Braintrust

Evals are integral to building quality agents. If you can't measure an agent's output, you can't ship it.
This talk covers how to eval agents with multi-step trajectories involving planning, tool calls, edits, and more.