I've found that once you run a few agents, the hard part isn't getting answers. It's knowing what happened between them. Seeing the hand-offs makes debugging feel a lot less like guesswork.