AI Agent Observability: Braintrust's Phil Hetzel Explains

AI agent performance needs new monitoring tools. Phil Hetzel, Head of Solution Engineering at Braintrust, explains the differences between traditional and agent observability. He spoke at an AI Engineer event. What's key here is that AI agents are non-deterministic. This means they can produce many different outputs from the same input. Traditional observability, built for predictable systems, just can't handle this variability. Hetzel has twelve years in consulting and implementation. He previously led the global Databricks business unit at Slalom. This experience gives him a deep understanding of managing complex systems. The bottom line: new approaches are essential to effectively evaluate and ensure the quality of AI agents.