feat: generate final answer when max_agent_steps reached

[dcondrey]

Related Issues

• fixes Agent can exit on tool call when max_agent_steps is reached, not guaranteeing a final text answer #10001

Proposed Changes:

Previously, if step_count ≥ max_agent_steps and last_msg ∈ ToolResult, output = ToolResult (no final text).

Now, if step_count ≥ max_agent_steps and last_msg ∈ ToolResult, optionally perform 1 extra LLM call → final_text_answer.

How did you test it?

• Files compile without Python syntax errors
• Fixed critical async bug
• Two new tests in test/components/agents/test_agent.py

Notes for the reviewer

hatch run test:unit test/components/agents/test_agent.py::TestAgent::test_final_answer_on_max_steps_enabled
hatch run test:unit test/components/agents/test_agent.py::TestAgent::test_final_answer_on_max_steps_disabled
hatch run test:unit test/components/agents/test_agent.py::TestAgent::test_exceed_max_steps

Checklist

✓ I have read the contributors guidelines and the code of conduct

• I have updated the related issue with new insights and changes
  ✓ I added unit tests and updated the docstrings
  ✓ I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test: and added ! in case the PR includes breaking changes.
  ✓ I documented my code
  ✓✓ I ran pre-commit hooks and fixed any issue

[vercel]

@dcondrey is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.

[dcondrey]

This fixes a contract break in Agent.run() when max_agent_steps is hit on a tool call.

Previously the agent could exit with a raw ToolResult as last_message.

With final_answer_on_max_steps=True (default), the agent makes one extra, tool-disabled LLM call only in that case to produce a proper final ChatMessage.

Setting it to False preserves the old behavior. Normal runs and step limits are otherwise unchanged.

[sjrl]

@dcondrey thanks for working on this so quickly but we are still discussing internally if this is a pattern we we want to support. For example, an alternative is to pipe the output of the Agent to other downstream components like a ChatGenerator component to generate a final answer instead of adding this complexity inside of the Agent component.