I'm noticing our AI agent is making some things up. I would love it if I could simply approve the messages being sent during the "training phase" of the AI agent, and then once it has all. the information to make good responses, I can let it go on its own.