Enterprise AI evaluation shifts from single-conversation scoring to cohort comparison

Leaders from LangChain, Conviva, and CoreWeave said at VB Transform 2026 that enterprises are moving away from scoring individual AI agent conversations in isolation—which can appear flawless but mask broken products—toward comparing groups of users against a baseline. The industry is also adopting cheaper, narrower judge models instead of larger ones.