ILLUSTRATION: ALISA STERN; SHUTTERSTOCK/SEVENTYFOUR, EARLY SPRING
FOR THREE YEARS the prophets of technological unemployment have had a field day. ChatGPT’s debut in late 2022 unleashed a wave of forecasts predicting that generative artificial intelligence would hollow out swathes of the workforce with a speed unseen since the spinning jenny.
By last summer more than half of working-age Americans had used the technology, according to the Real-Time Population Survey. That such a transformative force would soon show up in hiring data seemed a matter of when, not if. A new paper by Jessica Liu and Douglas Webber of the Federal Reserve Board suggests the answer, for now, is never.
The pair have done something straightforward yet oddly rare: they have looked for the smoking gun of an AI-induced hiring slump and found none. Using two rich data sources—the Census Bureau’s Business Trends and Outlook Survey (BTOS) and job-postings from Lightcast—they track the relationship between AI adoption and subsequent hiring behaviour. If the technology were destroying jobs en masse, you would expect firms or industries that have embraced it to be posting fewer vacancies.
Yet across a battery of specifications, the coefficients are either positive or indistinguishable from zero. Industries where BTOS respondents reported using AI in the past two weeks showed no sign of pulling back on hiring in the following year. Firms that had posted a job requiring AI or machine-learning skills—a more intensive form of adoption—went on to post more vacancies, not fewer, in the subsequent months.
The magnitudes are tiny. For the average firm, where only 1.6% of postings mention AI, the effect on total hiring is a rounding error. But that is the point. After a technology shock that had economists reaching for their Schumpeter, the labour market is reacting with the enthusiasm of a teenager asked to tidy their room. If there is carnage in specific occupations—and the authors do not rule out pockets of pain—it has been more than offset by hiring elsewhere. The net effect, so far, is a shrug.
This absence of evidence is not yet evidence of absence. The United States remains in the early innings of AI diffusion. Even as usage has soared, only around 10% of firms in the BTOS survey reported actually using the technology in production as of late 2025. That figure, while up from zero a few years ago, is still a sliver of the economy. It is plausible that the real disruption lies ahead, as adoption spreads and employers figure out which roles are truly redundant.
Yet the paper points to another possibility: that AI is proving to be less a destroyer of jobs than a reshuffler of them. The authors note that past research has found that AI-exposed occupations tend to see wage gains, rather than job losses, as productivity rises. Other work suggests that while entry-level roles in automatable fields may shrink, opportunities for more experienced workers hold steady or grow. If AI is acting as a skill supplement for the adept rather than a replacement for the novice, the headline hiring numbers would stay stable even as the composition of work changes beneath them.
That would be cold comfort for the workers caught in the churn. But for policymakers watching for signs of a structural break in the labour market, the Liu-Webber findings confer a moment of calm. The slowdown in national job postings since the post-pandemic frenzy has had many culprits—from higher interest rates to fading fiscal stimulus and corporate caution—but AI does not appear to be among them. The revolution, for now, is a quiet one.
The authors are careful to note that their exercise is backward-looking and that they plan to keep watching. So should everyone else. The history of technology is littered with productivity miracles that took years to materialise, and with them, labour-market dislocations that arrived not with a bang but with a persistent shift. The absence of an AI jobs panic in the data through 2025 is reassuring. But it is not a guarantee.
The machines may yet make their presence felt in the hiring numbers. For the moment, however, they seem content to let firms get on with the business of employing people the old-fashioned way. ■
