Launch HN: Laminar (YC S24) – We trained a model to catch silent agent failures

Laminar, a YC S24 startup, released flow-1, an AI model trained with reinforcement learning to detect silent failures in agent traces at 23x lower cost than GPT-4o while maintaining comparable detection quality. The model analyzes agent behavior by treating traces like code repositories, allowing it to identify implicit errors such as incorrect button clicks or wrong citations that typical observability tools miss. By automatically analyzing all production traces and grouping similar errors into clusters, Laminar addresses the challenge of scaling agent observability without expensive sampling or manual trace review.