People notice a robot when it is announced and stop noticing once it is audited

When Britain talks about machines taking work, it talks about moments. A humanoid robot on a stage. A chief executive's prediction. A launch event, a demo, a clip of an arm doing something uncanny. Meanwhile, last September, a document recording the real thing, measured and audited, was filed for a British grocer, and the coverage lasted about a day and reported the revenue. That is not a failing of the journalists involved: the trade press did its job, reporting results as results. It is something more interesting, and more human. It is a fact about what attention is built to catch.

Start with the charitable read, because the behaviour makes sense. An announcement is an event: it has a date, a stage, a face, a before and an after. You can react to it, share it, be early to it. Being the person who noticed carries a little status, and noticing is social long before it is analytical. A ratio moving from 240 to 267 has none of that equipment. There is no moment at which noticing it makes you interesting at dinner. The general findings underneath this, that human attention goes to the sudden and the social over the gradual and the ambient, are among the oldest and best-replicated in the behavioural sciences, and none of it needs a laboratory to recognise. We are built for events. Substitution, it turns out, is not an event. It is a gradient.

You can watch the mismatch shape Britain's argument about work. Hatfield, Ocado's oldest warehouse, made national news when it closed in 2023, because a closure is an event with a car park and a padlock. The published productivity series that explains the closure, climbing quietly in filings for years before and after, passed without remark. So the argument forms around the wrong data type: we debate the padlock and ignore the gradient, then are surprised by the next padlock. The pattern is not confined to warehouses, and it is not confined to the public. Institutions inherit the same wiring, because institutions are made of people, and a select committee can be convened about an announcement far more easily than about a rate of change.

Is "nobody noticed" itself true? Hold that claim lightly, because an absence is easy to overclaim and hard to prove. What we can say precisely: we searched for anyone reading these filings as a record of substitution, rather than as quarterly retail results, and found top-line coverage only. That is a search, not a proof, and if a reader knows better we would genuinely like the reference.

Here is what the mismatch means for the question this paper asks. If machines absorb work as a rate rather than as a day, then the arrival of the post-work economy may have no arrival. There may be no morning on which it happens, no footage, no padlock large enough. The question "when will AI take the jobs" would then have no answer, not because the change is not happening but because the question uses the wrong grammar: it asks for a when, and the phenomenon is a rate. A society tuned to whens can be told a rate twice a year, on schedule, in audited documents, and still be surprised.

The adaptation is not vigilance; nobody sustains vigilance, and no one should be asked to. It is calendars. The numbers arrive on a schedule, so the noticing can too, the way households already treat a Budget or an energy price cap change as a date to look up. The machines are not sneaking up on anyone. They are filing. The gap is not in the disclosure; it is in the reading, and reading can be scheduled.