Ask HN: What do u use for agent/agentic evals?

Right now looking at MLFlow/Braintrust but find it hard to compare acrosss versions of agents, and a/b testing of agents, and mcp tools. Also obvious things like runaway agents (stuck in a loop), or token/spend optimalisation.

What do you all use?