Be careful telling AI agents how to act, since they might end up hacking their way to an end goal. Agents assigned routine ...