AI21 Labs

Walk through SWE-bench, AgentBench, GAIA, and TAU-bench: what they measure, their shared blind spots, and how to interpret agent benchmark numbers.

Describe each step of the agent loop and explain which component, LLM or runtime, is responsible for it

Order the five steps of the canonical agent loop from first to last

Explain the LLM vs runtime split in an agent and why the distinction matters for debugging

Which component is 'the agent' in an LLM-based agent system?

How does LangGraph's StateGraph enable agent loops that a standard DAG pipeline cannot express?

What LangGraph feature allows the agent loop to cycle back to an earlier node rather than terminating?

What conditions produce an infinite agent loop and what runtime safeguards prevent one?

Which runtime safeguards prevent an agent from looping indefinitely?

Describe three patterns for inserting human approval into an agent workflow and their tradeoffs