Comment by thephyber

> Reasoning with AI achieves at most changing that one agent's behavior.

Wrong. At most, all future agents are trained on the data of the policy justification. Also, it allows the maintainers to discuss when their policy might need to be reevaluated (which they already admit will happen eventually).