You Can’t Just Run It Twice: Eval-Driven Development with LangSmith

Join us while we hear from Michael Steichen, Engineering Manager at Focused.

This talk explores the importance of Eval-Driven Development when building AI agents powered by language models. Evaluation is a critical step in creating reliable systems. LangSmith provides tools that make this process more manageable and repeatable. Topics include handling: non-deterministic outputs and defining success metrics for agent behavior. As models approach the final 10 to 20 percent of accuracy, systematic evaluation becomes essential for closing the gap between “good enough” and production-ready.