Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen published a revised working paper on August 12, 2026, using ADP payroll records through June 2026.[1] The Stanford Digital Economy Lab update still finds no evidence of widespread, economy-wide job displacement. Employment of 22-to-25-year-olds in highly AI-exposed occupations now stands about 19 percent below where it would be had it kept pace with less-exposed peers. Experienced workers in the same comparison show no comparable gap.
The authors now lead with that simpler kept-pace shortfall. Earlier vintages had headlined regression estimates. By this descriptive measure the shortfall was 15 percent at the July 2025 data vintage and 19 percent as of June 2026. In levels, employment of 22-to-25-year-olds in the two most exposed quintiles fell about 11 percent between November 2022 and June 2026, while the same age group in the three least-exposed quintiles grew about 10 percent. The widening is the story they are asking readers to watch.
The flow behind that shortfall is intake. The gap runs mainly through reduced hiring of young workers. Separations are not the driver. Declines concentrate in occupations where AI usage substitutes for human tasks. Where usage complements workers, employment is flat or rising, especially for experienced workers.[1]
Several other stories do not close the gap. It persists when the authors exclude technology firms and computer occupations, when they control for interest-rate exposure and remote work, and across alternative AI-exposure measures. By November 2022 the relative position of exposed occupations had returned to about its pre-pandemic level, so the later decline moves the gap below that baseline.
The paper still withholds a causal claim. The authors call the six facts early descriptive indicators, canaries, not estimates of what generative AI caused. They flag divergent trends that predate ChatGPT, attenuation when education is controlled, and a stronger pattern in the ADP sample than in national surveys. They cite Tucker (2026) only as showing consistent raw patterns in government administrative data by age and industry exposure.[1]
Yale's Budget Lab, updating through July 2026 CPS microdata, still does not see a clear AI footprint in occupational mix or in synthetic comparisons of exposed and unexposed jobs. Its younger-worker cut is a broader 16-to-34 band and returns mixed, statistically insignificant results.[2] That is a different unit than Stanford's 22-to-25 slice inside ADP.
Chris Churchman, who leads Goldman Sachs Marquee and co-chairs its Global Banking and Markets AI working group, has warned that automating junior work can hollow the path that produces senior judgment.[3] That is a practitioner account of skill formation. It is not payroll evidence. It is the question the 19 percent figure leaves on the table: if the intake door narrows, where does the next cohort learn the work that still has to be judged, fixed, and owned.
Stanford publishes a public set of AI Economic Indicators for ongoing tracking.[1] The next vintage is the check, not a slogan about jobs ending.
Verification bottleneck
Payrolls can look calm while the apprenticeship layer is the thing that moved.
- The 19 percent figure is a kept-pace comparison in ADP, a large private payroll sample, not a national census of jobs destroyed.
- The authors do not claim causation. Pre-trends, education controls, and survey-versus-ADP differences remain on the table.
- Yale's broader occupational tracker can stay quiet while Stanford's narrow age-and-exposure cut widens. Those two results can both be true.
- Hiring, not firing, is the flow that carries the canary. A firm can report stable headcount and still stop training the next bench.
Opportunities
Employers that still run junior programs in writing, analysis, coding support, or research can count their own 22-to-25 intake against last year. A missing hire does not show up as a layoff. Idea fodder only. Not advice.
Schools and other training programs have a related paperwork job if on-the-job grind is no longer where the receipt of practice lives. The public dashboard is the watchlist: whether the kept-pace shortfall widens, stalls, or reverses in later ADP vintages.
Sources
[2] The Budget Lab at Yale, Tracking the Impact of AI on the Labor Market, updated July 16, 2026
[3] Goldman Sachs Exchanges, Building AI systems for capital markets, Chris Churchman