Last week this newsletter told you July payrolls fell by 23,000. On Friday the Bureau of Labor Statistics revised that to a gain of 21,000, then reported August at 162,000, more than five times the average monthly gain over the prior year. Unemployment held at 4.1 percent. Participation edged up to 61.6 percent, still half a point below January. The gains were concentrated in restaurants, local school districts, and manufacturing, which is up 58,000 since its December low. The one sector that lost ground was information, down 23,000, with the losses in computing infrastructure, data processing, publishing, and broadcasting. That is the AI-exposed corner of the economy, and it is the only corner shrinking.
Now read the Challenger report next to it. Employers announced 52,881 cuts in August, the lowest August since 2022. Restructuring led, followed by market conditions and closings. Artificial intelligence, which had been the top stated reason every month since March, came fourth at 3,462 cuts, the fewest since December. It is still the leading reason for the year at roughly 22 percent of announced cuts, so nobody should read one month as a verdict. But the story that every layoff is an AI layoff has lost its lead, and the sector where AI is actually displacing work is a narrow one that employs a small share of the people reading this.
The Federal Reserve’s Beige Book, out the same week, put it in plainer language. Districts reported both positive and negative effects of AI on labor demand, and the thing employers could not find was skilled trades and technical workers. The economy is hiring. The open question is whether the software in the middle is letting the right people through.
Technology companies cut about 14,700 positions in August. Technology employment rose by 86,000. Both numbers come from CompTIA’s read of Friday’s BLS data, and the gap between them is the hiring signal of the week. The jobs left the tech sector and showed up in manufacturing, administrative services, retail, and finance, at companies that are past the AI pilot and now need someone to run the thing. There were nearly 600,000 open technology postings in August, and 58 percent of them were carried over from earlier months, unfilled.
Challenger’s hiring-plan data says the same thing from the other side. Employers have announced plans to hire 119,825 workers this year, up 37 percent, the strongest start since 2023, with 46 percent of those plans in manufacturing. Andy Challenger’s own caveat was that the positions do not appear to be filling quickly. The Kansas City Fed called labor the top growth constraint in its district and said hiring difficulties had broadened.
I have run enough sourcing programs to know what a requisition that stays open all summer looks like from the inside. The budget is real. The manager is serious. The screen was built for the candidate they imagined, and it keeps rejecting the candidate who has already done the job. So who fills a role that has been open since spring? The person who has filled it before, once the filter gets out of the way. That is the question in front of Judge Rita Lin on Monday, when she hears argument on class certification in Mobley v. Workday, the case asking whether an AI screening tool can be held liable under the Age Discrimination in Employment Act for the applicants over 40 it rejected. The hearing is set for September 14 in the Northern District of California.
A group of management researchers in the U.S., U.K., and Switzerland published a paper last month warning that managers who lean too hard on generative AI suffer something they call epistemic de-skilling. Stripped of the Latin, it means this: judgment comes from doing the work and living with the consequences, and if you outsource the doing, the judgment does not form.
Nobody who started work before the browser needed a study to learn that. We built judgment the only way it was available, by getting things wrong in front of people who remembered. The pattern-matching came later, the hard way, and it sits in a place no prompt can reach.
So the market is now paying to find something that thirty years of work manufactures as a byproduct. The researchers’ prescription is to use AI to test your reasoning instead of replacing it. Fine. We have been doing that with a colleague down the hall since 1994. Experience is the one asset that appreciates because it cannot be shortcut.