The data are in: AI is not causing a jobs apocalypse

For two years, some of the loudest voices on AI have spoken on white-collar layoffs, vanishing entry-level jobs, and a generation of graduates shut out. Chief executives routinely point to AI when they announce cuts. Is AI the shiny new scapegoat on the block, or a legitimate culprit behind what would otherwise be faster job growth? We argue it is neither.

A new Bureau of Economic Analysis working paper from one of us provides early evidence linking actual AI utilization across states and industries with real output and employment, providing some of the first national evidence on the potentially important upside that AI has for not only aggregate growth, but also for the nation’s small businesses, which generate nearly half of U.S. economic output and are often run by solopreneurs.

For context, a study released in June reaches the same conclusion using an entirely different kind of data: money. Economists at Ramp and Revelio Labs linked actual AI spending to hiring records across more than 21,000 U.S. companies. Firms that adopted AI most intensively grew their employee headcount by about 10 percent over the following two years; firms that only dabbled saw no gain at all.

The result holds precisely where the alarm has been loudest. At the heaviest adopters, entry-level hiring grew 12 percent, faster than any other category, as those firms went looking for young workers who already knew how to use the tools. Similar patterns are seen in small business adoption of AI. A JP Morgan Chase study found newer firms were more likely to adopt AI sooner and ramp up use more quickly.

Adoption itself is no longer in question. Ramp’s spending index, based on transactions at more than 70,000 businesses, shows the share of companies paying for AI crossed 50 percent in early 2026, up from 35 percent a year earlier. Even the Census Bureau’s survey of firms , which the BEA’s new working paper finds captures a narrower slice of AI use than worker surveys do, shows a steep increase, particularly in 2025. Ask workers, ask employers, or watch the corporate card — every measure agrees that AI has spread through the economy with unusual speed.

Critics of the Ramp and Revelio study rightly caution that the results are not causal, nor necessarily nationally representative. That is where the BEA working paper comes in.

Using nationally representative survey data — Gallup’s Workforce Panel, which asks more than 20,000 U.S. workers each quarter — we find that the states and industries where workers use AI most intensively show stronger output and productivity after 2020, and employment that rose rather than fell.

Whatever AI is doing in those industries, it is not showing up as displacement, and it is not all “slop” — that is to say, productivity has actually grown, at least on average.

Our statistical model compares each state and industry with itself over time and strips out shocks that hit entire industries in a given year — a pandemic-era technology boom, for instance — so the comparison rests on differences across states within the same industry.

To be sure, our results are not fully causal either. The states and industries using AI most intensively in 2025 and 2026 might also be the ones that grew fastest earlier on. Our event study lets us test for this directly, by checking whether these places were already pulling ahead before 2020. In terms of output, they were not.

As a further check, we instrument AI use with each area’s pre-pandemic occupational mix. The results also line up with earlier work , which measured how states and industries more exposed to generative AI before the pandemic fared on output, wages, and employment — with highly similar results.

The upside looks largest for the businesses least likely to seize it. Ramp finds that small firms adopt AI less often than large ones, but the small firms that do adopt tend to use it more intensively. And because AI can stand in for functions a small company could never staff on its own or never planned to staff — for example, writing software, keeping the books or answering customers — thus the gains from AI can be outsized. The BEA study shows a matching pattern: the post-2020 employment gains appear soonest and most clearly among the smallest employers, close to 10 percent within a few years, while at the largest firms no clear gain shows up until 2022.

If there is a problem in these numbers, it is distribution, not destruction. The firms capturing AI’s gains so far are the ones that already had engineers, capital, and a habit of adopting new tools. Ramp finds that who funded a company predicts its AI use better than the industry it operates in, and that California firms adopt faster than otherwise-similar firms in New York — AI travels through networks, and those networks leave many small and midsize firms outside.

Realizing AI’s full productivity and growth potential will require coordinated efforts to address this distribution problem. Indeed, the Milken Institute recently called for a national council on AI for small businesses to provide leadership and guidance on how and where small business owners can apply AI most effectively to networks and small business support organizations nationwide to facilitate a wider distribution and utilization of AI.

However, the most useful responses to solving the distribution problem are often the unglamourous ones. The Small Business Development Centers, community colleges, state workforce boards, and other technical assistance providers are already positioned to get AI tools, training, and practical know-how to the businesses and regions on the wrong side of that gap.

The evidence is early on AI. It will shift as more data arrive. But for now, the numbers do not support the story of an AI jobs apocalypse. The companies using AI most are growing and hiring — including the young workers everyone was warned about.

The objective before us all is to widen that circle and direct resources toward businesses where AI can be a productivity force multiplier, instead of bracing for a collapse that, so far, the data does not show.

Christos A. Makridis is an associate research professor at Arizona State University, digital fellow at the Stanford Digital Economy Lab, and associate faculty at the Complexity Science Hub in Vienna. Kristen Fanarakis leads a small business, entrepreneurship, and economic growth research initiative at the Milken Institute.