The Economist is right, and I want to say that before the disagreement.

Payrolls added 162,000 jobs in August against a consensus of 53,000. Unemployment held at 4.1%, lower than in nearly 90% of months over the past half century. The gap between 20-to-24-year-old unemployment and the overall rate sits near a multi-decade low. The magazine published on September 4 that the apocalypse is postponed and an AI jobs boom is under way, with something like a million AI-created roles.

Their arithmetic goes further than the headline and deserves quoting in full. Against that million they set roughly 200,000 layoffs attributed to AI since mid-2023, and they find engineers, software developers, mathematicians and data scientists added some 730,000 positions above trend since 2022.

The forecast of mass unemployment lost. Fairly, on the data.

Here is where it stops helping you. A payroll survey counts employed people. It does not count tasks. Everything AI is currently doing inside companies happens within jobs that are still held by the same person, and that motion has nowhere to land in this statistic.

Thirty-six years of measuring the wrong thing first

I started in 1990, fifteen years old, on a Commodore 64 and a TI.

Every wave since then got argued as an employment question before anyone treated it as a process question. Spreadsheets did not fire accountants. They changed what an accountant did between nine and six, and the profession grew.

Run the employment data of those years and it looks exactly like this month’s. Healthy, stable, no signal.

What did change, visible inside firms long before it was visible in any national series, was who checked what, how long a close took, and how many people it took to produce the same report. That is where the story lived both times.

The instrument is fine, it answers a different question

The BLS does its job well. 162,000 jobs, 7.0 million unemployed, average hourly earnings at $37.75 after a ten-cent rise, up 3.1% on the year. All accurate and all useful.

Your question is whether the quoting process that takes four days today still takes four days next quarter, and whether the two people running it will be running it or doing something more valuable. National statistics have no resolution for that and never will.

Occupation-level work gets closer. Karpathy scored all 342 occupations one by one, which we went through in the AI exposure map, and it remains the most useful public instrument for this argument. On the demand side, AI-exposed postings started growing again in the Indeed Hiring Lab data.

Both things hold at once. Healthy aggregate employment, and deep reorganization of what sits inside each role. That combination is exactly what you would predict if the technology works.

One more figure from the same release deserves a mention, because it cuts against a lazy reading of the good news. The 162,000 came against an average monthly gain of 31,000 across the previous twelve months. August was an outlier month, not the establishment of a trend, and a single strong print is thin evidence for a structural claim in either direction. Anyone using it to argue that AI creates jobs is making the same mistake as the people who used three weak months to argue the opposite.

What you can actually measure

Three numbers, none of them federal.

How long a specific repetitive task takes now against a year ago. Who reviews the output before it leaves the building, and what share of their week that consumes. And where a saved hour lands in a financial statement.

That third one is usually missing, and it is the one that turns an improvement into a result. Saving four hours a week on a team that will be fully staffed next quarter is not a saving. It is free capacity, and it only counts if somebody decides what to spend it on. We worked through that shift in where the work moved.

The Economist won the headline argument with real data. The headline was never the work.

Two months of AI Maestro discovery exists to measure precisely what a payroll survey cannot, process by process, with a map of where each decision actually lives. If your board forwarded you the September 4 piece as a reason to wait, that is the conversation to have this week.

Measure what the payroll survey cannot see