This is one of the most important system-level AI posts I’ve read in a while. Many public services are not designed to be accessible, they are designed to be survivable: complexity and bad UX function as informal rationing. AI agents will erode that fast, not by changing eligibility, but by making search, form-filling, appeals, and persistence essentially free.
Some friction will remain (identity checks, authentication, in-person verification), but once everything else becomes effortless those choke points will come under pressure, and where governments do not streamline them, third parties will build workarounds.
A current UK preview is the driving test booking mess: bots and middlemen (including apps that auto-check and auto-book cancellations) exploit scarcity, scoop up slots faster than humans, and help fuel reselling at inflated prices, forcing government into an anti-bot crackdown while the backlog persists.
So we might see something interesting: less “AI automating the state,” more AI automating citizens’ ability to claim from the state, at scale. If the state does not automate internally to absorb the surge, the voice-responsiveness gap will widen, and frustration and distrust will predictably rise.