Evals are key to defining quality and preventing regressions, but they only test what teams already know to look for. Once agents reach production, new behaviors and failure modes emerge across more traces than any team can review manually.
This talk introduces active observability: a continuous quality loop that connects production traces, pattern discovery, debugging, and evals. Attendees will learn how they can help their teams turn production evidence into prioritized improvements, validate changes, and measure whether agent quality improves over time.
