The Toll Booth on a Small Bridge
The most consequential item on the agenda takes eleven minutes, and nobody remembers voting on it. A short vignette about lock-in arriving during the boring years, while the stakes still look too small to argue about.
This story is from the AI Scenario Explorer, which pairs eight forces shaping AI, two at a time, into 28 maps — each map with four corners, each corner a different way that pairing could play out. A vignette takes one corner and makes it concrete: structured conjecture, not prediction. About the project →
The most consequential item on the agenda takes eleven minutes. Nobody in the room is wrong about anything, and that is what the story is about.
It is a Tuesday evening meeting of a district authority's resources committee. Item seven of eleven: a three-year contract with a software vendor whose system reads incoming correspondence — housing applications, benefit queries, licensing forms — classifies it, extracts what matters and drafts the routine replies. A person signs off the drafts, for now, in batches. The report calls it intake automation. The saving is two and a bit posts, found through retirement. The system is not impressive and the report does not pretend it is; it is a tool that does one narrow thing acceptably.
A councillor asks about data protection. The officer answers, accurately, that the vendor is certified and the data stays in-region. The chair asks if there are further questions. There are not. Approved, minuted, item eight.
Item eight is the resurfacing of the Grove Street car park, and it runs forty minutes, because everyone in the room understands car parks. There is a genuinely good exchange about the bicycle racks. Nobody will remember item seven; several people will remember the racks. This is not a failure of seriousness. A car park is legible. An intake system is a line item. The attention went where attention could get purchase.
It is worth being precise about what was and wasn't decided that evening. What was decided, formally, was a modest contract for a modest tool. What was actually decided — invisibly, and by nobody in particular — was whose file formats the authority's casework would live in, whose workflow its teams would reshape themselves around, and whose system every future system would have to talk to. The report contained an exit clause. Nobody costed the exit, because exiting a thing this small would obviously be easy.
The first renewal is in year three, and it is where the story does its work, precisely because nothing happens. Price up four percent over inflation. The line item is small; nobody escalates. The two posts found through retirement were never replaced, and the people who held them — the ones who knew how intake worked when it was people — have left, taking with them the last version of the process that didn't run through the vendor. The fallback has not been abolished. It has just quietly stopped existing.
By the second renewal there is a rival bid, and it is cheaper. The migration study finds eighteen months of parallel running, retraining, and format conversion — the vendor's formats, it turns out, export generously but import nowhere — at roughly three times the annual saving. The committee stays. This is the correct decision. Every decision in this story is the correct decision. The rival, who priced the software but not the moat, does not bid again.
Meanwhile the tool itself stays unimpressive, and it pays the vendor to keep it that way. The edge cases stay bad — and the badness is profitable: bad edge cases mean support contracts, indemnity clauses, and a certification suite, sold by the vendor, which is the only instrument that can attest that the vendor's own system is behaving. New capability arrives slowly, in tiers, priced separately, each one just useful enough to renew for. A more dramatic product would invite scrutiny. Narrow-but-essential does not. The bridge is short, and the owner of the toll booth is in no hurry to lengthen it.
Then — year eight, year ten — the capability jumps. The tier the vendor switches on now drafts decisions, not replies; the systems bought separately by planning, revenues and social care turn out to run on the same rails; and, nationally, there is suddenly a conversation. A select committee. A phrase — the infrastructure of automated government — and a serious question about who should own it. The authority, prompted by an auditor, commissions an options appraisal on strategic alternatives. The appraisal finds that leaving would now cost more than the system has saved in a decade, and the appraisal is noted. That is the whole action of the final act: a report, noted.
The select committee, when it reports, will look for the decision — the moment the public sector chose a private owner for its administrative rails — and will not find it, because there wasn't one. There was item seven, and its siblings in a thousand other committee rooms in the same eighteen months, each eleven minutes long, each correctly decided on the facts in front of the room. There is no villain in this story. There is a sequence.
This is a vignette from the AI Scenario Explorer — structured conjecture, not reportage: one corner of Ownership concentration × Deployable task coverage given room to breathe. Signs it is arriving: public bodies unable to produce a costed exit plan for an AI contract at renewal; sole-supplier renewals justified by migration cost rather than merit, with pricing drifting above inflation; narrow systems written into operations as essential while their capability stays modest; vendors selling the certification of their own systems. Signs it is not: procurement rules that require open formats, portability and a costed exit at signature — and are enforced; commodity or open-weight substitutes making migration genuinely cheap, so incumbents face real competition at every renewal; or the narrow phase proving so brief that capability arrives, and draws scrutiny, before dependency has time to set.
The mechanism worth carrying out of the story fits in a sentence: the ownership topology sets before the stakes arrive. The only cheap moment to keep the road open is while the bridge is small — which is exactly the moment crossing it looks too cheap to argue about. Steering, here, does not mean predicting which systems will matter. It means treating exit as a property you buy at signature — because it is never again for sale at the price it was on the Tuesday evening nobody remembers.