For two years, the loudest claim in the AI-jobs debate has been a prediction: white-collar layoffs, vanishing graduate roles, a generation locked out. What has been rarer is federal statistical agencies measuring what actually happens in the places where AI use is most intense. A Bureau of Economic Analysis working paper, published in August and written up by two of its co-authors in The Hill on 8 September, does exactly that — and its early numbers cut against the apocalypse story.
🔍 THE BOTTOM LINE — The BEA paper combines worker-reported AI use from Gallup’s Workforce Panel with federal economic data across US states and industries. Cells with the most frequent AI use show stronger post-2020 output paths and positive — though imprecisely estimated — employment differences. The authors themselves call the evidence descriptive, not causal. Both halves of that sentence matter.
What the Paper Actually Found
The study, WP2026-18 by Christos Makridis, Tina Highfill and Jon Samuels, pools Gallup’s worker surveys from Q2 2025 through Q1 2026 — a nationally representative panel asking people how often they use AI at work — and matches it to state-by-industry employment, earnings and real output data from the LEHD program and BEA national accounts.
Three findings stand out:
- Output first, jobs alongside. State-industry cells with higher worker-reported AI use show consistently stronger real output growth after 2020. Employment effects are positive but less statistically certain — the honest reading is “no displacement signal, weak growth signal.”
- The dose matters. Per the authors’ earlier presentation of the results, a one percentage point increase in frequent AI use is associated with roughly 0.2–0.4% higher employment and 0.1–0.2% higher real output after ChatGPT’s diffusion.
- Small firms see it soonest. The strongest employment patterns appear in the smallest establishments (0–19 employees), consistent with AI acting as a force-multiplier for businesses that could never have staffed certain functions at all.
Gallup’s panel shows any AI use climbing from about 20% of workers in mid-2023 to nearly 50% by early 2026, with frequent use rising from roughly 10% to more than 25%. Whatever AI is doing to work, it is doing it fast.
The Caveats Are in the Paper, Not Just the Comments
This is where the write-up and the underlying study deserve equal attention. The BEA authors state directly that their estimates are descriptive rather than causal: states and industries that adopted AI most intensively by 2025–26 might simply be the ones that were already growing. Their event-study design checks for exactly that — the pre-2020 coefficients for high-adoption cells were “slightly negative or near zero” — but they do not claim the question is settled, and they write that results “should be read accordingly.”
The Hill op-ed, by Makridis and Kristen Fanarakis, runs further than the paper does, arguing “the numbers do not support the story of an AI jobs apocalypse.” That is opinion layered on early evidence — the reader is entitled to know the difference.
The Corroborating Evidence
The op-ed cites a second, independent data source that reaches a similar place by a different route: a June study by economists at Ramp and Revelio Labs linking actual AI spending to hiring records across more than 21,000 US firms. Firms that adopted AI most intensively grew headcount about 10% over the following two years, with entry-level hiring up 12% — faster than any other category, as those firms sought young workers already fluent in the tools. As covered on this site earlier this month, that study has its own survivorship questions, but two unrelated datasets pointing the same direction is how a picture builds.
The op-ed also names the real problem the data does show: distribution, not destruction. Ramp’s transaction data finds small firms adopt AI less often than large ones, and the gains concentrate among businesses that already had engineers, capital and a habit of adopting tools. The Milken Institute has since called for a national council on AI for small businesses; the authors argue the unglamorous fix is funding the community colleges and workforce boards that already reach the firms outside the network.
A NZ Reading — With the Usual Honesty
New Zealand has no equivalent to Gallup’s quarterly panel or the BEA’s state-industry linking, so no NZ agency can currently answer the question this paper asks: where is AI actually being used, and what happened to employment there? What NZ does have is a small-business economy — roughly half of all US output comes from small business, and NZ’s economy skews smaller still. If the BEA’s small-establishment pattern holds, the countries that help their smallest firms actually use these tools may capture employment gains the doomsayers never projected.
The caution travels with it: this is one working paper, descriptive by design, with the authors’ own caveat attached. The numbers do not prove AI is a jobs boom. What they do is shift the burden of proof. “AI is destroying jobs” is now the claim that needs evidence, not the default assumption.
FAQ
What did the BEA working paper find about AI and jobs? US state-industry cells with the most frequent worker-reported AI use showed stronger real output growth after 2020 and positive, though imprecisely estimated, employment differences. The authors call the estimates descriptive, not causal.
Is AI causing a jobs apocalypse, according to this research? The paper finds no displacement signal in the places using AI most intensively — but it explicitly does not claim proof. The Hill op-ed by the co-author goes further than the paper itself.
Does AI use help small businesses? The strongest employment associations in the study appeared in the smallest establishments (0–19 employees), consistent with AI substituting for functions a small firm could never afford to staff.
Has NZ studied AI’s effect on employment this way? No equivalent NZ study links worker-level AI use to official employment data. Stats NZ does not yet publish an AI-utilisation panel comparable to Gallup’s.
Read next: Companies Spending the Most on AI Are Hiring the Most People, The Economist Counts a Million New AI-Era Jobs in America, and The NY Fed Says AI Is Reshaping Hiring, Not Killing Jobs.
CJ Murden is editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
