\nTake a Tour\\n] + \\n[CONTACT US] [GET STARTED] \\n[![Rierino Logo]] [![Menu]] \\n# AI Agent Governance: + Building Trust in the Age of Autonomy\\nOctober 27, 2025\u20229minutes\\n[] + [] [] \\n![AI Agent Governance: Building Trust in the Age of Autonomy] \\nKey + Topics\\nGen AI\\nAI Governance\\nLow Code\\nAgentic AI\\nOrchestration\\nObservability\\nArtificial + intelligence is entering a new phase defined not just by intelligence, but by**autonomy**. + From marketing assistants that launch personalized campaigns to supply chain + bots that reorder inventory on their own, the rise of*agentic AI*marks a turning + point in how software operates. Yet as these systems gain more independence, + the need for governance and guardrails becomes exponentially more critical.\\nAccording + to*McKinsey*,[just 1%] of surveyed organizations believe that their AI adoption + has reached full maturity. While many enterprises are experimenting with autonomous + agents and generative models, few have established the safety, accountability, + and alignment mechanisms required to deploy them responsibly at scale. The challenge + isn\u2019t just about building smarter agents; it\u2019s about ensuring those + agents act within**trusted, observable boundaries**.\\nThe*Institute for AI + Policy and Strategy*\u2019s recent field guide highlights this emerging tension: + as decision-making moves from algorithms to autonomous agents, governance must + evolve from reactive oversight to[proactive orchestration]. In other words, + it\u2019s no longer enough to manage data and models, organizations must now + manage AI behavior.\\nAt Rierino, we\u2019ve seen this shift firsthand through + enterprises scaling their generative and agentic AI initiatives from pilots + to production. One insight stands out: the more powerful and connected AI agents + become, the more essential**governed orchestration**is to balance autonomy with + control.\\nThis article explores what effective AI agent governance looks like + \u2014the risks it mitigates, the layers that make it work, and how**orchestration-first + low-code**architecture enables trust, transparency, and security in the age + of autonomous systems.\\n# Why Governance Matters More for Agentic AI\\nThe + concept of**AI governance**isn\u2019t new \u2014most organizations already apply + policies around data use, model validation, and compliance. But agentic AI changes + the scope of what must be governed. Traditional systems make predictions or + recommendations; agents go a step further by initiating and executing actions + in live environments. That shift from decision to autonomous execution fundamentally + transforms the governance challenge.\\n> > In classical AI, risk management + focuses on **> how\\n**> a model behaves. With agentic systems, it\u2019s equally + about **> what\\n**> the model can do \u2014and under what circumstances.\\n> + Each agent\u2019s capacity to perceive, decide, and act introduces new layers + of operational risk that data ethics frameworks or model fairness audits alone + cannot address. This new paradigm requires oversight across three fronts:\\n* + **Action Control:**ensuring agents can act only within predefined scopes and + policies.\\n* **Inter-Agent Coordination:**managing how multiple agents interact + or collaborate to prevent conflict or duplication.\\n* **Continuous Accountability:**maintaining + full traceability over every decision, event, and action for audit and compliance.\\nAs + enterprises move beyond proof-of-concept deployments into production-scale systems, + the absence of such controls can quickly lead to unintended automation, data + exposure, or decision drift, issues that traditional governance frameworks were + never designed to detect in real time.\\nAgentic AI, in other words, doesn\u2019t + just need smarter governance; it needs**dynamic, execution-aware**governance + that can monitor, interpret, and intervene in the flow of autonomous operations + as they unfold.\\n# Top Governance Challenges for Autonomous Agents\\nDespite + growing interest in agentic AI, most enterprises still underestimate the complexity + of governing autonomy. The very qualities that make agents valuable, independence, + adaptability, and collaboration, also introduce new and often invisible risks. + Below are five of the most critical governance challenges organizations face + as they move from experimentation to large-scale deployment.\\n**1. Unbounded + Autonomy:**Without defined boundaries, autonomous agents may take actions that + exceed their intended authority or cascade across systems. For example, a logistics + optimization agent might reroute shipments to improve efficiency but inadvertently + disrupt existing service-level agreements or inventory priorities. This challenge + arises when autonomy is granted without clearly codified limits. Agents may + operate with good intent but without awareness of strategic, ethical, or operational + constraints.\\n**2. Opaque Decision Flows:**As agents self-adapt or collaborate, + their internal decision logic can become difficult to trace or explain. In practice, + this means a business cannot easily determine why a specific action occurred, + what data triggered it, or whether it aligned with compliance policies. When + decision paths lack transparency, accountability mechanisms fail, leaving enterprises + unable to reconstruct cause-and-effect relationships during audits or incident + analysis.\\n**3. Runtime Drift:**Over time, agents may diverge from their initial + parameters as they learn from evolving data and context. A recommendation agent, + for instance, might start prioritizing outcomes that maximize short-term engagement + over long-term customer value if its reward signals shift subtly. Governance + frameworks must therefore include continuous validation and feedback loops to + detect behavioral drift before it compounds into systemic bias or unintended + automation.\\n**4. Coordination Conflicts:**When multiple agents operate within + shared environments, their independent actions can overlap or contradict one + another. A pricing agent might apply a discount while a revenue management agent + simultaneously adjusts base prices, both rational individually but conflicting + collectively. These conflicts reveal the need for orchestration and dependency + management, ensuring that agents don\u2019t act in isolation but within an agreed-upon + coordination protocol.\\n**5. Auditability and Accountability Gaps:**Traditional + monitoring systems were not designed for real-time, autonomous decision chains. + Without detailed logs of every decision and action, enterprises risk compliance + blind spots, especially in regulated sectors like banking, telecom, or government. + True accountability requires fine-grained observability: a continuous record + of inputs, outputs, and actions that can be queried or rolled back at any point + in time.\\n**6. Resource and Cost Inefficiency:**Unsupervised agents can also + introduce hidden operational costs when allowed to execute actions or queries + without optimization. Excessive API calls, non-strategic task repetition, or + uncontrolled use of large language model tokens can quickly inflate spend without + delivering proportional value. For instance, a content generation agent might + consume thousands of unnecessary tokens refining an output that never reaches + production. Effective governance must therefore extend to resource orchestration + and budget control, ensuring that autonomy remains economically as well as operationally + sustainable.\\nAgentic AI requires governance that extends beyond model oversight + to execution-time control \u2014the ability to monitor, interpret, and intervene + in agentic behavior as it unfolds. This shift sets the foundation for the next + layer: designing effective,**multi-tiered governance frameworks**that balance + autonomy with accountability.\\n# What Effective AI Agent Governance Looks Like\\nEffective + governance for agentic AI isn\u2019t about limiting innovation, it\u2019s about + shaping autonomy within safe, observable boundaries. As enterprises move from + experimentation to scaled deployment, governance must evolve into a multi-layered + system that defines rules, enforces them during execution, and ensures continuous + accountability afterward.\\nAn emerging best practice is to view AI agent governance + across three complementary layers: Policy, Platform, and Oversight. Each plays + a distinct role in ensuring that autonomy remains productive, compliant, and + aligned with human intent.\\n## 1. Policy Layer: Defining the Rules of Autonomy\\nThe + foundation of any governance model is a clear, codified policy framework. This + layer sets the**limits of autonomy**, outlining what an agent can do, under + what conditions, and with what access rights. For example, in a government services + environment, an AI intake agent may automatically classify and route citizen + requests but must flag any case that involves sensitive personal data for manual + verification.\\nThis layer acts as the blueprint for accountability. When implemented + through[composable low-code architectures], these rules can be versioned, reused, + and adapted across teams, ensuring consistent governance as agents evolve.\\n## + 2. Platform Layer: Enforcing Guardrails at Runtime\\nEven the best policies + mean little if they can\u2019t be enforced dynamically. The platform layer ensures + that rules set at design time translate into**real-time control and observability**during + execution. This is where orchestration becomes essential.\\nThrough centralized + execution flows, enterprises can monitor agent actions as they happen, validate + them against defined policies, and trigger fallback or escalation mechanisms + when anomalies occur. As an example, in financial operations, an agent responsible + for reconciling cross-border transactions may autonomously process low-value + payments but must request human validation for high-value or cross-jurisdiction + transfers.\\nThis\\nSummary: None\\n\\n\\nTitle: AI Agent Governance: A Field + Guide (IAPS, Apr 2025)\\nURL: https://www.aigl.blog/ai-agent-governance-a-field-guide-iaps-apr-2025/\\nID: + https://www.aigl.blog/ai-agent-governance-a-field-guide-iaps-apr-2025/\\nScore: + None\\nPublished Date: 2025-11-07T00:00:00.000Z\\nAuthor: Jakub Szarmach\\nImage: + https://www.aigl.blog/content/images/2025/11/AI-Agent-Governance-A-Field-Guide.jpg\\nFavicon: + https://www.aigl.blog/content/images/size/w256h256/2025/03/LOGO_no_background-removebg-preview.png\\nExtras: + None\\nSubpages: None\\nText: AI Agent Governance: A Field Guide (IAPS, Apr + 2025)\\n[AI Governance Library] **[Subscribe] \\n[**Sign up] [**Sign in] \\n[Log + in] [Subscribe] \\n**\\n# AI Agent Governance: A Field Guide (IAPS, Apr 2025)\\nA + practical, outcomes-driven playbook for governing AI agents\u2014covering risks, + adoption pathways, and a five-part taxonomy of interventions (Alignment, Control, + Visibility, Security & robustness, Societal integration), with examples + and vignettes.\\n**********\\n* [![Jakub Szarmach]] \\n[Jakub Szarmach] \\nNov + 7, 20254 min read\\n![AI Agent Governance: A Field Guide (IAPS, Apr 2025)] \\nOn + this page**\\n[\\nAI Agent Governance A Field Guide\\nAI Agent Governance A + Field Guide.pdf\\n2 MB\\ndownload-circle\\n] \\n## **\u26A1 Quick Summary**\\nThis + field guide from the Institute for AI Policy and Strategy maps the near-term + reality of agentic AI\u2014LLM-scaffolded systems that plan, remember, use tools, + and act\u2014and the governance work needed before they scale to \u201Cmillions + or billions.\u201D It distinguishes hype from evidence: agents can already add + value (customer support, AI R&D, some cyber tasks) but underperform humans + on long, open-ended workflows. The core governance question is how to keep autonomous, + tool-using systems safe, steerable, and accountable at scale. The report\u2019s + keystone is a five-category intervention taxonomy\u2014Alignment, Control, Visibility, + Security & robustness, and Societal integration\u2014spanning model, system, + and ecosystem layers, with concrete mechanisms (e.g., shutdown/rollback, agent + IDs, activity logging, liability regimes). The authors argue the capability + curve (especially \u201Ctest-time compute\u201D models like o-series) will bend + upward faster than governance capacity unless institutions, standards, and infrastructure + are built now. (See the intervention table and scenarios on pp. 10\u201312 and + 34\u201347; agent architecture diagram on p.14.)\\n## \U0001F9E9What\u2019s + Covered\\n* **Current state of agents.**A crisp definition (\u201CAI systems + that can autonomously achieve goals in the world\u201D) and the scaffolding + pattern around frontier models: reasoning/planning, memory, tool-use, and multi-agent + collaboration (p.14\u201317). Benchmarks summarized across GAIA, METR Autonomy + Evals, SWE-bench (incl. Verified), WebArena, OSWorld, RE-bench, CyBench, etc., + showing sharp fall-offs as human task time exceeds \\\\~1 hour and on open, + visually grounded, or long-horizon tasks (pp. 16\u201321; Appendix pp. 52\u201355).\\n* + **Pathways to better agents.**Improvements via stronger \u201Ccontroller\u201D + models and better scaffolding/orchestration; rise of**test-time compute**(o1/o3, + DeepSeek-R1) enabling longer, higher-quality reasoning; memory/tool ecosystems; + multi-agent orchestration (pp. 21\u201323).\\n* **Adoption lens.**Where early + ROI lands: customer relations (Klarna case), AI/ML R&&D (engineering-heavy + steps, fast feedback), and cybersecurity (mixed\u2014but promising\u2014signals). + Adoption governed by performance, cost (often 1/30th of US bachelor median wage + on certain tasks), and**reliability**(pp. 22\u201328).\\n* **Risk landscape.**Four + buckets: (1)**Malicious use**(scaled disinformation, cyber offense, dual-use + bio); (2)**Accidents & loss of control**(from mundane reliability failures + to \u201Crogue replication\u201D and scheming/deception); (3)**Security**(expanded + attack surface from tools, memory, and agent-agent dynamics, incl. \u201Cinfectious + jailbreaks\u201D); (4)**Systemic risks**(labor displacement, power concentration, + democratic erosion, market \u201Chyperswitching\u201D) (pp. 27\u201333).\\n* + **Agent governance as a distinct field.**Why agents change the governance calculus: + direct world actions, opacity + speed, multi-agent effects, and new levers at + the ecosystem layer (payments, browsers, IDs). Priorities include capability/risk + monitoring, lifecycle controls, incentivizing defensive/beneficial use, adapting + law/policy (pp. 31\u201335).\\n* **Intervention taxonomy (core contribution).**\\n* + **Alignment**(e.g., multi-agent RL, risk-attitude alignment, CoT paraphrasing, + alignment evals).\\n* **Control**(e.g., rollback infrastructure, shutdown/interrupt/timeouts, + tool/action restrictions, control protocols && evaluations).\\n* **Visibility**(e.g.,**Agent + IDs**, activity logging, cooperation-relevant capability evals, RL**reward reports**).\\n* + **Security & robustness**(e.g., access control tiers, adversarial robustness + testing,**sandboxing**, rapid response for adaptive defense).\\n* **Societal + integration**(e.g.,**liability regimes**, commitment devices/smart contracts, + equitable agent access schemes,**law-following agents**).Each category includes + plain-language vignettes showing how a measure works in practice (pp. 36\u201347).\\n* + **Two futures.**\u201CAgent-Driven Renaissance\u201D vs. \u201CAgents Run Amok\u201D + scenarios make the stakes concrete\u2014highlighting the need for IDs, logging, + rollbacks, defensive agents, and access/benefit-sharing (pp. 10\u201313).\\n* + **Research & capacity gap.**Few teams are working on agent governance relative + to the resources pouring into capability-building; the guide flags funders, + workshops, and priority research threads (pp. 5\u20137, 31\u201335, 48\u201350).\\n(See + the**taxonomy table**on p.6 and pp. 36\u201347 for detailed exemplars;**agent + architecture diagram**on p.14;**benchmark summaries**on pp. 18\u201321 and Appendix.)\\n# + **\U0001F4A1 Why it matters?**\\nAgents aren\u2019t just chatbots\u2014they\u2019re**actors**. + Once systems can plan, remember, and click \u201Cbuy\u201D (orexec), classic + safety levers (content filters, pre-deployment evals) aren\u2019t enough. Governance + must move**into the loop**: IDs and logs for traceability; rollback/shutdown + for damage control; sandboxing and access controls for containment; liability + and law-following to align incentives; and alignment methods suited to long-horizon, + multi-agent settings. Building this stack**before**mass deployment is the difference + between autonomous productivity and autonomous externalities. The report delivers + a concrete, actionable blueprint. (See pp. 34\u201347.)\\n# **\u2753 What\u2019s + Missing**\\n* **Operational maturity levels.**A readiness/assurance model (e.g., + Agent-ML levels with required controls/evals per tier) would help buyers and + regulators phase adoption.\\n* **Quantified control efficacy.**The guide catalogs + controls but offers limited empirical evidence on success rates (e.g., mean + time-to-shutdown, rollback coverage, bypass rates).\\n* **Supply-chain specifics.**More + detail on browser/OS, payment, and cloud co-regulation (who hosts agent IDs, + where logs live, cross-jurisdiction handling).\\n* **Human factors.**Guidance + for organizational design: who owns rollback authority, separation of duties, + incident playbooks, and \u201Ckill-switch\u201D drills.\\n* **Evaluation economics.**Costs + and sampling strategies to make agent evals reproducible at scale (the Appendix + notes time/cost, but buyers need budgeting templates).## \U0001F465Best Fo**r**\\n* + **CISOs/CIOs & platform owners**designing safe agent platforms (browser/OS, + payments, cloud, enterprise suites).\\n* **Risk, policy, & compliance leaders**crafting + internal standards and procurement criteria for agentic systems.\\n* **Regulators + & standards bodies**mapping controls to duties (IDs, logging, rollback, + liability apportionment).\\n* **Research leads & safety teams**prioritizing + evaluations, red-teaming, and adaptive defenses for agents.\\n* **Product strategists + & founders**building vertical agents who need a governance-as-a-feature + roadmap.## \U0001F4C4Source De**tails**\\n* **Title:***AI Agent Governance: + A Field Guide*\\n* **Authors:**Jam Kraprayoon, Zoe Williams, Rida Fayyaz\\n* + **Publisher:**Institute for AI Policy and Strategy (IAPS)\\n* **Date / Length:**April + 2025 / 63 pp.\\n* **Standout visuals:**Agent architecture diagram (p.14); intervention + taxonomy (pp. 6 && 36\u201347); benchmark tables && notes (pp. + 18\u201321; Appendix).## \U0001F4DDThanks**to**\\nAcknowledged contributors + include Alan Chan, Cullen O\u2019Keefe, Shaun Ee, Ollie Stephenson, Cristina + Schmidt-Ib\xE1\xF1ez, Clara Langevin, and Matthew Burtell (p.50).\\nIf desired, + an AIGL one-pager can be prepared distilling the taxonomy into procurement and + control checklists per deployment tier.\\n[Guide] \\n[**Share] [**Share] [**Share] + [**Share] [**Email] **Copy\\nAbout the author\\n[![Jakub Szarmach]] \\n## [Jakub + Szarmach] \\nLatest\\n[![Architecting Secure Enterprise AI Agents with MCP] + \\n### Architecting Secure Enterprise AI Agents with MCP\\n06 Feb 2026\\n] [![AI + Security Institute \u2013Frontier AI Trends Report (December 2025)] \\n### AI + Security Institute \u2013Frontier AI Trends Report (December 2025)\\n06 Feb + 2026\\n] [![AI Model Risk Management Framework] \\n### AI Model Risk Management + Framework\\n06 Feb 2026\\n] \\n#### AI Governance Library\\nCurated Library + of AI Governance Resources\\nSubscribe\\nGreat! Check your inbox and click the + link.\\nSorry, something went wrong. Please try again.\\nRead next\\n[### Architecting + Secure Enterprise AI Agents with MCP\\n**] [### AI Incident Response: Adapting + Proven Complex Systems Engineering Practices for AI-Enabled Systems\\n**] [### + AI for Security and Security for AI: Navigating Opportunities and Challenges\\n**] + [### Administering and Governing Agents\\n**] [### A practical guide to implementing + AI ethics governance\\n**] \\n## AI Governance Library\\nCurated Library of + AI Governance Resources\\nSubscribe\\nGreat! Check your inbox and click the + link.\\nSorry, something went wrong. Please try again.\\n![AI Governance Library] + \\n******\\nGreat! You\u2019ve successfully signed up.\\nWelcome back! You've + successfully signed in.\\nYou've successfully subscribed to AI Governance + Library.\\nYour link has expired.\\nSuccess! Check your email for magic link + to sign-in.\\nSuccess! Your billing info has been updated.\\nYour billing was + not updated.\\n**\\nSummary: None\\n\\n\\nTitle: AI Governance in the Era of + Agentic Generative AI and AGI: Frameworks, Risks, and Policy Directions\\nURL: + https://www.ijircst.org/DOC/3-AI-Governance-in-the-Era-of-Agentic-Generative-AI-and-AGI-Frameworks-Risks-and-Policy-Directions.pdf\\nID: + https://www.ijircst.org/DOC/3-AI-Governance-in-the-Era-of-Agentic-Generative-AI-and-AGI-Frameworks-Risks-and-Policy-Directions.pdf\\nScore: + None\\nPublished Date: 2025-09-04T00:00:00.000Z\\nAuthor: Satyadhar Joshi\\nImage: + None\\nFavicon: None\\nExtras: None\\nSubpages: None\\nText: International Journal + of Innovative Research in Computer Science and Technology (IJIRCST)\\nISSN (Online): + 2347-5552, Volume-13, Issue-5, September 2025\\nDOI: https:/doi.org/10.55524/ijircst.2025.13.5.3\\nArticle + ID IRP-1674, Pages 14-24\\nwww.ijircst.org\\nInnovative Research Publication + 14\\nAI Governance in the Era of Agentic Generative AI and AGI: \\nFrameworks, + Risks, and Policy Directions\\nSatyadhar Joshi\\nIndependent Researcher, Alumnus, + International MBA, Bar-Ilan University, Israel\\nCorrespondence should be addressed + to Satyadhar Joshi;\\n Received 27 July 2025; Revised 11 August 2025; Accepted + 26 August 2025\\nCopyright \xA9 2025 Made Satyadhar Joshi. This is an open-access + article distributed under the Creative Commons Attribution License, \\nwhich + permits unrestricted use, distribution, and reproduction in any medium, provided + the original work is properly cited.\\nABSTRACT- The accelerating development + of agentic \\nartificial intelligence (AI) and the prospect of artificial \\ngeneral + intelligence (AGI) create unprecedented \\nopportunities alongside complex governance + challenges. \\nThis paper examines the ethical, regulatory, and technical \\ndimensions + of governing highly autonomous AI systems, \\ndrawing upon more than fifty contemporary + academic and \\npolicy sources. Three core insights emerge. First, current \\ngovernance + structures provide limited coverage of risks \\nlinked to recursive self-improvement + and multi-agent \\ncoordination, with only an estimated 10\u201315% of safety + \\nresearch addressing impacts that arise after deployment. \\nSecond, economic + projections suggest that agentic AI could \\ngenerate between 2.6 and 4.4 trillion + USD in added global \\noutput by 2030, yet automation could replace \\napproximately + 28\u201342% of existing job tasks, making \\nproactive workforce transition + strategies a policy necessity. \\nThird, fragmented regulatory approaches remain + a concern; \\nin the United States, for example, 70\u201375% of critical \\ninfrastructure + is considered vulnerable to adversarial \\nautonomous systems. To address these + issues, we propose a \\ngovernance model built on three pillars: modular agent + \\ndesign, adaptive safety mechanisms, and international \\ncoordination. Policy + measures such as licensing thresholds \\nfor high-computer systems exceeding + 10^25 FLOPs, \\nstructured red-team testing across public and private \\nsectors, + and fiscal incentives for governance-by-design \\npractices are advanced as + actionable pathways. Overall, the \\nstudy argues for adaptive, globally coordinated + governance \\nframeworks that balance innovation with systemic risk \\nmitigation + in the era of agentic AI and AGI. his is a pure \\nreview paper and all results, + proposals and findings are from \\nthe cited literature. \\nKEYWORDS- AI Governance, + Agentic AI, Generative \\nAI, Artificial General Intelligence (AGI), Ethics, + Policy, \\nRisk Management, Recursive Self-Improvement, Multi\\u0002Agent Systems, + Workforce Transition, International \\nCoordination\\nI. INTRODUCTION\\nThe + emergence of agentic artificial intelligence (AI)\u2014\\nsystems capable of + autonomous goal-setting, decision\\u0002making, and task execution\u2014marks + a fundamental \\nparadigm shift in computational intelligence. Unlike \\nconventional + generative AI, which operates within static \\nprompt\u2013response frameworks, + agentic AI demonstrates \\ndynamic adaptability, recursive self-improvement + (RSI), \\nand multi-agent collaboration [1], [2]. These capabilities \\nenable + agentic systems not only to generate content but also \\nto plan, execute, and + optimize tasks with minimal human \\noversight. Parallel advancements in artificial + general \\nintelligence (AGI) further amplify both the transformative \\npotential + and systemic risks of these technologies. Current \\neconomic projections suggest + that agentic AI could \\ncontribute between $2.6 and $4.4 trillion to global + GDP by \\n2030, while simultaneously automating 28\u201342% of job\\u0002related + tasks [3], [4]. Such forecasts underscore the dual \\nchallenge of harnessing + productivity gains while mitigating \\nwidespread labor market disruptions.\\nDespite + growing awareness, existing governance \\nframeworks remain ill-equipped to + manage the \\ncomplexities of agentic AI and AGI. This paper identifies \\nthree + central gaps in current approaches. First, regulatory \\nand ethical models + provide insufficient guidance for \\naddressing risks associated with recursive + self\\u0002improvement and autonomous multi-agent coordination. \\nSecond, regulatory + fragmentation across jurisdictions \\nhinders coherent oversight; in the United + States alone, \\nestimates indicate that 70\u201375% of critical infrastructure + \\nremains exposed to adversarial exploitation by autonomous \\nsystems [5]. + Third, policy tools remain largely static, \\nlacking the adaptive mechanisms + necessary to keep pace \\nwith rapidly evolving agentic technologies.\\nTo address + these issues, this paper synthesizes insights from \\nover fifty contemporary + sources, including academic \\nliterature (32%), industry reports (28%), and + government \\npublications (20%), with a particular emphasis on policy \\ndevelopments + in 2024\u20132025, such as the European Union\u2019s \\nAI Act [6] and the United + States\u2019 Executive Order 14110\\n[7]. Our contributions are fourfold:\\n\uF0B7 + Conceptual foundations: We review the terminology \\nand governance challenges + of agentic AI and AGI, \\nsupported by definitional clarity (Table I) and forward\\u0002looking + projections (Figure 1).\\n\uF0B7 Technical governance frameworks: We introduce + \\ncompliance models such as governance scoring (Eq. 1) \\nand RSI optimization + methods (Algorithm 1).\\n\uF0B7 Comparative policy analysis: We evaluate governance + \\nstrategies across jurisdictions and sectors, summarized \\nin Table 4.\\n\uF0B7 + Tripartite governance architecture: We propose an \\nintegrated model combining + modular agent design, \\nInternational Journal of Innovative Research in Computer + Science and Technology (IJIRCST)\\nInnovative Research Publication 15\\nevolutionary + safety algorithms, and international \\ncoordination, illustrated in Figure + 3.\\nBy bridging technical, economic, and policy perspectives, \\nthis study + provides actionable recommendations for \\nstakeholders navigating the agentic + AI era. These include \\nstandardized licensing requirements for high-computer\\nFigure + 1: Future timeline of agentic AI impacts (2025-2035) showing adoption rates + (blue), economic effects (green/orange), \\nand workforce changes (red/purple) + with data source references\\nsystems (>10^25 FLOPs), structured public\u2013private + red\\u0002teaming protocols, and governance-by-design tax \\nincentives. The + concluding section outlines urgent priorities \\nfor research and regulation, + with the aim of developing \\nadaptive, globally coordinated governance frameworks + that \\nbalance innovation with existential risk mitigation.\\nII. LITERATURE + REVIEW\\nThe literature review presented in this study is based on a \\nsystematic + synthesis of more than fifty sources, \\nencompassing academic journals, industry + reports, \\ngovernment publications, conference proceedings, and \\ntrade media. + The objective of this review is to capture the \\nevolving technical, ethical, + and policy dimensions of \\nagentic artificial intelligence (AI) and artificial + general \\nintelligence (AGI) governance across multiple \\nperspectives.\\nAcademic + journals constituted the largest share of reviewed \\nmaterials, accounting + for approximately 32% of the \\nsources. These works were primarily drawn from + peer\\u0002reviewed venues such as IEEE Computer [8], specialized \\noutlets + focusing on robotics, safety, and alignment [4], and \\nprominent policy journals. + The emphasis of this body of \\nwork was on technical architecture (15%), governance + \\nframeworks (10%), and quantitative risk analyses (7%), \\nproviding the theoretical + and conceptual foundation for \\nsubsequent developments.\\nIndustry contributions + represented 28% of the corpus, with \\ninfluential white papers from technology + corporations such \\nas IBM [9] and Deloitte [10], alongside reports by global + \\nthink tanks including the Global Skill Development Council \\n(GSDC) [5]. + These reports concentrated on market \\nprojections (12%), deployment case studies + (9%), and \\nemerging self-regulation standards (7%). Although industry \\nsources + provided pragmatic insights into implementation, \\nthey also reflected a tendency + toward optimism regarding \\nscalability and adoption.\\nGovernment publications + accounted for 20% of the \\nreferences, with key contributions including analyses + of the \\nU.S. Executive Orders on AI [7], the European Union\u2019s AI \\nAct + [6], and the United Nations\u2019 perspectives on AI \\ngovernance [11]. These + materials collectively focused on \\nregulatory frameworks (14%) and national + strategies (6%), \\noffering a policy-oriented complement to technical studies. + \\nIn parallel, conference proceedings and preprints comprised \\n12% of the + literature, particularly from NeurIPS workshops \\non AI safety and IEEE Symposia + addressing AGI \\ngovernance. These emerging works provided insights into \\nalgorithmic + innovations (8%) and governance prototypes \\n(4%), reflecting the experimental + stage of research in this \\narea. Finally, 8% of the sources were derived from + news \\noutlets and trade media such as TechRadar [12] and \\nMarkTechPost [13]. + These contributions highlighted real\\u0002world deployments (5%) and expert + interviews (3%), \\nserving as a bridge between academic analysis and public + \\ndiscourse.\\nIn terms of temporal distribution, nearly half of the \\nreviewed + works (50%) were published in 2025, reflecting \\nthe intensification of debates + surrounding agentic AI \\ndeployment, risk analyses [18], and U.S.-specific + \\nregulatory actions [19]. A further 32% were concentrated in \\n2024, coinciding + with a surge in ethical frameworks [16] \\nand global policy proposals [11]. + The earlier period of \\n2021\u20132023 accounted for 18% of the references, + focusing \\nlargely on distinctions between generative and agentic AI \\n[14] + and the first wave of governance frameworks [15].\\nGeographically, the review + reflects both national and \\ninternational perspectives. U.S.-centric studies + comprised \\n45% of the dataset, particularly in relation to federal and \\nstate-level + pol\\nSummary: None\\n\\n\\nTitle: Preparing for AI Agent Governance\\nURL: + https://partnershiponai.org/resource/preparing-for-ai-agent-governance/\\nID: + https://partnershiponai.org/resource/preparing-for-ai-agent-governance/\\nScore: + None\\nPublished Date: 2025-09-30T00:00:00.000Z\\nAuthor: Jacob Pratt\\nImage: + https://partnershiponai.org/wp-content/uploads/2025/10/policy-agenda-feature-img-1800x832-1.webp\\nFavicon: + https://partnershiponai.org/wp-content/uploads/2021/08/PAI-Favicon.png\\nExtras: + None\\nSubpages: None\\nText: Preparing for AI Agent Governance - Partnership + on AI\\n[![logo]![logo]![logo]] \\n* [] \\n* [DONATE] \\n×\\n[SEARCH] + \\nShare\\n* [**] \\n* [**] \\n* [**] \\n[Our Resources] ##### /\\n##### Report\\nPolicy\\n# + Preparing for AI Agent Governance\\n### A research agenda for policymakers and + researchers\\n![] \\nJacob Pratt\\nSeptember 30, 2025\\n![$hero_image['alt']] + \\n### Key Takeaways\\n* **How AI agents will impact society is still uncertain.**While + the body of scholarly work and publicly available evidence are growing, we don\u2019t + yet know enough about how AI agents will be used or what impacts they may have. + However, AI investments continue to soar, so policymakers should begin to prepare + now.\\n* **Policymakers should prioritize evidence and information gathering, + including through sandboxes and testbeds.**Given this uncertainty, policymakers + should promote activities to generate evidence, rather than advancing prescriptive + regulations. Promising options policymakers can use to build expertise and track + developments are sandboxes and testbeds, which enable experimentation of new + systems under regulatory supervision.\\n* **Subsequent rule-making will require + substantial research from inside and outside of government.**Academia, civil + society, government, and industry should work together to generate evidence + on AI agents’ capabilities, risks, societal impacts, and potential policy + interventions. This will support future policy development.\\n* **This paper + provides a roadmap for this research.**We outline three foundational requirements + for governing AI agents and detail a comprehensive research agenda, including + 12 top-level and 45 sub-level questions, designed to directly support policymakers + in developing evidence-based policy.\\nIndustry leaders have named 2025 the + \u201Cyear of agentic exploration,\u201D foretelling the adoption of systems + that will change how we interact, what jobs we perform, and even how we think. + However, these innovations face significant hurdles: widespread AI agent adoption + has been stifled by persistent reliability and security challenges, giving policymakers + an opportunity to prepare thoughtfully for how to promote the benefits and protect + against the risks of AI agents.\\nMaking the right public policy decisions on + AI governance, including on the institutions, policies, regulations, and tools + that ensure that AI systems operate in the public interest, requires substantial + research and evidence. Though there is significant research done on AI more + broadly and the theoretical impacts of AI agents, we don\u2019t yet know enough + about how AI agents will be used or what impacts they may have. The key challenge + for policymakers is not whether to regulate now, but how to govern AI agents + while their impacts are still uncertain, and understand what evidence will be + needed to make informed decisions when decisive action is needed.\\n> The key + challenge for policymakers is not whether to regulate now, but how to govern + AI agents while their impacts are still uncertain.\\n**In this paper, we outline + a research agenda to guide policymakers and researchers in preparing for the + governance of AI agents.**Structured around three core technology governance + requirements for policymakers – foundational understanding, impact assessment, + and intervention evaluation – we present 12 top-level questions and 45 + detailed open sub-questions from literature and partner discussions that policymakers + and researchers should explore. We also explore how monitoring, sandboxes, and + testbeds can be an important first step for policymakers.\\nPAI has already + started conducting research that supports this agenda, and have recently published + work on[prioritizing real-time failure detection] and[global governance]. We + will continue to build on this work, and look forward to collaborating with + our community to advance AI agent governance.\\nDownload the full research agenda + and be a part of guiding evidence-based AI agent governance now.\\n[Back to + All Resources]\\nSummary: None\\n\\n\\nTitle: We Need a New Ethics for a World + of AI Agents\\nURL: https://arxiv.org/abs/2509.10289\\nID: https://arxiv.org/abs/2509.10289\\nScore: + None\\nPublished Date: 2025-09-12T00:00:00.000Z\\nAuthor: [Submitted on 12 Sep + 2025]\\nImage: /static/browse/0.3.4/images/arxiv-logo-fb.png\\nFavicon: https://arxiv.org/static/browse/0.3.4/images/icons/favicon-32x32.png\\nExtras: + None\\nSubpages: None\\nText: [2509.10289] We Need a New Ethics for a World + of AI Agents\\n[Skip to main content] \\n[![Cornell University]] \\nWe gratefully + acknowledge support from the Simons Foundation,[member institutions], and all + contributors.[Donate] \\n[] \\n[![arxiv logo]] >[cs] >arXiv:2509.10289\\n[Help] + |[Advanced Search] \\nAll fieldsTitleAuthorAbstractCommentsJournal referenceACM + classificationMSC classificationReport numberarXiv identifierDOIORCIDarXiv author + IDHelp pagesFull text\\nSearch\\n[![arXiv logo]] \\n[![Cornell University Logo]] + \\nopen search\\nGO\\nopen navigation menu\\n# Computer Science \\\\> Computers + and Society\\n**arXiv:2509.10289**(cs)\\n[Submitted on 12 Sep 2025]\\n# Title:We + Need a New Ethics for a World of AI Agents\\nAuthors:[Iason Gabriel],[Geoff + Keeling],[Arianna Manzini],[James Evans] \\nView a PDF of the paper titled We + Need a New Ethics for a World of AI Agents, by Iason Gabriel and 3 other authors\\n[View + PDF] > > Abstract:\\n> The deployment of capable AI agents raises fresh questions + about safety, human-machine relationships and social coordination. We argue + for greater engagement by scientists, scholars, engineers and policymakers with + the implications of a world increasingly populated by AI agents. We explore + key challenges that must be addressed to ensure that interactions between humans + and agents, and among agents themselves, remain broadly beneficial. Comments:|6 + pages, no figures|\\nSubjects:|Computers and Society (cs.CY); Artificial Intelligence + (cs.AI)|\\nACMclasses:|I.2.0; K.4.1|\\nCite as:|[arXiv:2509.10289] [cs.CY]|\\n|(or[arXiv:2509.10289v1] + [cs.CY]for this version)|\\n|[https://doi.org/10.48550/arXiv.2509.10289] \\nFocus + to learn more\\narXiv-issued DOI via DataCite\\n|\\nJournalreference:|Nature, + 644 (8075), 2025, 38-40|\\nRelated DOI:|[https://doi.org/10.1038/d41586-025-02454-5] + \\nFocus to learn more\\nDOI(s) linking to related resources\\n|\\n## Submission + history\\nFrom: Iason Gabriel [[view email]]\\n**[v1]**Fri, 12 Sep 2025 14:29:14 + UTC (162 KB)\\nFull-text links:## Access Paper:\\nView a PDF of the paper titled + We Need a New Ethics for a World of AI Agents, by Iason Gabriel and 3 other + authors\\n* [View PDF] \\n[view license] \\nCurrent browse context:\\ncs.CY\\n[<<prev] + | [next>>] \\n[new] |[recent] |[2025-09] \\nChange to browse by:\\n[cs] + \\n[cs.AI] \\n### References & Citations\\n* [NASA ADS] \\n* [Google Scholar] + \\n* [Semantic Scholar] \\nexport BibTeX citationLoading...\\n## BibTeX formatted + citation\\n×\\nloading...\\nData provided by:\\n### Bookmark\\n[![BibSonomy + logo]] [![Reddit logo]] \\nBibliographic Tools\\n# Bibliographic and Citation + Tools\\nBibliographic Explorer Toggle\\nBibliographic Explorer*([What is the + Explorer?])*\\nConnected Papers Toggle\\nConnected Papers*([What is Connected + Papers?])*\\nLitmaps Toggle\\nLitmaps*([What is Litmaps?])*\\nscite.ai Toggle\\nscite + Smart Citations*([What are Smart Citations?])*\\nCode, Data, Media\\n# Code, + Data and Media Associated with this Article\\nalphaXiv Toggle\\nalphaXiv*([What + is alphaXiv?])*\\nLinks to Code Toggle\\nCatalyzeX Code Finder for Papers*([What + is CatalyzeX?])*\\nDagsHub Toggle\\nDagsHub*([What is DagsHub?])*\\nGotitPub + Toggle\\nGotit.pub*([What is GotitPub?])*\\nHuggingface Toggle\\nHugging Face*([What + is Huggingface?])*\\nLinks to Code Toggle\\nPapers with Code*([What is Papers + with Code?])*\\nScienceCast Toggle\\nScienceCast*([What is ScienceCast?])*\\nDemos\\n# + Demos\\nReplicate Toggle\\nReplicate*([What is Replicate?])*\\nSpaces Toggle\\nHugging + Face Spaces*([What is Spaces?])*\\nSpaces Toggle\\nTXYZ.AI*([What is TXYZ.AI?])*\\nRelated + Papers\\n# Recommenders and Search Tools\\nLink to Influence Flower\\nInfluence + Flower*([What are Influence Flowers?])*\\nCore recommender toggle\\nCORE Recommender*([What + is CORE?])*\\n* Author\\n* Venue\\n* Institution\\n* Topic\\nAbout arXivLabs\\n# + arXivLabs: experimental projects with community collaborators\\narXivLabs is + a framework that allows collaborators to develop and share new arXiv features + directly on our website.\\nBoth individuals and organizations that work with + arXivLabs have embraced and accepted our values of openness, community, excellence, + and user data privacy. arXiv is committed to these values and only works with + partners that adhere to them.\\nHave an idea for a project that will add value + for arXiv's community?[**Learn more about arXivLabs**].\\n[Which authors of + this paper are endorsers?] |[Disable MathJax] ([What is MathJax?])\\nSummary: + None\\n\\n\\nTitle: AI Agents Should be Regulated Based on the Extent of Their + Autonomous Operations\\nURL: https://arxiv.org/abs/2503.04750\\nID: https://arxiv.org/abs/2503.04750\\nScore: + None\\nPublished Date: 2025-02-07T00:00:00.000Z\\nAuthor: [Submitted on 7 Feb + 2025 (v1), last revised 26 May 2025 (this version, v2)]\\nImage: /static/browse/0.3.4/images/arxiv-logo-fb.png\\nFavicon: + https://arxiv.org/static/browse/0.3.4/images/icons/favicon-32x32.png\\nExtras: + None\\nSubpages: None\\nText: [2503.04750] AI Agents Should be Regulated Based + on the Extent of Their Autonomous Operations\\n[Skip to main content] \\n[![Cornell + University]] \\nWe gratefully acknowledge support from the Simons Foundation,[member + institutions], and all contributors.[Donate] \\n[] \\n[![arxiv logo]] >[cs] + >arXiv:2503.04750\\n[Help] |[Advanced Search] \\nAll fieldsTitleAuthorAbstractCommentsJournal + referenceACM classificationMSC classificationReport numberarXiv identifierDOIORCIDarXiv + author IDHelp pagesFull text\\nSearch\\n[![arXiv logo]] \\n[![Cornell University + Logo]] \\nopen search\\nGO\\nopen navigation menu\\n# Computer Science \\\\> + Computers and Society\\n**arXiv:2503.04750**(cs)\\n[Submitted on 7 Feb 2025 + ([v1]), last revised 26 May 2025 (this version, v2)]\\n# Title:AI Agents Should + be Regulated Based on the Extent of Their Autonomous Operations\\nAuthors:[Takayuki + Osogami] \\nView a PDF of the paper titled AI Agents Should be Regulated Based + on the Extent of Their Autonomous Operations, by Takayuki Osogami\\n[View PDF] + [HTML (experimental)] > > Abstract:\\n> This position paper argues that AI agents + should be regulated by the extent to which they operate autonomously. AI agents + with long-term planning and strategic capabilities can pose significant risks + of human extinction and irreversible global catastrophes. While existing regulations + often focus on computational scale as a proxy for potential harm, we argue that + such measures are insufficient for assessing the risks posed by agents whose + capabilities arise primarily from inference-time computation. To support our + position, we discuss relevant regulations and recommendations from scientists + regarding existential risks, as well as the advantages of using action sequences + -- which reflect the degree of an agent's autonomy -- as a more suitable + measure of potential impact than existing metrics that rely on observing environmental + states. Comments:|Updated to improve the overall clarity with additional recent + references; 22 pages, 2 figures|\\nSubjects:|Computers and Society (cs.CY); + Artificial Intelligence (cs.AI)|\\nCite as:|[arXiv:2503.04750] [cs.CY]|\\n|(or[arXiv:2503.04750v2] + [cs.CY]for this version)|\\n|[https://doi.org/10.48550/arXiv.2503.04750] \\nFocus + to learn more\\narXiv-issued DOI via DataCite\\n|\\n## Submission history\\nFrom: + Takayuki Osogami Ph.D. [[view email]]\\n**[[v1]] **Fri, 7 Feb 2025 09:40:48 + UTC (166 KB)\\n**[v2]**Mon, 26 May 2025 08:41:12 UTC (133 KB)\\nFull-text links:## + Access Paper:\\nView a PDF of the paper titled AI Agents Should be Regulated + Based on the Extent of Their Autonomous Operations, by Takayuki Osogami\\n* + [View PDF] \\n* [HTML (experimental)] \\n* [TeX Source] \\n[view license] \\nCurrent + browse context:\\ncs.CY\\n[<<prev] | [next>>] \\n[new] |[recent] + |[2025-03] \\nChange to browse by:\\n[cs] \\n[cs.AI] \\n### References & + Citations\\n* [NASA ADS] \\n* [Google Scholar] \\n* [Semantic Scholar] \\nexport + BibTeX citationLoading...\\n## BibTeX formatted citation\\n×\\nloading...\\nData + provided by:\\n### Bookmark\\n[![BibSonomy logo]] [![Reddit logo]] \\nBibliographic + Tools\\n# Bibliographic and Citation Tools\\nBibliographic Explorer Toggle\\nBibliographic + Explorer*([What is the Explorer?])*\\nConnected Papers Toggle\\nConnected Papers*([What + is Connected Papers?])*\\nLitmaps Toggle\\nLitmaps*([What is Litmaps?])*\\nscite.ai + Toggle\\nscite Smart Citations*([What are Smart Citations?])*\\nCode, Data, + Media\\n# Code, Data and Media Associated with this Article\\nalphaXiv Toggle\\nalphaXiv*([What + is alphaXiv?])*\\nLinks to Code Toggle\\nCatalyzeX Code Finder for Papers*([What + is CatalyzeX?])*\\nDagsHub Toggle\\nDagsHub*([What is DagsHub?])*\\nGotitPub + Toggle\\nGotit.pub*([What is GotitPub?])*\\nHuggingface Toggle\\nHugging Face*([What + is Huggingface?])*\\nLinks to Code Toggle\\nPapers with Code*([What is Papers + with Code?])*\\nScienceCast Toggle\\nScienceCast*([What is ScienceCast?])*\\nDemos\\n# + Demos\\nReplicate Toggle\\nReplicate*([What is Replicate?])*\\nSpaces Toggle\\nHugging + Face Spaces*([What is Spaces?])*\\nSpaces Toggle\\nTXYZ.AI*([What is TXYZ.AI?])*\\nRelated + Papers\\n# Recommenders and Search Tools\\nLink to Influence Flower\\nInfluence + Flower*([What are Influence Flowers?])*\\nCore recommender toggle\\nCORE Recommender*([What + is CORE?])*\\n* Author\\n* Venue\\n* Institution\\n* Topic\\nAbout arXivLabs\\n# + arXivLabs: experimental projects with community collaborators\\narXivLabs is + a framework that allows collaborators to develop and share new arXiv features + directly on our website.\\nBoth individuals and organizations that work with + arXivLabs have embraced and accepted our values of openness, community, excellence, + and user data privacy. arXiv is committed to these values and only works with + partners that adhere to them.\\nHave an idea for a project that will add value + for arXiv's community?[**Learn more about arXivLabs**].\\n[Which authors of + this paper are endorsers?] |[Disable MathJax] ([What is MathJax?])\\nSummary: + None\\n\\n\\nTitle: AI Agent Governance: A Field Guide\\nURL: https://arxiv.org/abs/2505.21808\\nID: + https://arxiv.org/abs/2505.21808\\nScore: None\\nPublished Date: 2025-05-27T00:00:00.000Z\\nAuthor: + [Submitted on 27 May 2025]\\nImage: /static/browse/0.3.4/images/arxiv-logo-fb.png\\nFavicon: + https://arxiv.org/static/browse/0.3.4/images/icons/favicon-32x32.png\\nExtras: + None\\nSubpages: None\\nText: [2505.21808] AI Agent Governance: A Field Guide\\n[Skip + to main content] \\n[![Cornell University]] \\nWe gratefully acknowledge support + from the Simons Foundation,[member institutions], and all contributors.[Donate] + \\n[] \\n[![arxiv logo]] >[cs] >arXiv:2505.21808\\n[Help] |[Advanced Search] + \\nAll fieldsTitleAuthorAbstractCommentsJournal referenceACM classificationMSC + classificationReport numberarXiv identifierDOIORCIDarXiv author IDHelp pagesFull + text\\nSearch\\n[![arXiv logo]] \\n[![Cornell University Logo]] \\nopen search\\nGO\\nopen + navigation menu\\n# Computer Science \\\\> Computers and Society\\n**arXiv:2505.21808**(cs)\\n[Submitted + on 27 May 2025]\\n# Title:AI Agent Governance: A Field Guide\\nAuthors:[Jam + Kraprayoon],[Zoe Williams],[Rida Fayyaz] \\nView a PDF of the paper titled AI + Agent Governance: A Field Guide, by Jam Kraprayoon and 2 other authors\\n[View + PDF] > > Abstract:\\n> This report serves as an accessible guide to the emerging + field of AI agent governance. Agents - AI systems that can autonomously achieve + goals in the world, with little to no explicit human instruction about how to + do so - are a major focus of leading tech companies, AI start-ups, and investors. + If these development efforts are successful, some industry leaders claim we + could soon see a world where millions or billions of agents autonomously perform + complex tasks across society. Society is largely unprepared for this development. + A future where capable agents are deployed en masse could see transformative + benefits to society but also profound and novel risks. Currently, the exploration + of agent governance questions and the development of associated interventions + remain in their infancy. Only a few researchers, primarily in civil society + organizations, public research institutes, and frontier AI companies, are actively + working on these challenges. Subjects:|Computers and Society (cs.CY)|\\nCite + as:|[arXiv:2505.21808] [cs.CY]|\\n|(or[arXiv:2505.21808v1] [cs.CY]for this version)|\\n|[https://doi.org/10.48550/arXiv.2505.21808] + \\nFocus to learn more\\narXiv-issued DOI via DataCite\\n|\\n## Submission history\\nFrom: + Jam Kraprayoon Mr. [[view email]]\\n**[v1]**Tue, 27 May 2025 22:26:51 UTC (2,346 + KB)\\nFull-text links:## Access Paper:\\nView a PDF of the paper titled AI Agent + Governance: A Field Guide, by Jam Kraprayoon and 2 other authors\\n* [View PDF] + \\n[![license icon] view license] \\nCurrent browse context:\\ncs.CY\\n[<<prev] + | [next>>] \\n[new] |[recent] |[2025-05] \\nChange to browse by:\\n[cs] + \\n### References & Citations\\n* [NASA ADS] \\n* [Google Scholar] \\n* + [Semantic Scholar] \\nexport BibTeX citationLoading...\\n## BibTeX formatted + citation\\n×\\nloading...\\nData provided by:\\n### Bookmark\\n[![BibSonomy + logo]] [![Reddit logo]] \\nBibliographic Tools\\n# Bibliographic and Citation + Tools\\nBibliographic Explorer Toggle\\nBibliographic Explorer*([What is the + Explorer?])*\\nConnected Papers Toggle\\nConnected Papers*([What is Connected + Papers?])*\\nLitmaps Toggle\\nLitmaps*([What is Litmaps?])*\\nscite.ai Toggle\\nscite + Smart Citations*([What are Smart Citations?])*\\nCode, Data, Media\\n# Code, + Data and Media Associated with this Article\\nalphaXiv Toggle\\nalphaXiv*([What + is alphaXiv?])*\\nLinks to Code Toggle\\nCatalyzeX Code Finder for Papers*([What + is CatalyzeX?])*\\nDagsHub Toggle\\nDagsHub*([What is DagsHub?])*\\nGotitPub + Toggle\\nGotit.pub*([What is GotitPub?])*\\nHuggingface Toggle\\nHugging Face*([What + is Huggingface?])*\\nLinks to Code Toggle\\nPapers with Code*([What is Papers + with Code?])*\\nScienceCast Toggle\\nScienceCast*([What is ScienceCast?])*\\nDemos\\n# + Demos\\nReplicate Toggle\\nReplicate*([What is Replicate?])*\\nSpaces Toggle\\nHugging + Face Spaces*([What is Spaces?])*\\nSpaces Toggle\\nTXYZ.AI*([What is TXYZ.AI?])*\\nRelated + Papers\\n# Recommenders and Search Tools\\nLink to Influence Flower\\nInfluence + Flower*([What are Influence Flowers?])*\\nCore recommender toggle\\nCORE Recommender*([What + is CORE?])*\\n* Author\\n* Venue\\n* Institution\\n* Topic\\nAbout arXivLabs\\n# + arXivLabs: experimental projects with community collaborators\\narXivLabs is + a framework that allows collaborators to develop and share new arXiv features + directly on our website.\\nBoth individuals and organizations that work with + arXivLabs have embraced and accepted our values of openness, community, excellence, + and user data privacy. arXiv is committed to these values and only works with + partners that adhere to them.\\nHave an idea for a project that will add value + for arXiv's community?[**Learn more about arXivLabs**].\\n[Which authors of + this paper are endorsers?] |[Disable MathJax] ([What is MathJax?])\\nSummary: + None\\n\\n\\nTitle: AI Agents Race in July 2025\\nURL: https://medium.com/@enrico.papalini/ai-agents-race-in-july-2025-77c1cb711f14\\nID: + https://medium.com/@enrico.papalini/ai-agents-race-in-july-2025-77c1cb711f14\\nScore: + None\\nPublished Date: None\\nAuthor: Enrico Papalini\\nImage: https://miro.medium.com/v2/resize:fit:1024/1*IQ_4v80JCQesyRnZVscQxQ.png\\nFavicon: + https://miro.medium.com/v2/5d8de952517e8160e40ef9841c781cdc14a5db313057fa3c3de41c6f5b494b19\\nExtras: + None\\nSubpages: None\\nText: AI Agents Race in July 2025. Cutting Through the + Hype to Uncover\u2026 | by Enrico Papalini | Medium\\n[Sitemap] \\n[Open in + app] \\nSign up\\n[Sign in] \\n[Medium Logo] \\n[\\nWrite\\n] \\n[\\nSearch\\n] + \\nSign up\\n[Sign in] \\n![] \\n# **AI Agents Race in July 2025**\\n## **Cutting + Through the Hype to Uncover Real Impact and Returns**\\n[\\n![Enrico Papalini] + \\n] \\n[Enrico Papalini] \\n38 min read\\n\xB7Jul 21, 2025\\n[\\n] \\n--\\n[] + \\nListen\\nShare\\nPress enter or click to view image in full size\\n![] \\nImage + generated by Bing## Executive Summary\\nThe discourse surrounding Artificial + Intelligence (AI) Agentic systems in July 2025 is rife with both immense promise + and considerable misunderstanding. A critical distinction has solidified: on + one hand, there are sophisticated AI-powered tools that automate predefined + workflows, often mislabeled as \u201Cagents\u201D; on the other, there are truly + autonomous AI agents capable of independent decision-making and goal execution. + While the former offers significant, achievable efficiency gains, the latter, + though rapidly advancing, remains in a \u201Cresearch preview\u201D phase for + high-stakes applications, necessitating a pragmatic approach to adoption.\\nLeading + technology players like OpenAI, Google DeepMind, and Meta are actively deploying + and integrating autonomous agent capabilities. These agents demonstrate impressive + abilities, from automating complex web tasks and office chores to orchestrating + multi-step personal projects and generating advertising content. However, their + current limitations are clear: they struggle with sensitive, high-stakes tasks, + complex interfaces, and certain web access restrictions, often requiring human + oversight for critical actions.\\nEnterprise adoption is accelerating, with + a substantial majority of executives already deploying or planning to deploy + AI agents, driven by the perception of significant competitive advantage. This + adoption often follows a pragmatic, risk-mitigated approach, starting with rule-based + processes and incorporating robust human-in-the-loop oversight. The regulatory + environment, particularly the EU AI Act, is shaping deployment strategies, emphasizing + data governance, transparency, and accountability, which are critical for building + trust. Despite the hype, confidence in fully autonomous agents has declined, + highlighting a \u201Ctrust gap\u201D that organizations must address through + ethical governance and explainability.\\nThe realistic return on investment + (ROI) for AI agents extends beyond immediate cost reductions to encompass substantial + gains in operational efficiency, productivity, enhanced customer experience, + and accelerated innovation. While quantifying ROI can be complex, early adopters + are reporting measurable value and a significant competitive edge. The path + forward for AI Agentic involves a continued focus on addressing technical limitations, + navigating ethical complexities, and fostering a collaborative human-AI operational + model to unlock its full transformative potential, moving beyond aspirational + \u201Cfull autonomy\u201D to achievable, impactful augmentation.\\n## 1. Demystifying + AI Agentic: What Are True Autonomous Agents?\\nThe term \u201CAI Agentic\u201D + has become a prevalent buzzword, often leading to confusion regarding the actual + capabilities of various AI systems. As of July 2025, a clear definition has + emerged, distinguishing between AI-powered tools that enhance existing processes + and genuinely autonomous AI agents that operate with independent decision-making + authority. Understanding this distinction is crucial for strategic planning + and avoiding the \u201Cinnovation theatre\u201D that can fuel unrealistic expectations + [30].\\n### 1.1 Defining AI Agents: From LLMs to Autonomous Decision-Making\\nThe + fundamental concept of AI Agentic, as understood in July 2025, centers on*autonomous + agents*. These are systems designed to receive a high-level goal and then independently + execute multiple operations, including making choices and potentially initiating + actions such as payments [30]. This capability represents a significant advancement + beyond traditional AI tools. For instance, OpenAI describes its agents, including + the newly integrated ChatGPT Agent (formerly Operator), as \u201CAIs capable + of doing work for you independently \u2014you give it a task and it will execute + it\u201D [1]. This directly aligns with the principle of autonomous execution.\\nPress + enter or click to view image in full size\\n![] \\nThe Agentic mode in ChatGPT\\nLarge + Language Models (LLMs) serve as foundational components, acting as the \u201Cbrains\u201D + or \u201Creasoning engines\u201D that enable these autonomous solutions. However, + it is important to recognize that LLMs themselves are not the agents. The evolution + of AI has seen LLMs overcome their inherent limitations, such as poor calculation + abilities or fixed training data, by gaining the capacity to interface with + external tools. This includes calling a calculator for complex mathematical + operations, retrieving information from external document collections via architectures + like Retrieval-Augmented Generation (RAG), connecting to the internet, or interfacing + with email and notification systems [30]. This integration of tools was a pivotal + step towards agentic behavior.\\nThe transition from LLMs as standalone processing + units to LLMs functioning as enablers within larger, orchestrated autonomous + systems signifies a fundamental architectural evolution in AI. This shift moves + the primary focus of innovation from merely scaling model size to designing + sophisticated system architectures and enhancing reasoning capabilities. For + businesses, this implies that successful AI agent adoption will require more + than simply acquiring access to the latest LLM. It necessitates a strategic + emphasis on integrating these models into robust, tool-augmented architectures + and developing a deep understanding of the system-level design required for + truly autonomous operation. The future gains in AI are increasingly expected + to come from \u201Csmarter algorithms\u201D and architectural innovation, rather + than solely from building bigger models [2].\\nPress enter or click to view + image in full size\\n![] \\nMulti agents interaction \u2014source[https://blog.langchain.com/langgraph-multi-agent-workflows/] + \\nThe rapid proliferation of the \u201CAI Agentic\u201D term, often misapplied + to less autonomous systems, has created considerable market excitement and a + sense of \u201Cfear of missing out\u201D (FOMO) among top management [30]. This + dynamic necessitates a clear, grounded definition for effective strategic decision-making. + The competitive pressures in the AI market incentivize companies to rebrand + existing automation or LLM-powered tools as \u201Cagents\u201D to capitalize + on the hype. This marketing-driven terminology can obscure the true capabilities + for non-technical executives. Consequently, a critical component of any organization\u2019s + AI strategy must involve thorough due diligence and an internal understanding + of the actual functionalities of any \u201Cagent\u201D solution. Without this + clarity, organizations risk misallocating resources, setting unrealistic expectations, + and experiencing disillusionment, which could undermine broader AI adoption + efforts.\\n### 1.2 The Critical Distinction: Programmed Tools vs. Autonomous + Agents\\nA clear demarcation exists between programmed AI tools and truly autonomous + AI agents, a distinction articulated through a software versioning analogy in + the initial discussion.\\n**Programmed Agents (Version 2.1.0): The Achievable + Automation**\\nThese systems are fundamentally \u201CRPA on steroids or RPA + powered by GenAI\u201D [30]. They integrate LLMs with other LLMs or broader + IT systems but operate with \u201Czero freedom of decision\u201D [30]. Their + execution is \u201Cguided and controlled,\u201D strictly following \u201Cpredetermined + tracks\u201D [30]. Frameworks such as LangChain and LangGraph are instrumental + in developing these more complex, multi-step workflows. Typical applications + include automated document processing, email sorting, meeting summaries, and + workflow management, where tasks are specific and circumscribed [4]. These solutions + are \u201Cextraordinary and can truly help and improve many processes\u201D + [30], delivering significant, measurable value within established operational + frameworks.\\n**Autonomous Agents (Version 4.0.0): The Aspirational Frontier**\\nThese + represent the genuine \u201CAgentic AI\u201D under discussion in 2025, characterized + by true \u201Cdecision-making autonomy\u201D [30]. The user provides a high-level + goal, such as \u201Cplan a weekend trip,\u201D and the agent independently performs + all necessary operations, including navigating the internet, making choices, + and potentially initiating payments [30]. OpenAI\u2019s ChatGPT Agent, launched + on July 17, 2025, is presented as a leading example of this paradigm shift.1 + Other key players include Google DeepMind, with AlphaEvolve optimizing algorithms + and Project Mariner automating web tasks, and Anthropic, which develops Claude + agents for complex problem-solving [5].\\nThe versioning analogy used to illustrate + this evolution is as follows from [30]:\\n* **Version 2.1.0:**Represents programmed + agents, where individual LLMs are guided to perform specific, circumscribed + tasks.\\n* **Version 3.0.0:**Marks the emergence of reasoning engines, such + as OpenAI\u2019s o1 (launched September 2024), which natively integrate complex + architectures to increasingly approximate human-like reasoning.\\n* **Version + 3.1.0:**Encompasses open-source advancements in reasoning, notably accelerated + by projects like DeepSeek\u2019s R1.\\n* **Version 4.0.0:**Denotes truly autonomous + agents that not only have access to tools but \u201Cautonomously choose which + to use and which decisions to take\u201D .\\nThe evolution from programmed tools + to autonomous agents signifies a fundamental shift from the automation of*steps*to + the automation of*goals*. This fundamentally alters how businesses can approach + process optimization and problem-solving. Programmed agents are designed\\nSummary: + None\\n\\nResolved Search Type: neural\\nCostDollars: total=0.015\\n - search: + {'neural': 0.005}\\n - contents: {'text': 0.01}\"}],\"model\":\"gpt-4o-mini\",\"tool_choice\":\"auto\",\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"exa_search_tool\",\"description\":\"Search + the internet using Exa\",\"strict\":true,\"parameters\":{\"properties\":{\"search_query\":{\"description\":\"Mandatory + search query you want to use to search the internet\",\"title\":\"Search Query\",\"type\":\"string\"},\"start_published_date\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}],\"default\":null,\"description\":\"Start + date for the search\",\"title\":\"Start Published Date\"},\"end_published_date\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}],\"default\":null,\"description\":\"End + date for the search\",\"title\":\"End Published Date\"},\"include_domains\":{\"anyOf\":[{\"items\":{\"type\":\"string\"},\"type\":\"array\"},{\"type\":\"null\"}],\"default\":null,\"description\":\"List + of domains to include in the search\",\"title\":\"Include Domains\"}},\"required\":[\"search_query\",\"start_published_date\",\"end_published_date\",\"include_domains\"],\"type\":\"object\",\"additionalProperties\":false}}}]}" + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '182647' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NfMy7GCzjcOsghGYl8ZQANo7L8g\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413312,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"### Summary of Key Findings on Autonomous + AI Agents (2025)\\n\\n**Overview**\\nThe advancements in autonomous AI agents + in 2025 highlight their potential to significantly enhance productivity and + operational efficiency across various industries. Alongside this potential, + emerging challenges in ethical governance and reliability are becoming increasingly + apparent.\\n\\n#### **1. Enhancing Productivity and Operational Efficiency**\\n- + **Integration into Workflows**: Autonomous AI agents are recognized as transformative + tools, shifting from standard automation to sophisticated entities capable + of executing complex tasks autonomously. They eliminate mundane tasks, enabling + employees to focus on higher-level initiatives.\\n- **Substantial ROI**: Companies + leveraging AI agents report efficiency gains of 30-80%, with measurable increases + in productivity, cost savings, and improved decision-making processes. For + instance, PwC's survey highlights that 66% of businesses adopting AI agents + see tangible productivity benefits.\\n- **Examples of Applications**: Industries + such as finance and healthcare are adopting AI agents for tasks like processing + invoices, managing patient schedules, and conducting complex R&D tasks. These + applications streamline operations and drive down costs.\\n\\n#### **2. Ethical + Governance Challenges**\\n- **Accountability and Trust**: As AI agents take + on more autonomous roles, the necessity for robust governance frameworks becomes + paramount. Organizations must navigate challenges such as unbounded autonomy, + opaque decision-making processes, and risks associated with multi-agent systems.\\n- + **Dynamic Governance Frameworks**: Traditional governance models are inadequate + for managing the fluidity and complexity of AI agents. New approaches must + include continuous monitoring, accountability measures, and mechanisms to + ensure compliance with ethical standards.\\n- **Policy Recommendations**: + Essential guidelines include defining the scope of autonomy for agents, enhancing + transparency in their decision-making processes, and ensuring compliance with + regulatory standards. Research suggests that policymakers should prioritize + evidence gathering through supportive structures like testbeds and sandboxes.\\n\\n#### + **3. Addressing Reliability Concerns**\\n- **Emergent Risks**: The deployment + of AI agents is accompanied by risks including decision drift, coordination + conflicts among multiple agents, and potential ethical violations. The governance + of these systems must adapt to manage and mitigate these risks effectively.\\n- + **Collaboration with Human Oversight**: Despite their capabilities, AI agents + cannot function optimally without human involvement in high-stakes decisions. + Balancing autonomy with human oversight is key to reducing risks and ensuring + reliable performance.\\n\\n#### **Conclusion**\\nThe trajectory of autonomous + AI agents in 2025 reveals a dual-edged sword: while they offer vast potential + to redefine productivity and operational dynamics, they also raise significant + ethical and reliability challenges that necessitate proactive governance and + strategic oversight. As organizations integrate these agents, a focus on establishing + trustworthy frameworks will be critical for harnessing their full capabilities + while safeguarding against potential pitfalls.\",\n \"refusal\": null,\n + \ \"annotations\": []\n },\n \"logprobs\": null,\n \"finish_reason\": + \"stop\"\n }\n ],\n \"usage\": {\n \"prompt_tokens\": 42474,\n \"completion_tokens\": + 543,\n \"total_tokens\": 43017,\n \"prompt_tokens_details\": {\n \"cached_tokens\": + 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_6c0d1490cb\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:29:00 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '28556' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: "{\"messages\":[{\"role\":\"system\",\"content\":\"You are a Planning Agent + observing execution progress. After each step completes, you analyze what happened + and decide whether the remaining plan is still valid.\\n\\nReason step-by-step + about:\\n1. What new information was learned from this step's result\\n2. Whether + the remaining steps still make sense given this new information\\n3. What refinements, + if any, are needed for upcoming steps\\n4. Whether the overall goal has already + been achieved\\n\\nBe conservative about triggering full replans \u2014 only + do so when the remaining plan is fundamentally wrong, not just suboptimal.\"},{\"role\":\"user\",\"content\":\"## + Original task\\n\\n\\n## Expected output\\n\\n\\n## Previously completed steps:\\n + \ Step 1: Conduct research on recent developments in autonomous AI agents in + 2025.\\n Result: ### Recent Developments in Autonomous AI Agents (2025)\\n\\nThis + summary compiles recent findings on the developments in autonomous AI agents + for the year 2025, highlighting their evolution, applications,\\n\\n## Just + completed step 2\\nDescription: Summarize the key findings from the research + conducted on autonomous AI agents, highlighting their potential to enhance productivity + and operational efficiency while addressing emerging challenges related to ethical + governance and reliability.\\nResult: ### Summary of Key Findings on Autonomous + AI Agents (2025)\\n\\n**Overview**\\nThe advancements in autonomous AI agents + in 2025 highlight their potential to significantly enhance productivity and + operational efficiency across various industries. Alongside this potential, + emerging challenges in ethical governance and reliability are becoming increasingly + apparent.\\n\\n#### **1. Enhancing Productivity and Operational Efficiency**\\n- + **Integration into Workflows**: Autonomous AI agents are recognized as transformative + tools, shifting from standard automation to sophisticated entities capable of + executing complex tasks autonomously. They eliminate mundane tasks, enabling + employees to focus on higher-level initiatives.\\n- **Substantial ROI**: Companies + leveraging AI agents report efficiency gains of 30-80%, with measurable increases + in productivity, cost savings, and improved decision-making processes. For instance, + PwC's survey highlights that 66% of businesses adopting AI agents see tangible + productivity benefits.\\n- **Examples of Applications**: Industries such as + finance and healthcare are adopting AI agents for tasks like processing invoices, + managing patient schedules, and conducting complex R&D tasks. These applications + streamline operations and drive down costs.\\n\\n#### **2. Ethical Governance + Challenges**\\n- **Accountability and Trust**: As AI agents take on more autonomous + roles, the necessity for robust governance frameworks becomes paramount. Organizations + must navigate challenges such as unbounded autonomy, opaque decision-making + processes, and risks associated with multi-agent systems.\\n- **Dynamic Governance + Frameworks**: Traditional governance models are inadequate for managing the + fluidity and complexity of AI agents. New approaches must include continuous + monitoring, accountability measures, and mechanisms to ensure compliance with + ethical standards.\\n- **Policy Recommendations**: Essential guidelines include + defining the scope of autonomy for agents, enhancing transparency in their decision-making + processes, and ensuring compliance with regulatory standards. Research suggests + that policymakers should prioritize evidence gathering through supportive structures + like testbeds and sandboxes.\\n\\n#### **3. Addressing Reliability Concerns**\\n- + **Emergent Risks**: The deployment of AI agents is accompanied by risks including + decision drift, coordination conflicts among multiple agents, and potential + ethical violations. The governance of these systems must adapt to manage and + mitigate these risks effectively.\\n- **Collaboration with Human Oversight**: + Despite their capabilities, AI agents cannot function optimally without human + involvement in high-stakes decisions. Balancing autonomy with human oversight + is key to reducing risks and ensuring reliable performance.\\n\\n#### **Conclusion**\\nThe + trajectory of autonomous AI agents in 2025 reveals a dual-edged sword: while + they offer vast potential to redefine productivity and operational dynamics, + they also raise significant ethical and reliability challenges that necessitate + proactive governance and strategic oversight. As organizations integrate these + agents, a focus on establishing trustworthy frameworks will be critical for + harnessing their full capabilities while safeguarding against potential pitfalls.\\n\\n\\nAnalyze + this step's result and provide your observation.\"}],\"model\":\"gpt-4o-mini\",\"response_format\":{\"type\":\"json_schema\",\"json_schema\":{\"schema\":{\"description\":\"Planner's + observation after a step execution completes.\\n\\nReturned by the PlannerObserver + after EVERY step \u2014 not just failures.\\nThe Planner uses this to decide + whether to continue, refine, or replan.\\n\\nBased on PLAN-AND-ACT (Section + 3.3): the Planner observes what the Executor\\ndid and incorporates new information + into the remaining plan.\\n\\nAttributes:\\n step_completed_successfully: + Whether the step achieved its objective.\\n key_information_learned: New + information revealed by this step\\n (e.g., \\\"Found 3 products: A, + B, C\\\"). Used to refine upcoming steps.\\n remaining_plan_still_valid: + Whether pending todos still make sense\\n given the new information. + True does NOT mean no refinement needed.\\n suggested_refinements: Minor + tweaks to upcoming step descriptions.\\n These are lightweight in-place + updates, not a full replan.\\n Example: [\\\"Step 3 should select product + B instead of 'best product'\\\"]\\n needs_full_replan: The remaining plan + is fundamentally wrong and must\\n be regenerated from scratch. Mutually + exclusive with\\n remaining_plan_still_valid (if this is True, that should + be False).\\n replan_reason: Explanation of why a full replan is needed (None + if not).\\n goal_already_achieved: The overall task goal has been satisfied + early.\\n No more steps needed \u2014 skip remaining todos and finalize.\",\"properties\":{\"step_completed_successfully\":{\"description\":\"Whether + the step achieved what it was asked to do\",\"title\":\"Step Completed Successfully\",\"type\":\"boolean\"},\"key_information_learned\":{\"default\":\"\",\"description\":\"What + new information this step revealed\",\"title\":\"Key Information Learned\",\"type\":\"string\"},\"remaining_plan_still_valid\":{\"default\":true,\"description\":\"Whether + the remaining pending todos still make sense given new information\",\"title\":\"Remaining + Plan Still Valid\",\"type\":\"boolean\"},\"suggested_refinements\":{\"anyOf\":[{\"items\":{\"type\":\"string\"},\"type\":\"array\"},{\"type\":\"null\"}],\"description\":\"Minor + tweaks to descriptions of upcoming steps (lightweight, no full replan)\",\"title\":\"Suggested + Refinements\"},\"needs_full_replan\":{\"default\":false,\"description\":\"The + remaining plan is fundamentally wrong and must be regenerated\",\"title\":\"Needs + Full Replan\",\"type\":\"boolean\"},\"replan_reason\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}],\"description\":\"Explanation + of why a full replan is needed\",\"title\":\"Replan Reason\"},\"goal_already_achieved\":{\"default\":false,\"description\":\"The + overall task goal has been satisfied early; no more steps needed\",\"title\":\"Goal + Already Achieved\",\"type\":\"boolean\"}},\"required\":[\"step_completed_successfully\",\"key_information_learned\",\"remaining_plan_still_valid\",\"suggested_refinements\",\"needs_full_replan\",\"replan_reason\",\"goal_already_achieved\"],\"title\":\"StepObservation\",\"type\":\"object\",\"additionalProperties\":false},\"name\":\"StepObservation\",\"strict\":true}},\"stream\":false}" + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '7712' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-helper-method: + - beta.chat.completions.parse + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NfpGkucYiZb5268lDmNexWbKHty\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413341,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"{\\\"step_completed_successfully\\\":true,\\\"key_information_learned\\\":\\\"The + research on autonomous AI agents in 2025 reveals that they enhance productivity + and operational efficiency while also presenting ethical governance and reliability + challenges.\\\",\\\"remaining_plan_still_valid\\\":true,\\\"suggested_refinements\\\":[\\\"Consider + expanding on specific case studies of businesses that successfully implemented + autonomous AI agents, particularly in finance and healthcare.\\\"],\\\"needs_full_replan\\\":false,\\\"replan_reason\\\":null,\\\"goal_already_achieved\\\":false}\",\n + \ \"refusal\": null,\n \"annotations\": []\n },\n \"logprobs\": + null,\n \"finish_reason\": \"stop\"\n }\n ],\n \"usage\": {\n \"prompt_tokens\": + 1371,\n \"completion_tokens\": 99,\n \"total_tokens\": 1470,\n \"prompt_tokens_details\": + {\n \"cached_tokens\": 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:29:02 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '1656' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: '{"messages":[{"role":"system","content":"You are refining upcoming plan + steps based on new information. Update the step descriptions to be more specific + and actionable given what was learned. Keep the same step numbers.\n\nRespond + with one line per step in the format:\nStep N: "},{"role":"user","content":"## + New information learned\nThe research on autonomous AI agents in 2025 reveals + that they enhance productivity and operational efficiency while also presenting + ethical governance and reliability challenges.\n\n## Suggested refinements\nConsider + expanding on specific case studies of businesses that successfully implemented + autonomous AI agents, particularly in finance and healthcare.\n\n## Current + pending steps\n\n\nUpdate the step descriptions to incorporate the new information."}],"model":"gpt-4o-mini"}' + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '838' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6Nfr1259U15GetiyfADYjGADAQxx\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413343,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"Step 1: Conduct a literature review + on recent research regarding autonomous AI agents in 2025, focusing on their + impact on productivity and operational efficiency. \\nStep 2: Identify and + compile a list of specific case studies from businesses in finance and healthcare + that have successfully integrated autonomous AI agents and summarize their + outcomes. \\nStep 3: Analyze the ethical governance and reliability challenges + faced by these businesses in implementing autonomous AI agents and propose + potential solutions. \\nStep 4: Develop a best practices guideline for organizations + looking to adopt autonomous AI agents, incorporating insights from the case + studies and ethical considerations. \\nStep 5: Organize a workshop with industry + experts to discuss the findings from the research and case studies, sharing + strategies for overcoming governance and reliability challenges. \",\n \"refusal\": + null,\n \"annotations\": []\n },\n \"logprobs\": null,\n + \ \"finish_reason\": \"stop\"\n }\n ],\n \"usage\": {\n \"prompt_tokens\": + 134,\n \"completion_tokens\": 150,\n \"total_tokens\": 284,\n \"prompt_tokens_details\": + {\n \"cached_tokens\": 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:29:05 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '2335' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: "{\"messages\":[{\"role\":\"system\",\"content\":\"You are Research Analyst. + You have completed a multi-step task. Synthesize the results from all steps + into a single, coherent final response that directly addresses the original + task. Do NOT list step numbers or say 'Step 1 result'. Produce a clean, polished + answer as if you did it all at once.\"},{\"role\":\"user\",\"content\":\"## + Original Task\\nResearch the current state of autonomous AI agents in 2025. + Search for recent developments, then summarize the key findings.\\n\\n## Results + from each step\\nStep 1 (Conduct research on recent developments in autonomous + AI agents in 2025.):\\n### Recent Developments in Autonomous AI Agents (2025)\\n\\nThis + summary compiles recent findings on the developments in autonomous AI agents + for the year 2025, highlighting their evolution, applications, challenges, and + implications across various industries.\\n\\n#### Key Articles and Insights:\\n\\n1. + **AI Agents in 2025: Expectations vs. Reality** \\n - **Source**: [IBM](https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality) + \ \\n - **Published**: February 26, 2025 \\n - **Summary**: The article + discusses the transition from large language models (LLMs) to agentic AI, emphasizing + autonomous agents' potential to reshape work dynamics. Experts from IBM note + that while there is enthusiasm about AI agents, many current applications are + still in their infancy, requiring refinement in planning and reasoning capabilities + to reach true autonomy.\\n\\n2. **Autonomous AI Agents Are Reshaping the Business + Landscape** \\n - **Source**: [KPMG](https://kpmg.com/sk/en/insights/2025/06/autonomous-ai-agents-reshape-business-landscape.html) + \ \\n - **Published**: June 25, 2025 \\n - **Summary**: This report outlines + how agentic AI is transitioning from theory into practical use, impacting sectors + like finance and healthcare. It describes agentic AI as a \\\"digital workforce\\\" + that enhances operational efficiency and decision-making processes through autonomous + task management. \\n\\n3. **The Rise of Autonomous AI Agents: Automating Complex + Tasks** \\n - **Source**: [ResearchGate](https://www.researchgate.net/publication/393620165_The_Rise_of_Autonomous_AI_Agents_Automating_Complex_Tasks) + \ \\n - **Published**: July 11, 2025 \\n - **Summary**: This academic paper + explores the technological foundations and applications of autonomous agents + in various fields, noting their ability to autonomously resolve multi-step tasks. + It addresses the ethical considerations linked to the increasing autonomy of + these systems.\\n\\n4. **Developments in AI Agents: Q1 2025 Landscape Analysis** + \ \\n - **Source**: [The Science of Machine Learning & AI](https://www.ml-science.com/blog/2025/4/17/developments-in-ai-agents-q1-2025-landscape-analysis) + \ \\n - **Published**: April 17, 2025 \\n - **Summary**: Highlights significant + advancements in AI agent technology, including their capability for autonomous + task achievement. The article underscores ongoing challenges related to reliability, + safety, and ethical governance in AI performance.\\n\\n5. **Agentic AI in Software + Development** \\n - **Source**: [SculptSoft](https://www.sculptsoft.com/agentic-ai-in-action-how-autonomous-ai-agents-are-changing-software-development-in-2025/) + \ \\n - **Published**: October 15, 2025 \\n - **Summary**: Discusses how + autonomous AI agents are revolutionizing software development by automating + the development lifecycle\u2014planning, coding, testing, and deployment\u2014thus + enabling teams to innovate faster while reducing manual workloads.\\n\\n6. **AI + Agents Arrived in 2025 \u2013 Here\u2019s What Happened and the Challenges Ahead** + \ \\n - **Source**: [The Conversation](https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325) + \ \\n - **Published**: December 29, 2025 \\n - **Summary**: The article + reviews the major milestones reached in 2025, including the introduction of + protocols to enable agents to communicate and use external tools. It also addresses + the societal implications, such as the need for enhanced governance and ethical + considerations in AI deployment.\\n\\n7. **2025 Buildings Show: AI Agents Become + 'Talk of the Town'** \\n - **Source**: [ConstructConnect](https://news.constructconnect.com/2025-buildings-show-ai-agents-become-talk-of-the-town) + \ \\n - **Published**: December 11, 2025 \\n - **Summary**: At the Buildings + Show, digital innovation in construction technology showcased agentic AI's capability + to monitor and manage project tasks autonomously, streamlining operations in + the sector.\\n\\n### Conclusion\\nThe advancements in autonomous AI agents in + 2025 exhibit a shift towards integrating AI into real-world applications, enhancing + productivity across various industries. While there is significant potential + for these agents to improve efficiency and decision-making, challenges related + to ethical considerations, governance, and reliability remain pivotal in guiding + the future of AI development.\\n\\nStep 2 (Summarize the key findings from the + research conducted on autonomous AI agents, highlighting their potential to + enhance productivity and operational efficiency while addressing emerging challenges + related to ethical governance and reliability.):\\n### Summary of Key Findings + on Autonomous AI Agents (2025)\\n\\n**Overview**\\nThe advancements in autonomous + AI agents in 2025 highlight their potential to significantly enhance productivity + and operational efficiency across various industries. Alongside this potential, + emerging challenges in ethical governance and reliability are becoming increasingly + apparent.\\n\\n#### **1. Enhancing Productivity and Operational Efficiency**\\n- + **Integration into Workflows**: Autonomous AI agents are recognized as transformative + tools, shifting from standard automation to sophisticated entities capable of + executing complex tasks autonomously. They eliminate mundane tasks, enabling + employees to focus on higher-level initiatives.\\n- **Substantial ROI**: Companies + leveraging AI agents report efficiency gains of 30-80%, with measurable increases + in productivity, cost savings, and improved decision-making processes. For instance, + PwC's survey highlights that 66% of businesses adopting AI agents see tangible + productivity benefits.\\n- **Examples of Applications**: Industries such as + finance and healthcare are adopting AI agents for tasks like processing invoices, + managing patient schedules, and conducting complex R&D tasks. These applications + streamline operations and drive down costs.\\n\\n#### **2. Ethical Governance + Challenges**\\n- **Accountability and Trust**: As AI agents take on more autonomous + roles, the necessity for robust governance frameworks becomes paramount. Organizations + must navigate challenges such as unbounded autonomy, opaque decision-making + processes, and risks associated with multi-agent systems.\\n- **Dynamic Governance + Frameworks**: Traditional governance models are inadequate for managing the + fluidity and complexity of AI agents. New approaches must include continuous + monitoring, accountability measures, and mechanisms to ensure compliance with + ethical standards.\\n- **Policy Recommendations**: Essential guidelines include + defining the scope of autonomy for agents, enhancing transparency in their decision-making + processes, and ensuring compliance with regulatory standards. Research suggests + that policymakers should prioritize evidence gathering through supportive structures + like testbeds and sandboxes.\\n\\n#### **3. Addressing Reliability Concerns**\\n- + **Emergent Risks**: The deployment of AI agents is accompanied by risks including + decision drift, coordination conflicts among multiple agents, and potential + ethical violations. The governance of these systems must adapt to manage and + mitigate these risks effectively.\\n- **Collaboration with Human Oversight**: + Despite their capabilities, AI agents cannot function optimally without human + involvement in high-stakes decisions. Balancing autonomy with human oversight + is key to reducing risks and ensuring reliable performance.\\n\\n#### **Conclusion**\\nThe + trajectory of autonomous AI agents in 2025 reveals a dual-edged sword: while + they offer vast potential to redefine productivity and operational dynamics, + they also raise significant ethical and reliability challenges that necessitate + proactive governance and strategic oversight. As organizations integrate these + agents, a focus on establishing trustworthy frameworks will be critical for + harnessing their full capabilities while safeguarding against potential pitfalls.\\n\\nSynthesize + these results into a single, coherent final answer.\"}],\"model\":\"gpt-4o-mini\",\"response_format\":{\"type\":\"json_schema\",\"json_schema\":{\"schema\":{\"properties\":{\"topic\":{\"description\":\"The + research topic\",\"title\":\"Topic\",\"type\":\"string\"},\"key_findings\":{\"description\":\"List + of 3-5 key findings\",\"items\":{\"type\":\"string\"},\"title\":\"Key Findings\",\"type\":\"array\"},\"conclusion\":{\"description\":\"A + brief conclusion paragraph\",\"title\":\"Conclusion\",\"type\":\"string\"}},\"required\":[\"topic\",\"key_findings\",\"conclusion\"],\"title\":\"ResearchSummary\",\"type\":\"object\",\"additionalProperties\":false},\"name\":\"ResearchSummary\",\"strict\":true}},\"stream\":false}" + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '9244' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-helper-method: + - beta.chat.completions.parse + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6Nfuj5WDh8F9ZKOc8EID5zFSvbhc\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413346,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"{\\\"topic\\\":\\\"Current State of + Autonomous AI Agents in 2025\\\",\\\"key_findings\\\":[\\\"Autonomous AI agents + are transforming industries by automating complex tasks, significantly enhancing + productivity and operational efficiency.\\\",\\\"Businesses implementing these + agents report efficiency gains of 30-80%, particularly in sectors like finance + and healthcare where tasks are streamlined.\\\",\\\"The deployment of autonomous + AI agents raises critical ethical governance challenges, necessitating robust + frameworks to ensure accountability and manage risks.\\\",\\\"Reliability + concerns persist, including decision drift and coordination issues among multiple + agents, highlighting the need for a balance between autonomy and human oversight.\\\"],\\\"conclusion\\\":\\\"The + advancements observed in autonomous AI agents in 2025 indicate a significant + shift towards their integration in real-world applications across various + industries. While the potential for these agents to drive efficiency and productivity + is substantial, pressing challenges related to ethical governance and reliability + must be addressed. Establishing solid governance frameworks and ensuring accountability + will be crucial in harnessing their capabilities responsibly.\\\"}\",\n \"refusal\": + null,\n \"annotations\": []\n },\n \"logprobs\": null,\n + \ \"finish_reason\": \"stop\"\n }\n ],\n \"usage\": {\n \"prompt_tokens\": + 1777,\n \"completion_tokens\": 188,\n \"total_tokens\": 1965,\n \"prompt_tokens_details\": + {\n \"cached_tokens\": 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:29:09 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '3011' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +version: 1 diff --git a/lib/crewai/tests/cassettes/agents/TestResponseFormatWithKickoff.test_kickoff_response_format_without_planning.yaml b/lib/crewai/tests/cassettes/agents/TestResponseFormatWithKickoff.test_kickoff_response_format_without_planning.yaml new file mode 100644 index 000000000..e4304248b --- /dev/null +++ b/lib/crewai/tests/cassettes/agents/TestResponseFormatWithKickoff.test_kickoff_response_format_without_planning.yaml @@ -0,0 +1,228 @@ +interactions: +- request: + body: '{"messages":[{"role":"system","content":"You are Math Assistant. A precise + math assistant that always returns structured data\nYour personal goal is: Solve + math problems and return structured results"},{"role":"user","content":"\nCurrent + Task: What is 15 + 27?\n\nProvide your complete response:"}],"model":"gpt-4o-mini","response_format":{"type":"json_schema","json_schema":{"schema":{"properties":{"answer":{"description":"The + numeric answer","title":"Answer","type":"integer"},"explanation":{"description":"Brief + explanation of the solution","title":"Explanation","type":"string"}},"required":["answer","explanation"],"title":"MathResult","type":"object","additionalProperties":false},"name":"MathResult","strict":true}},"stream":false}' + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '739' + content-type: + - application/json + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-helper-method: + - beta.chat.completions.parse + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6NfzfSu3KkFRgIbm3Z222RJMTPy9\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413351,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"{\\\"answer\\\":42,\\\"explanation\\\":\\\"To + find the sum of 15 and 27, we simply add the two numbers together: 15 + 27 + = 42.\\\"}\",\n \"refusal\": null,\n \"annotations\": []\n },\n + \ \"logprobs\": null,\n \"finish_reason\": \"stop\"\n }\n ],\n + \ \"usage\": {\n \"prompt_tokens\": 117,\n \"completion_tokens\": 37,\n + \ \"total_tokens\": 154,\n \"prompt_tokens_details\": {\n \"cached_tokens\": + 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:29:12 GMT + Server: + - cloudflare + Set-Cookie: + - SET-COOKIE-XXX + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '976' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +- request: + body: '{"messages":[{"role":"system","content":"You are Math Assistant. A precise + math assistant that always returns structured data\nYour personal goal is: Solve + math problems and return structured results"},{"role":"user","content":"\nCurrent + Task: What is 15 + 27?\n\nProvide your complete response:"}],"model":"gpt-4o-mini","response_format":{"type":"json_schema","json_schema":{"schema":{"properties":{"answer":{"description":"The + numeric answer","title":"Answer","type":"integer"},"explanation":{"description":"Brief + explanation of the solution","title":"Explanation","type":"string"}},"required":["answer","explanation"],"title":"MathResult","type":"object","additionalProperties":false},"name":"MathResult","strict":true}},"stream":false}' + headers: + User-Agent: + - X-USER-AGENT-XXX + accept: + - application/json + accept-encoding: + - ACCEPT-ENCODING-XXX + authorization: + - AUTHORIZATION-XXX + connection: + - keep-alive + content-length: + - '739' + content-type: + - application/json + cookie: + - COOKIE-XXX + host: + - api.openai.com + x-stainless-arch: + - X-STAINLESS-ARCH-XXX + x-stainless-async: + - 'false' + x-stainless-helper-method: + - beta.chat.completions.parse + x-stainless-lang: + - python + x-stainless-os: + - X-STAINLESS-OS-XXX + x-stainless-package-version: + - 1.83.0 + x-stainless-read-timeout: + - X-STAINLESS-READ-TIMEOUT-XXX + x-stainless-retry-count: + - '0' + x-stainless-runtime: + - CPython + x-stainless-runtime-version: + - 3.13.3 + method: POST + uri: https://api.openai.com/v1/chat/completions + response: + body: + string: "{\n \"id\": \"chatcmpl-D6Ng1hVush1SBz7bgwbNvbWC8F3Lm\",\n \"object\": + \"chat.completion\",\n \"created\": 1770413353,\n \"model\": \"gpt-4o-mini-2024-07-18\",\n + \ \"choices\": [\n {\n \"index\": 0,\n \"message\": {\n \"role\": + \"assistant\",\n \"content\": \"{\\\"answer\\\":42,\\\"explanation\\\":\\\"To + find the sum of 15 and 27, you simply add the two numbers together: 15 + 27 + = 42.\\\"}\",\n \"refusal\": null,\n \"annotations\": []\n },\n + \ \"logprobs\": null,\n \"finish_reason\": \"stop\"\n }\n ],\n + \ \"usage\": {\n \"prompt_tokens\": 117,\n \"completion_tokens\": 37,\n + \ \"total_tokens\": 154,\n \"prompt_tokens_details\": {\n \"cached_tokens\": + 0,\n \"audio_tokens\": 0\n },\n \"completion_tokens_details\": + {\n \"reasoning_tokens\": 0,\n \"audio_tokens\": 0,\n \"accepted_prediction_tokens\": + 0,\n \"rejected_prediction_tokens\": 0\n }\n },\n \"service_tier\": + \"default\",\n \"system_fingerprint\": \"fp_f4ae844694\"\n}\n" + headers: + CF-RAY: + - CF-RAY-XXX + Connection: + - keep-alive + Content-Type: + - application/json + Date: + - Fri, 06 Feb 2026 21:29:14 GMT + Server: + - cloudflare + Strict-Transport-Security: + - STS-XXX + Transfer-Encoding: + - chunked + X-Content-Type-Options: + - X-CONTENT-TYPE-XXX + access-control-expose-headers: + - ACCESS-CONTROL-XXX + alt-svc: + - h3=":443"; ma=86400 + cf-cache-status: + - DYNAMIC + openai-organization: + - OPENAI-ORG-XXX + openai-processing-ms: + - '1057' + openai-project: + - OPENAI-PROJECT-XXX + openai-version: + - '2020-10-01' + x-openai-proxy-wasm: + - v0.1 + x-ratelimit-limit-requests: + - X-RATELIMIT-LIMIT-REQUESTS-XXX + x-ratelimit-limit-tokens: + - X-RATELIMIT-LIMIT-TOKENS-XXX + x-ratelimit-remaining-requests: + - X-RATELIMIT-REMAINING-REQUESTS-XXX + x-ratelimit-remaining-tokens: + - X-RATELIMIT-REMAINING-TOKENS-XXX + x-ratelimit-reset-requests: + - X-RATELIMIT-RESET-REQUESTS-XXX + x-ratelimit-reset-tokens: + - X-RATELIMIT-RESET-TOKENS-XXX + x-request-id: + - X-REQUEST-ID-XXX + status: + code: 200 + message: OK +version: 1