QABattle

AI Agent Frameworks / Scenario

The AgentExecutor That Never Stops

A LangChain agent burns 14 iterations, crashes on parse errors, and its 'answer' is the iteration-limit sentinel. The tests saw none of it.

Difficulty
Hard
Format
Scenario
Points
200
Estimate
15 min

// MISSION BRIEF

Your Mission

A support agent built with create_tool_calling_agent + AgentExecutor is misbehaving in production: some conversations crash, others grind through the iteration limit and reply with a canned stop message that users see verbatim.

The test suite is green, it only asserts one happy-path answer. Read the setup, the trace, and the test, then dissect the failure modes.

// FIRST CONTACT

Battle teaser

First artifact

agent.py

Why do some conversations crash outright?

  1. Ahandle_parsing_errors is False (the default), so when the model emits output the agent cannot parse, the executor raises instead of feeding the error back for recovery
  2. BThe LLM is rate-limited
  3. CAgentExecutor cannot run more than one conversation
  4. DPython cannot catch this class of exception
Answers, scoring, hints, and the full battle stay sealed.

// SKILL TAGS

langchainagentexecutortool-callingdebuggingai-agents