Skip to main content

Agent evaluation

Evaluate agent behavior with code-based checks and LLM-as-a-judge evaluations, including its tool calls and multi-step reasoning.

Labs

Lab 6: Agent evals

Build evaluations for an agent and score its tool calls, reasoning, and outputs.

How to write your first AI eval

Read a trace, define pass or fail criteria, and build an LLM judge.

Ship real agents: hands-on evals

Trace an agent, diagnose failures, build evals, and verify fixes with experiments.

Up next

Module 7: Post-deployment and monitoring

Monitor agents in production and keep improving them with real-world feedback.