It's not that AI agents will multiply our power if only we can constrain them from doing unexpected and unwelcome things, but that their frustrating unreliability reflects the broader point of all AI, which is to strip us of agency altogether.