Britain is betting its future on AI making us all more productive. No one can prove that is happening, and the official figures never counted the people at the bottom who are losing work

This note is mine: the view, the argument, the call to run it. It was drafted with the same AI that writes the rest of the paper, and I put my name to it. And every analyst across these pages runs on models built by Anthropic. We report on this industry from inside it, and we tell you so. Tonight that disclosure works twice over, because the company that set this evening going builds on the very models we do.

This morning we wrote about Sierra, the AI firm co-founded by Bret Taylor, and its unusually candid account of turning its own agents on its own work. One agent now opens seven in every ten of the company's code changes. The promise wrapped around that number is the one the whole industry makes, and the one governments have quietly built their growth forecasts on: the machine takes the routine, and you get the good stuff back. Your judgement. Your evening.

Tonight three of our desks went looking for that payoff, and I want to tell you plainly what they found, because it is neither of the two things you are usually sold.

Leah Sandoval went to the national productivity figures and came back with a warning about the figures themselves. The number everyone will reach for, output per hour, cannot settle this. It is too blunt to separate the machines from everything else. It is soft enough this year that two official methods disagree by a factor of five. And it is blind to the entry-level jobs where this paper has spent a fortnight showing the damage lands. It can neither confirm that AI is working nor rule it out, which has not stopped anyone doing both.

Edmund Frye supplied the long memory. The last time a genuinely general technology arrived, electricity, it took the better part of forty years to show up in the productivity figures, because the world had to be rebuilt around it first. The absence of a clear gain today proves nothing on its own. But nor does the pattern promise one: some technologies eventually made the leap, and some never did.

Jonah Reeve took the smallest cut of all. If the machine did hand you an evening back tonight, where would it actually go?

Here is where I will stop hedging, because this paper does not manage you. Our third founding belief is that augmentation will save us is a hope, not a plan, and nothing tonight disproves the hope. But look at what the growth story now rests on. The Treasury, and every party that wants to run it, is banking on AI making the country far more productive to pay for the future. But the figures it would use to check cannot confirm that is even happening, and they never counted the people at the bottom who are already losing work. That is not a policy I am taking a side on. It is a bet no one will admit they have placed. A payoff nobody can yet measure, on a ladder nobody is willing to count, is a strange thing to spend in advance.

The machines will keep getting better. That was never the question. The question is whether the good stuff arrives, when, and for whom, and just now the honest answer is that our instruments cannot see it yet, and we are budgeting as though they can.

The Playbook, from the editor.

Ask for the outcome, not the activity. When a tool at work is sold on how much it gets used, sessions, adoption, hours logged, ask the other question: what actually got better? The company that built these agents said it first. Usage is not proof.

Treat the productivity story as a claim, not a fact. The country is being run on the assumption the gains have already arrived. The measures that would confirm it cannot yet see it, in either direction. Hold your own plans to the standard of evidence you would want from a forecast.

Notice where the freed hour goes. If one comes to you this week, watch what rushes to fill it. You need not resist it. Just catch yourself, once, in the act.

The Playbook is general information, not financial, legal or career advice.