I Evaluated My AI Agent. Three Decisions Were Wrong.

In my last post, I built a customer support agent with conditional routing, RAG-grounded responses, and human-in-the-loop approval for billing actions. I ended that post by listing evaluations as the