Online workshop: Find the AI failures that testing misses
Most AI quality issues appear first as support tickets, vague feedback, or production behavior that never came up during testing. Topics gives you a way to read those signals before they escalate: it reads production traces and groups them into named patterns that show what people are trying to do, where interactions are failing, and how experiences are landing.
In this session, Jess Wang walks through how PMs can use those production patterns to build evals, investigate regressions, and improve AI quality continuously.
What you'll learn
How Topics automatically discovers patterns across production AI traces
What the built-in Task, Issues, and Sentiment facets reveal about production performance
How to turn a Topics cluster into a dataset, an eval, and a validated fix
What a PM-owned AI quality review workflow looks like in Braintrust