diff --git a/lib/crewai/src/crewai/agents/crew_agent_executor.py b/lib/crewai/src/crewai/agents/crew_agent_executor.py
index 0707f59d6..d7b202a23 100644
--- a/lib/crewai/src/crewai/agents/crew_agent_executor.py
+++ b/lib/crewai/src/crewai/agents/crew_agent_executor.py
@@ -1072,6 +1072,7 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
             "result": result,
             "from_cache": from_cache,
             "original_tool": original_tool,
+            "error_occurred": error_event_emitted,
         }
 
     def _append_tool_result_and_check_finality(
@@ -1082,6 +1083,7 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
         result = cast(str, execution_result["result"])
         from_cache = cast(bool, execution_result["from_cache"])
         original_tool = execution_result["original_tool"]
+        error_occurred = execution_result.get("error_occurred", False)
 
         tool_message: LLMMessage = {
             "role": "tool",
@@ -1098,10 +1100,13 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
                 color="green",
             )
 
+        # Don't honor result_as_answer when the tool execution errored;
+        # let the agent reflect on the error instead.
         if (
             original_tool
             and hasattr(original_tool, "result_as_answer")
             and original_tool.result_as_answer
+            and not error_occurred
         ):
             return AgentFinish(
                 thought="Tool result is the final answer",
diff --git a/lib/crewai/src/crewai/experimental/agent_executor.py b/lib/crewai/src/crewai/experimental/agent_executor.py
index b785a102e..29dc24b4d 100644
--- a/lib/crewai/src/crewai/experimental/agent_executor.py
+++ b/lib/crewai/src/crewai/experimental/agent_executor.py
@@ -1635,6 +1635,7 @@ class AgentExecutor(Flow[AgentExecutorState], CrewAgentExecutorMixin):
                             "result": f"Error executing tool: {e}",
                             "from_cache": False,
                             "original_tool": None,
+                            "error_occurred": True,
                         }
                 execution_results = [
                     result for result in ordered_results if result is not None
@@ -1666,10 +1667,14 @@ class AgentExecutor(Flow[AgentExecutorState], CrewAgentExecutorMixin):
                         color="green",
                     )
 
+                # Don't honor result_as_answer when the tool execution errored;
+                # let the agent reflect on the error instead.
+                error_occurred = execution_result.get("error_occurred", False)
                 if (
                     original_tool
                     and hasattr(original_tool, "result_as_answer")
                     and original_tool.result_as_answer
+                    and not error_occurred
                 ):
                     self.state.current_answer = AgentFinish(
                         thought="Tool result is the final answer",
@@ -1704,10 +1709,14 @@ class AgentExecutor(Flow[AgentExecutorState], CrewAgentExecutorMixin):
                     color="green",
                 )
 
+            # Don't honor result_as_answer when the tool execution errored;
+            # let the agent reflect on the error instead.
+            error_occurred = execution_result.get("error_occurred", False)
             if (
                 original_tool
                 and hasattr(original_tool, "result_as_answer")
                 and original_tool.result_as_answer
+                and not error_occurred
             ):
                 # Set the result as the final answer
                 self.state.current_answer = AgentFinish(
@@ -1964,6 +1973,7 @@ class AgentExecutor(Flow[AgentExecutorState], CrewAgentExecutorMixin):
             "result": result,
             "from_cache": from_cache,
             "original_tool": original_tool,
+            "error_occurred": error_event_emitted,
         }
 
     def _extract_tool_name(self, tool_call: Any) -> str:
diff --git a/lib/crewai/src/crewai/tools/tool_usage.py b/lib/crewai/src/crewai/tools/tool_usage.py
index b6ce5adb6..92271c3d7 100644
--- a/lib/crewai/src/crewai/tools/tool_usage.py
+++ b/lib/crewai/src/crewai/tools/tool_usage.py
@@ -108,6 +108,7 @@ class ToolUsage:
         self.action = action
         self.function_calling_llm = function_calling_llm
         self.fingerprint_context = fingerprint_context or {}
+        self._last_execution_errored: bool = False
 
         # Set the maximum parsing attempts for bigger models
         if (
@@ -126,8 +127,10 @@ class ToolUsage:
     def use(
         self, calling: ToolCalling | InstructorToolCalling, tool_string: str
     ) -> str:
+        self._last_execution_errored = False
         if isinstance(calling, ToolUsageError):
             error = calling.message
+            self._last_execution_errored = True
             if self.agent and self.agent.verbose:
                 self._printer.print(content=f"\n\n{error}\n", color="red")
             if self.task:
@@ -138,6 +141,7 @@ class ToolUsage:
             tool = self._select_tool(calling.tool_name)
         except Exception as e:
             error = getattr(e, "message", str(e))
+            self._last_execution_errored = True
             if self.task:
                 self.task.increment_tools_errors()
             if self.agent and self.agent.verbose:
@@ -174,8 +178,10 @@ class ToolUsage:
         Returns:
             The result of the tool execution as a string.
         """
+        self._last_execution_errored = False
         if isinstance(calling, ToolUsageError):
             error = calling.message
+            self._last_execution_errored = True
             if self.agent and self.agent.verbose:
                 self._printer.print(content=f"\n\n{error}\n", color="red")
             if self.task:
@@ -186,6 +192,7 @@ class ToolUsage:
             tool = self._select_tool(calling.tool_name)
         except Exception as e:
             error = getattr(e, "message", str(e))
+            self._last_execution_errored = True
             if self.task:
                 self.task.increment_tools_errors()
             if self.agent and self.agent.verbose:
@@ -410,6 +417,7 @@ class ToolUsage:
                 except Exception as e:
                     self.on_tool_error(tool=tool, tool_calling=calling, e=e)
                     error_event_emitted = True
+                    self._last_execution_errored = True
                     self._run_attempts += 1
                     if self._run_attempts > self._max_parsing_attempts:
                         self._telemetry.tool_usage_error(llm=self.function_calling_llm)
@@ -642,6 +650,7 @@ class ToolUsage:
                 except Exception as e:
                     self.on_tool_error(tool=tool, tool_calling=calling, e=e)
                     error_event_emitted = True
+                    self._last_execution_errored = True
                     self._run_attempts += 1
                     if self._run_attempts > self._max_parsing_attempts:
                         self._telemetry.tool_usage_error(llm=self.function_calling_llm)
diff --git a/lib/crewai/src/crewai/utilities/agent_utils.py b/lib/crewai/src/crewai/utilities/agent_utils.py
index c1a341c39..88be08dd0 100644
--- a/lib/crewai/src/crewai/utilities/agent_utils.py
+++ b/lib/crewai/src/crewai/utilities/agent_utils.py
@@ -1575,11 +1575,13 @@ def execute_single_native_tool_call(
             color="green",
         )
 
-    # Check result_as_answer
+    # Check result_as_answer — but don't honor it when the tool errored;
+    # let the agent reflect on the error instead.
     is_result_as_answer = bool(
         original_tool
         and hasattr(original_tool, "result_as_answer")
         and original_tool.result_as_answer
+        and not error_event_emitted
     )
 
     return NativeToolCallResult(
diff --git a/lib/crewai/src/crewai/utilities/tool_utils.py b/lib/crewai/src/crewai/utilities/tool_utils.py
index 027f136ed..bb258b317 100644
--- a/lib/crewai/src/crewai/utilities/tool_utils.py
+++ b/lib/crewai/src/crewai/utilities/tool_utils.py
@@ -140,7 +140,12 @@ async def aexecute_tool_and_check_finality(
         except Exception as e:
             logger.log("error", f"Error in after_tool_call hook: {e}")
 
-        return ToolResult(modified_result, tool.result_as_answer)
+        # Don't honor result_as_answer when the tool execution errored;
+        # let the agent reflect on the error instead.
+        effective_result_as_answer = (
+            tool.result_as_answer and not tool_usage._last_execution_errored
+        )
+        return ToolResult(modified_result, effective_result_as_answer)
 
     tool_result = i18n.errors("wrong_tool_name").format(
         tool=sanitized_tool_name,
@@ -261,7 +266,12 @@ def execute_tool_and_check_finality(
         except Exception as e:
             logger.log("error", f"Error in after_tool_call hook: {e}")
 
-        return ToolResult(modified_result, tool.result_as_answer)
+        # Don't honor result_as_answer when the tool execution errored;
+        # let the agent reflect on the error instead.
+        effective_result_as_answer = (
+            tool.result_as_answer and not tool_usage._last_execution_errored
+        )
+        return ToolResult(modified_result, effective_result_as_answer)
 
     tool_result = i18n.errors("wrong_tool_name").format(
         tool=sanitized_tool_name,
diff --git a/lib/crewai/tests/agents/test_agent_executor.py b/lib/crewai/tests/agents/test_agent_executor.py
index c845bd458..26ac65491 100644
--- a/lib/crewai/tests/agents/test_agent_executor.py
+++ b/lib/crewai/tests/agents/test_agent_executor.py
@@ -879,6 +879,80 @@ class TestNativeToolExecution:
         assert len(tool_messages) == 1
         assert tool_messages[0]["tool_call_id"] == "call_1"
 
+    def test_execute_native_tool_result_as_answer_not_honored_on_error(
+        self, mock_dependencies
+    ):
+        """Test that result_as_answer is NOT honored when the tool errors.
+
+        Regression test for https://github.com/crewAIInc/crewAI/issues/5156
+        When a tool with result_as_answer=True raises an exception during
+        native tool execution, the agent should NOT treat the error as
+        the final answer.
+        """
+        executor = AgentExecutor(**mock_dependencies)
+
+        def failing_tool() -> str:
+            raise RuntimeError("Tool execution failed")
+
+        result_tool = Mock()
+        result_tool.name = "failing_tool"
+        result_tool.result_as_answer = True
+        result_tool.max_usage_count = None
+        result_tool.current_usage_count = 0
+
+        executor.original_tools = [result_tool]
+        executor._available_functions = {"failing_tool": failing_tool}
+        executor.state.pending_tool_calls = [
+            {
+                "id": "call_1",
+                "function": {"name": "failing_tool", "arguments": "{}"},
+            },
+        ]
+
+        result = executor.execute_native_tool()
+
+        # The tool errored, so it should NOT be treated as final answer
+        assert result == "native_tool_completed", (
+            "Expected 'native_tool_completed' (not 'tool_result_is_final') "
+            "when a result_as_answer tool errors"
+        )
+        assert not executor.state.is_finished, (
+            "Agent state should NOT be finished when tool errored"
+        )
+
+    def test_execute_native_tool_result_as_answer_honored_on_success(
+        self, mock_dependencies
+    ):
+        """Test that result_as_answer IS honored when the tool succeeds."""
+        executor = AgentExecutor(**mock_dependencies)
+
+        def success_tool() -> str:
+            return "final answer from tool"
+
+        result_tool = Mock()
+        result_tool.name = "success_tool"
+        result_tool.result_as_answer = True
+        result_tool.max_usage_count = None
+        result_tool.current_usage_count = 0
+
+        executor.original_tools = [result_tool]
+        executor._available_functions = {"success_tool": success_tool}
+        executor.state.pending_tool_calls = [
+            {
+                "id": "call_1",
+                "function": {"name": "success_tool", "arguments": "{}"},
+            },
+        ]
+
+        result = executor.execute_native_tool()
+
+        assert result == "tool_result_is_final", (
+            "Expected 'tool_result_is_final' when result_as_answer tool succeeds"
+        )
+        assert executor.state.is_finished is True
+        assert isinstance(executor.state.current_answer, AgentFinish)
+        assert executor.state.current_answer.output == "final answer from tool"
+
     def test_check_native_todo_completion_requires_current_todo(
         self, mock_dependencies
     ):
diff --git a/lib/crewai/tests/tools/test_tool_usage.py b/lib/crewai/tests/tools/test_tool_usage.py
index b68a41666..06af4a874 100644
--- a/lib/crewai/tests/tools/test_tool_usage.py
+++ b/lib/crewai/tests/tools/test_tool_usage.py
@@ -748,6 +748,246 @@ def test_tool_usage_finished_event_with_cached_result():
     assert event.type == "tool_usage_finished"
 
 
+def test_result_as_answer_not_honored_on_tool_error():
+    """Test that result_as_answer is NOT honored when a tool execution errors.
+
+    Regression test for https://github.com/crewAIInc/crewAI/issues/5156
+    When a tool with result_as_answer=True raises an exception, the agent
+    should be able to reflect on the error instead of treating it as the
+    final answer.
+    """
+    from crewai.tools.tool_calling import ToolCalling
+
+    class FailingResultTool(BaseTool):
+        name: str = "Failing Result Tool"
+        description: str = "A tool that always fails but has result_as_answer=True"
+        result_as_answer: bool = True
+
+        def _run(self, **kwargs) -> str:
+            raise ValueError("Something went wrong")
+
+    failing_tool = FailingResultTool().to_structured_tool()
+    assert failing_tool.result_as_answer is True
+
+    mock_agent = MagicMock()
+    mock_agent.key = "test_agent_key"
+    mock_agent.role = "test_agent_role"
+    mock_agent._original_role = "test_agent_role"
+    mock_agent.verbose = False
+    mock_agent.fingerprint = None
+    mock_agent.i18n.tools.return_value = {"name": "Failing Result Tool"}
+    mock_agent.i18n.errors.return_value = "Error: {error}"
+    mock_agent.i18n.slice.return_value = "Available tools: {tool_names}"
+
+    mock_task = MagicMock()
+    mock_task.delegations = 0
+    mock_task.name = "Test Task"
+    mock_task.description = "A test task"
+    mock_task.id = "test-task-id"
+
+    mock_action = MagicMock()
+    mock_action.tool = "failing_result_tool"
+    mock_action.tool_input = "{}"
+
+    tool_usage = ToolUsage(
+        tools_handler=MagicMock(cache=None, last_used_tool=None),
+        tools=[failing_tool],
+        task=mock_task,
+        function_calling_llm=None,
+        agent=mock_agent,
+        action=mock_action,
+    )
+
+    tool_calling = ToolCalling(tool_name="failing_result_tool", arguments={})
+    tool_usage.use(calling=tool_calling, tool_string="Action: failing_result_tool")
+
+    # After a failed execution, the error flag should be set
+    assert tool_usage._last_execution_errored is True, (
+        "Expected _last_execution_errored to be True after tool failure"
+    )
+
+
+def test_result_as_answer_honored_on_tool_success():
+    """Test that result_as_answer IS honored when a tool executes successfully."""
+    from crewai.tools.tool_calling import ToolCalling
+
+    class SuccessfulResultTool(BaseTool):
+        name: str = "Successful Result Tool"
+        description: str = "A tool that succeeds and has result_as_answer=True"
+        result_as_answer: bool = True
+
+        def _run(self, **kwargs) -> str:
+            return "This is the final answer"
+
+    successful_tool = SuccessfulResultTool().to_structured_tool()
+    assert successful_tool.result_as_answer is True
+
+    mock_agent = MagicMock()
+    mock_agent.key = "test_agent_key"
+    mock_agent.role = "test_agent_role"
+    mock_agent._original_role = "test_agent_role"
+    mock_agent.verbose = False
+    mock_agent.fingerprint = None
+    mock_agent.i18n.tools.return_value = {"name": "Successful Result Tool"}
+    mock_agent.i18n.errors.return_value = "Error: {error}"
+    mock_agent.i18n.slice.return_value = "Available tools: {tool_names}"
+
+    mock_task = MagicMock()
+    mock_task.delegations = 0
+    mock_task.name = "Test Task"
+    mock_task.description = "A test task"
+    mock_task.id = "test-task-id"
+
+    mock_action = MagicMock()
+    mock_action.tool = "successful_result_tool"
+    mock_action.tool_input = "{}"
+
+    tool_usage = ToolUsage(
+        tools_handler=MagicMock(cache=None, last_used_tool=None),
+        tools=[successful_tool],
+        task=mock_task,
+        function_calling_llm=None,
+        agent=mock_agent,
+        action=mock_action,
+    )
+
+    tool_calling = ToolCalling(tool_name="successful_result_tool", arguments={})
+    tool_usage.use(calling=tool_calling, tool_string="Action: successful_result_tool")
+
+    # After a successful execution, the error flag should NOT be set
+    assert tool_usage._last_execution_errored is False, (
+        "Expected _last_execution_errored to be False after tool success"
+    )
+
+
+def test_execute_tool_and_check_finality_not_final_on_error():
+    """Test that execute_tool_and_check_finality does NOT set result_as_answer
+    when the tool with result_as_answer=True raises an error.
+
+    Regression test for https://github.com/crewAIInc/crewAI/issues/5156
+    """
+    from crewai.agents.parser import AgentAction
+    from crewai.utilities.tool_utils import execute_tool_and_check_finality
+
+    class FailingFinalTool(BaseTool):
+        name: str = "Failing Final Tool"
+        description: str = "A tool that fails but has result_as_answer=True"
+        result_as_answer: bool = True
+
+        def _run(self, **kwargs) -> str:
+            raise RuntimeError("Tool execution failed")
+
+    failing_tool = FailingFinalTool().to_structured_tool()
+
+    mock_agent = MagicMock()
+    mock_agent.key = "test_agent_key"
+    mock_agent.role = "test_agent_role"
+    mock_agent._original_role = "test_agent_role"
+    mock_agent.verbose = False
+    mock_agent.fingerprint = None
+    mock_agent.i18n.tools.return_value = {"name": "Failing Final Tool"}
+    mock_agent.i18n.errors.return_value = "Error: {error}"
+    mock_agent.i18n.slice.return_value = "Available tools: {tool_names}"
+
+    mock_task = MagicMock()
+    mock_task.delegations = 0
+    mock_task.name = "Test Task"
+    mock_task.description = "A test task"
+    mock_task.id = "test-task-id"
+
+    from crewai.utilities.i18n import I18N
+
+    i18n = I18N()
+
+    action = AgentAction(
+        thought="I need to use this tool",
+        tool="failing_final_tool",
+        tool_input="{}",
+        text="Action: failing_final_tool\nAction Input: {}",
+    )
+
+    result = execute_tool_and_check_finality(
+        agent_action=action,
+        tools=[failing_tool],
+        i18n=i18n,
+        agent_key="test_agent_key",
+        agent_role="test_agent_role",
+        tools_handler=MagicMock(cache=None, last_used_tool=None),
+        task=mock_task,
+        agent=mock_agent,
+        function_calling_llm=None,
+    )
+
+    # The result should NOT be treated as the final answer when the tool errored
+    assert result.result_as_answer is False, (
+        "result_as_answer should be False when tool execution fails, "
+        "even if the tool has result_as_answer=True"
+    )
+
+
+def test_execute_tool_and_check_finality_is_final_on_success():
+    """Test that execute_tool_and_check_finality DOES set result_as_answer
+    when the tool with result_as_answer=True succeeds.
+    """
+    from crewai.agents.parser import AgentAction
+    from crewai.utilities.tool_utils import execute_tool_and_check_finality
+
+    class SuccessfulFinalTool(BaseTool):
+        name: str = "Successful Final Tool"
+        description: str = "A tool that succeeds and has result_as_answer=True"
+        result_as_answer: bool = True
+
+        def _run(self, **kwargs) -> str:
+            return "This is the correct final answer"
+
+    success_tool = SuccessfulFinalTool().to_structured_tool()
+
+    mock_agent = MagicMock()
+    mock_agent.key = "test_agent_key"
+    mock_agent.role = "test_agent_role"
+    mock_agent._original_role = "test_agent_role"
+    mock_agent.verbose = False
+    mock_agent.fingerprint = None
+    mock_agent.i18n.tools.return_value = {"name": "Successful Final Tool"}
+    mock_agent.i18n.errors.return_value = "Error: {error}"
+    mock_agent.i18n.slice.return_value = "Available tools: {tool_names}"
+
+    mock_task = MagicMock()
+    mock_task.delegations = 0
+    mock_task.name = "Test Task"
+    mock_task.description = "A test task"
+    mock_task.id = "test-task-id"
+
+    from crewai.utilities.i18n import I18N
+
+    i18n = I18N()
+
+    action = AgentAction(
+        thought="I need to use this tool",
+        tool="successful_final_tool",
+        tool_input="{}",
+        text="Action: successful_final_tool\nAction Input: {}",
+    )
+
+    result = execute_tool_and_check_finality(
+        agent_action=action,
+        tools=[success_tool],
+        i18n=i18n,
+        agent_key="test_agent_key",
+        agent_role="test_agent_role",
+        tools_handler=MagicMock(cache=None, last_used_tool=None),
+        task=mock_task,
+        agent=mock_agent,
+        function_calling_llm=None,
+    )
+
+    # The result SHOULD be treated as the final answer when the tool succeeds
+    assert result.result_as_answer is True, (
+        "result_as_answer should be True when tool execution succeeds "
+        "and the tool has result_as_answer=True"
+    )
+
+
 def test_tool_error_does_not_emit_finished_event():
     from crewai.tools.tool_calling import ToolCalling