Stanford's Jobs Data Says the Gap for Young Workers Keeps Widening
The August update to the Canaries in the Coal Mine paper puts employment for 22 to 25 year olds in AI-exposed jobs 19 percent below their peers, up from 13 percent. Older workers show no such gap.
The Stanford Digital Economy Lab published an updated version of its “Canaries in the Coal Mine” paper in August, and the central number moved in the direction nobody was hoping for. Employment for workers aged 22 to 25 in the occupations most exposed to AI now sits about 19 percent below where it would be had it kept pace with less-exposed peers. In the previous release the figure was 13 percent.
The study, led by Erik Brynjolfsson with Bharat Chandar and Ruyu Chen, uses high-frequency payroll records from ADP covering millions of US workers through June 2026, which is why it can see changes long before official statistics do. Three findings are worth separating out. First, the effect is concentrated almost entirely among the youngest workers; more experienced people in the same occupations show no comparable gap. Second, the mechanism is reduced hiring rather than layoffs, so the jobs are not disappearing from under people who hold them, they are not being opened in the first place. Third, and most usefully, the decline shows up in occupations where surveys say AI mostly substitutes for human tasks, while in occupations where AI mostly complements the work, employment is flat or rising, especially for experienced staff. The authors describe the entry-level effects as real, persistent and widening, and they have tracked the divergence steadily since first documenting it in August 2025.
What’s actually going on here: the substitute-versus-complement split is the part to hold onto, because it is doing most of the explanatory work. Entry-level jobs have historically been paid apprenticeships: a junior does the routine, well-defined portion of the work, and learns the judgment part by proximity. That routine portion is exactly what current AI tools handle best, so the rung that people used to climb is the one being sanded off. Where AI makes an experienced person faster instead of replacing a task outright, the data shows no damage at all. It is also worth being careful about causation. This is an observational study of payroll records during a period that also included high interest rates and a broad tech hiring slowdown, and the authors’ own earlier work spent time separating those threads.
What this means for you: if you are early in your career, the practical read is not “avoid AI-exposed fields” but “get to the judgment part faster”, and being visibly good with these tools is currently one of the few ways a junior person outperforms the automation of their own tasks. If you hire, the finding worth acting on is that gutting junior roles solves this year’s budget and creates a shortage of experienced people in five years. And if you are already established, the data says your position is, for now, not the one under pressure.
Sources
Uber Fined 825 Million Euros for Letting Software Decide
The Dutch data protection authority says Uber suspended and deactivated drivers through automated systems without proper human review. It is the second-largest GDPR fine ever issued.
