Today I dug deeper into evaluating our agents

What agent actions relate to which workflow rule?

What eval cases (this input, this scenario) tie to which workflow rules?

If the AI gives you an explanation you don’t understand, question it. Aggressively.