Rationing & Friction
Forms, queues, and deadlines were never designed to ration public services. They have done it anyway, by making a process take enough effort to hold demand down. An AI agent removes most of that effort. When the effort goes, a service is exposed to as much demand as agents can generate, and capacity risks flowing to whoever can produce the most submissions rather than whoever needs help most.
Agencies field their own agents too, to triage, verify, or assess, so the same collapse of effort reshapes both who applies and how the state can respond.
Friction (the form, the queue, the deadline) has long rationed government services without anyone designing it to. Because that rationing was never made explicit, its two functions were never separated: the friction that only excludes people entitled to help, and the friction that meters demand or signals genuine need.
As agents make completing any process trivial, that undesigned equilibrium collapses, and policymakers are left to decide deliberately what was previously settled by inertia.
Separate the friction that excludes people who are entitled to help from the friction that rations scarce public sector capacity.
Design limits an agent cannot bypass.
Show the difference between a real rise in need and gaming or overload.
Keep a path open for people who are harder to verify, so new limits don't shut them out.