if you want to learn how we think about evaluating agents - real agents, not simple llm prompts - this blog is a goldmine