Does AI Adoption Lead to Layoffs?

Does AI adoption lead to layoffs? The most direct evidence available says no. A study linking real AI vendor spending at 21,559 US firms to those same firms' actual monthly headcount found that companies making the largest AI investments grew employment by 10.2% over the following two years, while companies that barely used AI showed no significant change at all. The firms spending the most on AI are not the ones cutting staff. They are the ones hiring.

That finding matters because it contradicts the story in most headlines. Executives keep naming AI as the reason for a round of cuts, and the explanation spreads faster than anyone checks it. This is the first study built to check it directly, using what companies actually spent rather than what they said in a press release.

What this study actually measured

Most earlier research on AI and jobs had to guess at adoption. A researcher would score an occupation for how exposed it looks to AI, or ask executives in a survey whether their company uses it, then infer employment effects from there. This study instead linked Ramp's corporate card and bill-pay records, which show exactly when and how much a company pays AI vendors, to Revelio Labs' workforce records for the same firms, covering 21,559 companies observed monthly from January 2021 through February 2026. A firm counts as an adopter only after three straight months of sustained AI spending, and firms are sorted into low- and high-intensity adopters by how much they spend per employee in the first months after adopting.

The design also corrects for a real problem: firms that adopt AI heavily were already growing faster than everyone else before they adopted it. Comparing them to firms that never adopt would overstate AI's effect. The researchers instead compared adopters to other, similar firms that had not yet adopted at the same point in their own growth curve, which is the more conservative comparison and the one the headline numbers below come from.

So does AI adoption lead to layoffs? Heavy adopters hired, they did not cut

High-intensity AI adopters grew total headcount by 10.2% in the two years after adoption. Low-intensity adopters, the ones running light chat subscriptions rather than sustained investment, showed a change of -0.6%, not statistically different from zero. The gains were not confined to one department. Entry-level headcount rose 12.0% at high-intensity adopters, and headcount grew across sales, administration, engineering, customer service, and finance, all functions often named as exposed to AI.

Why the gains are not everywhere yet

The employment gains are not evenly spread. Statistically significant gains showed up only in the Information sector, where high-intensity adopters grew headcount by 13.4%; other sectors had positive point estimates that were not yet distinguishable from no effect. The gains also took time to appear: the effect was close to zero in the first three months after adoption and only became clearly visible six to twelve months in, after firms had time to establish workflows rather than run a pilot and stop. A subscription bought and left unused is not the same investment as one a company builds its work around, and the data draws that line clearly.

Why this does not match the headlines

A separate line of evidence backs this up. In the U.S. Census Bureau's own nationally representative business survey, only 5% of firms report any employment impact from AI at all, and of that small share, roughly as many report gains as report losses. Most companies naming AI as a reason for layoffs are describing a choice of framing, not a measured cause, a pattern this site has already traced through official job-cut reports. The timing of a broader hiring slowdown tells a similar story: a Federal Reserve Bank of New York study found the gap between AI-exposed and AI-safe occupations opened up before ChatGPT even existed.

None of this means AI has zero effect on any individual role, and it does not sit well against every other finding in this space. Stanford researchers have found a real, growing employment gap for 22 to 25 year olds specifically in the most AI-exposed occupations, measured across the whole economy, which is a different unit of analysis: an age-cohort effect across every firm in an occupation, most of which never adopted AI at all, next to a firm-level effect concentrated in the minority of companies that invested heavily. Both can be true at once. The honest read of both studies together is that the deciding factor is not whether AI exists in an industry, it is what a specific employer actually did with it, and what you can prove about the judgement you bring regardless of which one you work for. The check scores six of those judgement domains directly from your own answers, in about seven minutes.