No one agrees on the effect of AI on employment, and the data comes too late

Layoffs at Microsoft, 23,000 jobs destroyed in July in the United States, but a study of 21,000 companies showing that big investors in AI are hiring: the debate is stumbling over the measure itself.

The disagreement over the effect of AI on jobs isn't just about the conclusions: it's about the very ability to measure. Recent research credits technology with a positive effect on job creation, while economists and labor activists warn that it threatens to transform the financial system very quickly and that action is needed now. Big tech companies continue to cut their workforce during the boom: Microsoft laid off nearly 5,000 people in early July while pouring billions into its data centers after other cuts, and both Amazon and Oracle have cut thousands of jobs in the past two years.

It remains to be seen whether AI is the direct cause. "There has been some uncertainty about the extent to which the layoffs we are seeing are actually caused by AI. It's notoriously difficult to establish," says Till von Wachter, an economist at the University of California, Los Angeles. The confusion also comes from the executives themselves, who blame their layoffs on AI one month and then celebrate it as a driver of job creation the next — while Alphabet tells its investors that it plans to increase its workforce. Ben Zipperer, of the Economic Policy Institute, believes the impact so far is more limited than the worst-case scenarios announced. The latest U.S. jobs report, with 23,000 jobs lost in July against all odds, did not clarify the picture.

Some data points sharply in the other direction, with a study by financial services firm Ramp of more than 21,000 U.S. companies showing that those that have invested in AI have seen their headcount grow. Its biggest spenders, classified as "high-intensity" adopters, have increased their staff by 10% in two years and their entry-level hires by 12%, contrary to the narratives of a stricken market for recent graduates — while the bottom two-thirds of adopters have seen no growth. An important nuance is that these most intense adopters tend to be small organizations, perhaps already growing and more inclined to experiment, and they use advanced tools such as code agents or APIs.