Building an Agentic Eval Pipeline: Battle-Tested Lessons... - Kurtis Van Gent & Haoyu Wang, Google
Agents look impressive in a demo, then quietly get worse after a small change to a prompt, tool, or skill. In this talk, Kurtis Van Gent and Haoyu Wang of Google Cloud Databases explain how to test an AI agent on real user journeys so you catch breaks before your users do.
Agentic AI Foundation