I noticed something interesting. Whenever an AI agent starts doing strange things, the immediate response of any team is to add more of something.