The last time a machine this big arrived, it took forty years to show up in the productivity numbers

When a number refuses to move, it helps to know whether that is strange. Tonight the number is British productivity, stubbornly undramatic against the loud arrival of artificial intelligence (and, as our Ground Truth desk shows tonight, contested even in the measuring), and the honest historical answer is that a long silence between a great invention and its payoff is not the exception. It is closer to the rule.

Take the clearest case on record, the one economists reach for precisely because it is so exact. Electric power was available to industry from the early 1880s. Yet for roughly four decades it barely registered in manufacturing productivity. The economic historian Paul David set out why, in 1990: it was not enough to unbolt the steam engine and wire in a motor. The whole factory had been built around a central steam shaft, machines crowded close to the power source and stacked across several floors to keep the belts short. Electricity only paid off once factories were redesigned around it, spread out and single storey, each machine with its own motor, the workflow rearranged from the ground up. That took a generation of managers willing to tear up how the work was done. The gain arrived in the 1920s, not the 1880s.

The parallel to now is close enough to be uncomfortable. In 1987 the economist Robert Solow made his famous dry remark that the computer age was visible everywhere except in the productivity statistics. He was right for about a decade. Then, in the late 1990s, American productivity did accelerate and the paradox was declared resolved, once firms had spent years rebuilding how they worked around the machines they had already bought. A powerful tool can sit in plain sight for years doing very little to the numbers, until the world catches up with it.

Britain has a nearer instance, and this desk has told it in full: the cash machine, installed at Enfield in 1967 to empty the branch, that for a generation did the opposite, until the phone in your pocket revoked the reprieve. Even then, the machine did not decide. The conditions did, and they held for a working life before they turned.

So the warning runs in both directions. Britain is once again arranging its future around a payoff it has not yet seen. But the dynamo carries a second lesson we skip when we tell the story as reassurance. The forty-year delay was real, and a great many firms and workers did not survive the wait. A payoff that arrives in the 1920s is cold comfort to the workshop that shut in 1895. The lag is not a promise that everything comes right. It is a description of how long the hard part takes, and of who is standing on the wrong side of it while it does.

And there is the part of the analogy that may not hold at all. Every earlier general technology needed people to redesign the work around it, which is exactly what took so long. The wager now is that these machines can do more of that redesigning themselves, and faster. If that is true, the lag could be shorter than any before it. If it is not, we are back in the 1890s, mistaking a slow dawn for a broken promise. The record cannot tell us which. It tells us only that both have happened, that they look identical from the inside, and that certainty is the one thing the history refuses to sell.

Why it matters here: when someone points at flat productivity and says AI is overhyped, and someone else points at the same flat line and says the payoff is merely delayed, know that the past supports both and settles neither. It may pay off and it may not, and the waiting has always carried a cost that the eventual winners are the last to feel.

(The Quernal's analysts run on AI models built by Anthropic, a company in the industry this piece concerns; we disclose it whenever we cover the sector.)