may 2026
i talk to a lot of founders, investors and operators. genuinely brilliant people, most of them completely convinced this is the biggest technological shift in human history.
and then you get into the actual rooms. integration is hard. the data is a mess. the gains are real but modest. the pilot works and the rollout stalls. everyone says revolution in public and says it is going okay in private, and both are true at the same time.
i used to read that gap as early days. im less sure now.
electricity arrives commercially in the 1880s. everybody understands immediately that it changes everything. they are completely correct. and then productivity in american manufacturing does approximately nothing for about forty years.
the usual telling of this is “be patient, transformation takes time.” thats not what happened, and the real version is much more useful.
factories at the time were built around a steam engine driving a single line shaft down the length of the building, with belts running off it to each machine. when electricity arrived, factories bought electric motors and used them to drive the same line shaft. one big motor where the engine used to be. entirely sensible, and it produced almost nothing.
the gains came decades later, when someone worked out you could put a small motor on every machine. the shaft became unnecessary, the layout could follow the work instead of the power, and the whole floor could be rearranged around how the thing was actually made. thats when the numbers moved. not when the technology arrived, when the building was redesigned around it.
computers ran the same play. solow made his famous crack about seeing the computer age everywhere except in the productivity statistics in 1987, and the acceleration didnt really show up until the mid nineties. same shape. same reason.
the money going into this right now is enormous. and when i look at what it is actually buying, most of it is a faster version of a process that already existed.
summarise the thing someone was already reading. draft the email someone was already writing. cut the support team because the bot handles tier one. automate the report nobody read.
all of that works. all of it shows up as a line item. and all of it is one big motor bolted to the old line shaft.
nobody is doing the equivalent of throwing out the shaft, because throwing out the shaft means admitting the process itself was shaped by a constraint that no longer exists, and then redesigning around its absence. that is slow, it is politically expensive, and it does not fit in a quarter.
organisations are not built for transformation. they are built for stability, and they are extremely good at it. they will absorb almost anything you hand them and convert it into a slightly cheaper version of what they were already doing. thats not a failure of the technology. its the immune system working exactly as designed.
we had the compute. we had more capital than any technology has ever been handed. we had, genuinely, a very large share of the smartest people of a generation all pointed at the same problem.
and the most defensible thing most of us can point to is that some spreadsheets are faster and some support teams are smaller.
thats the future that unsettles me. not the one that kills us, which at least would mean we did something. the one where all of it works, and we live a slightly better version of the same life, and in thirty years the honest summary is that we made existing things marginally cheaper.
not that we tried and failed. that we never really tried.
maybe im just impatient. nine years in, on a curve that historically takes twenty five to forty, and complaining that the numbers havent moved is a bit like standing in a 1905 factory declaring electricity overrated.
thats fair, and its probably partly true. but the lag in those cases was not passive waiting. it was the slow, unglamorous work of rebuilding around the new thing, and someone had to start that work early for it to finish at all. the forty years wasnt the world being patient. it was the world being slow, and a small number of people not being.
so the question isnt whether the gains arrive eventually. its whether youre one of the people doing the rebuild or one of the people bolting a motor to a shaft and reporting a saving.
the same thing holds one level down, when youre building with these models rather than buying them. the interesting decisions are architectural rather than technical, and the model is the part everyone has. this is that claim at the scale of an organisation.
the model is not the advantage. everyone can buy the model. the advantage is being willing to redesign the process around what the model makes possible, which almost nobody wants to do, because the existing process is where all the politics, the headcount and the last five years of somebodys career live.
and the tell is very simple. if the change you shipped this quarter would still make sense in a world where the model didnt exist, you rearranged furniture.
comfortable and consequential are not the same thing. i keep having to remind myself of that, because comfortable is genuinely nice and consequential is mostly just stressful, and there is no external force that will ever push you from one to the other.
if this really is the most important technological moment in a long time, and i think it might be, then the only honest question is what youre doing today that reflects believing it.
im not sure my own answer is good enough yet.