How to Know If Your AI Agent Is Any Good: Evals With Phoenix & Claude — Laurie Voss, Arize AI
Arize Phoenix enables structured evals to catch AI agent regressions before production
Laurie Voss of Arize AI ran a hands-on workshop showing how to evaluate AI agents using Arize Phoenix and Claude, targeting the common failure mode where an agent seems fine in testing but breaks unexpectedly in production or when prompts are adjusted. The session provided a TypeScript repository with checkpoint-based recovery so attendees could follow along even with setup delays. The talk is practical and tooling-focused rather than a research or product announcement.