Susan Chang shares how Elastic built a production-grade AI agent evaluation framework, detailing lessons on tracing, LLM-as-a-judge, programmatic evals to prevent regression across engineering teams.