For three years the prophecy has been strikingly consistent. Sam Altman suggested AI could eliminate up to half of all existing jobs. Dario Amodei warned it could wipe out half of entry-level positions and push unemployment to between 10 and 20 percent by 2030. This Labor Day, the American economy answered with a shrug. August brought 162,000 new jobs, comfortably above forecasts, and unemployment held at 4.1 percent, lower than in almost 90 percent of months over the past half century. The apocalypse, as The Economist and others put it this week, has been postponed.
The New Yorker, the economist Noah Smith and the aggregator AllSides all circled the same puzzle from different angles: a technology everyone calls a job-killer that stubbornly refuses to kill jobs. Layoffs since mid-2023 have averaged around 1.7 million a month, essentially unchanged from the 2010s. McKinsey found that while 40 percent of large firms now deploy AI agents, only 14 percent reported actual AI-related job cuts this year, down from the 32 percent who a year earlier said they expected them. Yale's Budget Lab put it flatly: the occupational mix is not yet changing in ways that clearly track the arrival of AI.
The most interesting explanations converge on a single insight, that most jobs are messier than they look. In a new book titled "Messy Jobs," economists argue that roles resist automation because they bundle far more tasks than an outsider sees. Radiologists, long cast as AI's first casualties, also talk to patients, coordinate with surgeons and adjudicate hard cases. Smith offers a sharper version: people often do not know how they produce value at work, so when AI absorbs one task, it tends to make the worker more valuable as a piece of the larger machine, not less. Even translators, a seemingly obvious target, are employed in roughly the same numbers as in 2022.
There is a second, less-noticed story: AI may be creating more work than it removes. The Economist estimates the technology has generated about a million American jobs, against roughly 200,000 layoffs attributed to it since 2023. Some are new white-collar categories, AI engineers, data annotators, freshly minted "heads of AI," but many are stubbornly physical: the electricians, HVAC installers and data-center technicians building the boom's infrastructure, often at wages 40 percent above comparable work. Census Bureau data shows that among firms reporting any employment change from AI, more report hiring than cutting.
None of this means the danger was imagined, only that it is arriving slowly. The New Yorker flags the real pressure points: recent graduates, whose jobless rate has climbed above the national average for the first time in decades, and junior roles in law and finance where clients such as Citigroup are already demanding AI-driven fee cuts. The labor share of national income has quietly slid to a historic low of 52.8 percent. The lesson of this Labor Day is not that AI is harmless, but that inertia has bought some time. As Altman himself now concedes, "the economy just has so much inertia." The open question is what gets built with the reprieve.