Early-career AI jobs now shrinking 3.8% a year
Daily Briefing | June 28, 2026
Editor’s note: This is one of the first things I remember distinctly crossing my mind as we started talking about AI impacting the future of work ~2 years ago. As a recovering consultant, I saw my own career path be automated by AI - the meeting notes, the first draft of the PowerPoint deck, the survey data cleansing, etc. There’s even more evidence that prediction is (not becoming… is) reality.
Erik Brynjolfsson spent the past year fielding every objection to his finding that AI is hollowing out the bottom rung of the job market. Interest rates, tech overhiring, remote work, pandemic noise: critics offered each in turn. He kept updating the data with ADP Research, and the pattern held through every cut. The Stanford Digital Economy Lab now runs the numbers live on its Canaries Dashboard, extended to April 2026, and the headline figure stays calm while the figure underneath it does not. Across all workers, AI-exposed occupations are nearly flat. For workers ages 22 to 25 in those same occupations, employment is shrinking 3.8% a year, and the decline has deepened by about half a percentage point every month since the original paper.
By the Numbers
Workers ages 22 to 25 in highly AI-exposed occupations: employment shrinking 3.8% per year as of April 2026 (Stanford Digital Economy Lab / ADP Research).
The same occupations across all ages: down only 0.2% year over year, against 0.1% growth for the least-exposed roles (Stanford Digital Economy Lab / ADP Research).
The early-career decline has steepened from 2.8% to over 4% per year since the original August 2025 paper, falling roughly half a point per month (Stanford Digital Economy Lab / ADP Research).
Mid-career workers ages 31 to 34 in AI-exposed roles: down 1.7% year over year; workers ages 35 to 40: up 2% (Stanford Digital Economy Lab / ADP Research).
Dashboard coverage: 4.6 million workers across more than 730 occupations, roughly one in six American workers (Stanford Digital Economy Lab / ADP Research).
Research and the Labor Market
AI is not cutting jobs across the board. It is cutting the on-ramp.
Last August, Brynjolfsson and collaborators published the first large-scale evidence that AI was pulling down employment for the youngest workers in the most exposed jobs. The pushback was immediate. Google economists pointed to interest rates; others blamed tech-sector overhiring, remote-work distortions, and pandemic-era noise. This month Apollo’s Torsten Slok was still asking where the AI jobs crisis was. Rather than defend the paper as written, Brynjolfsson partnered with ADP Research, which sees payroll data on about one in six U.S. workers, and built the Canaries Dashboard to track the effect in close to real time.
The aggregate picture stays quiet. AI-exposed occupations contracted 0.2% year over year as of April 2026, while the least-exposed grew 0.1%. Split the data by career stage and a fault line opens. Employment for workers ages 22 to 25 in highly AI-exposed occupations is shrinking 3.8% a year, and the trend has deepened by roughly half a percentage point each month since the original paper. The least-exposed jobs in that age group are growing 2%. Mid-career workers ages 31 to 34 are down 1.7%; workers ages 35 to 40 are up 2%. Nela Richardson, ADP’s chief economist, traces the split to a simple mechanism: AI reaches tasks before it reaches jobs, and the tasks it reaches first, retrieving, summarizing, scheduling, formatting, are the ones handed to people at the start of their careers. Brynjolfsson removed the entire tech sector, isolated remote-work effects, and tested the rate-sensitivity argument, which points the wrong way because rate-sensitive work like construction has low AI exposure. The pattern survived each test.
Source: Fortune, June 27, 2026, “It’s not going away: The Stanford economist who called the AI entry-level jobs crisis early has the receipts.” Read it here.
Why it matters: The headline labor numbers will keep looking fine while the entry-level pipeline thins, which means companies can damage their future talent base without it showing up in any topline they watch. Track hiring and retention by age band and AI exposure, not just total headcount. The cost of cutting the bottom rung lands years later, when there is no trained mid-career bench to promote.
What Workforce Leaders Are Watching
If AI absorbs the retrieving, summarizing, and formatting that junior roles run on, where will your people learn the judgment that senior roles require? What replaces the apprenticeship that those tasks used to be?
Your total headcount can hold steady while your 22-to-25 cohort quietly shrinks. Are you tracking hiring and attrition by age band and AI exposure, or only the topline that hides the shift?
The data shows augmented occupations growing and automated ones contracting. For each role on your team, are you deploying AI to extend what people do or to remove the task entirely, and do you know which you are choosing?
The early-career decline has compounded about half a point a month for nearly two years with no sign of reverting. If that holds another year, what does your entry-level hiring plan for 2027 look like, and who owns rebuilding the on-ramp if you let it close?
This briefing was prepared automatically by the Workforce Rewired research assistant. All stories include direct source links.