The awkward fact about the AI boom is showing up in the aggregate statistics: the tools are spectacular and the productivity numbers are merely fine. Economists call it a diffusion problem. Operators know it by a homelier name — the distance between buying software and changing how work happens.
The firms pulling ahead share habits that predate the technology. They measure outcomes rather than activity, so automation shows up as capacity instead of threat. They redesign the process before wiring the tool into it, having learned that automating a bad process gives you bad results at scale. And they push decisions to where the information lives, because a model that answers in seconds is wasted inside an approval chain that answers in weeks.
The unfashionable conclusion
None of this is glamorous, which is why it is rare. The winners of the last technological platform shift were not the companies with the earliest licences but the ones that reorganised around the new cost structure. The pattern is repeating. The constraint is not the model's intelligence. It is the organisation's willingness to change what it does with an answer once it has one — and that, unlike the software, cannot be bought by the seat.
