The productivity promise of generative AI had a specific pitch: automate the dull stuff, free humans for higher work. What nobody modelled was the demand-side explosion — what happens when the cost of producing a formal complaint, legal challenge, or freedom-of-information request drops to near zero while the cost of processing each one stays fixed in human hours. That asymmetry is now breaking things. UK local authorities report that average complaint length has ballooned from two pages to twenty; one council has received a single complaint stretching to 200 pages. Transport for London logged a record 4,769 FOI requests in 2025-26, up 5% on its previous peak. Employment tribunal claims opened between April and June surged 28% year-on-year, representing 14,000 single-claimant cases. The correspondence is verbose, peppered with hallucinated case law from foreign jurisdictions, and frequently omits the one thing a complaint needs: what resolution the complainant actually wants. MP offices are buckling under the same weight. Staff in Labour MP Graeme Downie's constituency office are described as "exhausted" and "working more hours than they are paid." The GMB union reports that casework in some MP offices has doubled or more. One senior caseworker estimates half of all incoming emails now involve some form of AI generation. Her office has stopped responding to automated "click to email your MP" campaigns entirely, a triage decision that would have been unthinkable five years ago. The extraction pattern is clean: individuals gain free access to a tool that produces professional-sounding output at negligible personal cost. The processing burden lands entirely on under-resourced public institutions bound by statutory response deadlines. Every pound a council spends wading through a 200-page AI-generated parking appeal is a pound not spent on services. The Information Commissioner's Office has acknowledged the problem but still requires human responses to AI-generated requests, even those containing inaccuracies. Some organisations are beginning to adapt. Ombudsman services are imposing word limits. Wiltshire council is exploring AI summarisation tools. Researchers have floated fees for submitting requests or bot-blocking measures. But each countermeasure introduces its own costs — financial, democratic, or reputational. Charging for FOI requests, for instance, would gut a transparency mechanism that exists precisely because access should be free. The deeper problem is structural. AI tools are optimised to be validating and supportive, which gives users misplaced confidence to pursue cases they might otherwise have dropped. One tech startup founder expects to spend £30,000 defending an employment tribunal case they believe was initiated with AI-assisted correspondence by a former employee dismissed on performance grounds. The 70,000 open single claims sitting in the tribunal system at the end of June represent real human and financial costs on both sides. Researchers at the Hertie School have labelled this "agentic flooding" — a term that captures the systemic nature of what is happening. This is not a technology problem awaiting a technology fix. It is a governance design problem: public systems built for a world where producing formal correspondence required effort are now operating in a world where it does not. The mismatch will widen until someone redesigns the intake architecture, not just the response tools.
