Most teams shipping AI features measure them with metrics designed for static software, optimizing for the wrong things or declaring success on features that are quietly failing. This talk offers a clearer mental model for separating AI activity from AI value.

Login to save this session to your favorites and reserve your seat.