A Stanford Digital Economy Lab study, built on ADP payroll data from 4.6 million U.S. workers, finds that employment for 22-to-25-year-olds in the most AI-exposed occupations now sits 19% below where it would be if it had tracked their less-exposed peers. Experienced workers in the same occupations show no comparable gap — the hit is landing almost entirely on people starting their careers.
Short answer: Yes — Stanford's "Canaries in the Coal Mine" study, using ADP payroll data through mid-2026, found employment for 22-25 year olds in AI-exposed jobs is 19% below its expected trend, up from 13% a year earlier. The gap comes from reduced hiring of young workers, not layoffs of existing staff, and it's concentrated in "codifiable" entry-level tasks like retrieval, summarization, and formatting.
I read a lot of AI-and-jobs research for this site, and most of it is either too broad to act on ("AI will affect 40% of jobs globally") or too speculative to trust (single-company layoff PR dressed up as a labor trend). In my testing this week, I went straight to the source behind the headline — Stanford economist Erik Brynjolfsson's Canaries Dashboard and the underlying paper he wrote with Bharat Chandar and Ruyu Chen — instead of the aggregator summaries repeating a number without the methodology behind it. What's actually in the data is narrower than "AI is killing jobs" and more concrete than most of the coverage suggests.
What you'll need
You don't need a statistics background to check this yourself, just three things: fifteen minutes, a browser, and your own age bracket and occupation in mind if you want to see how the data applies to you specifically. The core source is the Stanford Digital Economy Lab’s paper, which draws on ADP payroll records covering roughly one in six American workers across more than 730 occupations — a much larger and more current dataset than the Current Population Survey samples most competing studies rely on. Fortune's June 2026 interview with Brynjolfsson is the fastest plain-English summary if you want the findings without reading the paper itself.
Step-by-step: what the Stanford study actually found
1. Start with the headline number, and its trend
The core finding: employment for workers aged 22-25 in the most AI-exposed occupations is 19% below where it would be had it kept pace with less-exposed peers, as of the paper's August 2026 revision. That gap was 13% a year earlier, in the researchers' original August 2025 version of the same paper — meaning the divergence is widening, not stabilizing.
2. Check who's actually affected — and who isn't
This is the detail most secondhand coverage drops: the gap is almost entirely a young-worker phenomenon. Experienced workers in the exact same AI-exposed occupations show no comparable decline. Employment for 22-25 year olds in high-exposure jobs has been shrinking at roughly 3.8% a year, while low-exposure jobs for the same age group grew about 2%. Workers aged 31-34 in exposed occupations saw a smaller, 1.7% year-over-year dip; workers 35-40 actually grew about 2%.
3. Understand the mechanism: hiring, not firing
The researchers were specific about this, and it matters for how you read the number: the disparity "operates primarily through reduced hiring of young workers rather than increased separations." Companies aren't laying off entry-level staff in AI-exposed roles in large numbers — they're simply hiring fewer of them in the first place. That's a quieter, slower-moving trend than a layoff wave, and it's part of why it took payroll-level data to surface it instead of headline unemployment numbers.
4. Look at which tasks are actually getting automated
Brynjolfsson's team ties the effect to "codifiable" tasks — retrieval, summarization, scheduling, formatting — that don't require much on-the-job experience to perform. Those tasks make up a disproportionate share of what entry-level workers are hired to do, which is why the effect concentrates there rather than spreading evenly across an occupation's full range of tasks. Roles where AI complements rather than substitutes for a task showed flat or rising employment, including for younger workers.
5. Note what the study does not claim
Brynjolfsson has been explicit that he doesn't think a broad "AI jobs apocalypse" is likely, even as he stands by the entry-level finding — his quote to Fortune was "whatever it is, it's not going away." Total employment in AI-exposed occupations has still grown since ChatGPT's late-2022 launch (about 1.1% annually), just more slowly than less-exposed occupations (about 2%). This is a narrowing on-ramp, not a shrinking labor market overall.
Example prompts you can copy
Paste these into ChatGPT, Claude, or Gemini to apply the findings to your own situation:
- Exposure check: "Based on the Stanford 'Canaries in the Coal Mine' framework — where AI first automates codifiable tasks like retrieval, summarization, and formatting — how exposed is a [job title] role for someone with 0-2 years of experience?"
- Resume repositioning: "I'm applying for entry-level [field] roles. Given that AI is automating routine, codifiable tasks first, what specific skills or project experience should I highlight to show I can do judgment-based work, not just task execution?"
- Interview prep: "What questions might an interviewer for an entry-level [role] ask to distinguish a candidate who can supervise and correct AI output from one who can only produce a first draft?"
- Company research: "Search for whether [company name] has publicly discussed reducing entry-level hiring due to AI adoption, and separate any confirmed statements from speculation."
Common mistakes to avoid
The mistake I see most often is treating this study as proof of a general AI jobs apocalypse — it isn't; the paper's own authors reject that framing, and total employment in AI-exposed fields is still growing, just more slowly. Second is ignoring the hiring-versus-firing distinction: if you already have a job in an AI-exposed field, this data isn't describing your layoff risk, it's describing how hard the door is to open for people trying to get in behind you. Third is quoting the 19% figure without its trend line — a number that was 13% a year earlier tells you the gap is accelerating, which changes how urgently you should act on it. Fourth is assuming every entry-level role is equally at risk; the effect concentrates in codifiable, low-judgment tasks, not roles built around apprenticeship-style mentorship or hands-on physical work. Fifth, and this cost me time when I first read the coverage: several aggregator posts cited the 19% figure from the original 2025 paper without noting it had already been revised upward in the August 2026 update — always check the publication date on a stat like this before repeating it.
Stanford study claim vs. what the data actually shows
| Claim | What the study actually shows |
|---|---|
| "AI is causing a broad jobs apocalypse" | Total employment in AI-exposed occupations grew ~1.1% annually since late 2022 — slower than less-exposed roles' ~2%, but still growing |
| "Entry-level workers are being laid off because of AI" | The gap comes mainly from reduced hiring, not increased separations, per the study's authors |
| "Every young worker is equally affected" | Experienced workers in the same AI-exposed occupations show no comparable employment gap |
| "This is a one-time finding" | The 22-25 employment gap widened from 13% (Aug. 2025) to 19% (Aug. 2026) in the same ongoing research |
| "AI affects all entry-level tasks equally" | The effect concentrates in "codifiable" tasks — retrieval, summarization, scheduling, formatting |
Tools that make this easier
If you want the fuller labor-market picture beyond this one study, I've covered the same underlying tension in why the AI jobs apocalypse probably isn’t coming anytime soon and in separating AI jobs hype from reality, both of which pull in the Yale, Brookings, PwC, and Challenger Gray data that sits alongside this Stanford research. If you're early in your career and want to act on this rather than just read about it, my best AI tool for job searching guide and how to use ChatGPT to write a resume walkthrough cover positioning yourself as someone who directs AI output rather than someone whose tasks it replaces. My AI tool ratings hub is a good next stop if you're deciding which tools are actually worth building fluency in, and how to use ChatGPT for learning is useful if you're trying to close a skills gap fast. If budget's the barrier to any of this, my free AI tools roundup covers where to start at no cost.
My take
This is one of the more credible pieces of AI-jobs research I've read this year, mostly because it's built on payroll data instead of survey samples or single-company anecdotes, and because the authors are careful not to oversell it into a general apocalypse claim. The 19% gap for 22-25 year olds in AI-exposed fields is real and, per the study's own year-over-year comparison, getting wider rather than settling down. If you're already employed, this isn't your risk signal. If you're trying to break into an AI-exposed field for the first time, it's worth treating as a real headwind — and the practical response isn't panic, it's making sure your resume and interview answers show judgment and oversight skills that go beyond the codifiable tasks AI is absorbing first.
Frequently Asked Questions
Is AI actually hitting entry-level jobs hardest right now?
Yes, according to Stanford's Canaries in the Coal Mine research. Employment for 22-25 year olds in the most AI-exposed occupations sits 19% below its expected trend as of the August 2026 update, while experienced workers in the same occupations show no comparable gap.
Is this the same as an "AI jobs apocalypse"?
No. Total employment in AI-exposed occupations has still grown since ChatGPT launched in late 2022, about 1.1% annually — just more slowly than the roughly 2% growth in less-exposed occupations. The study's own authors, including lead researcher Erik Brynjolfsson, don't describe this as an economy-wide collapse.
Are companies laying off entry-level workers because of AI?
Mostly not directly. The researchers found the employment gap comes primarily from reduced hiring of young workers, not increased separations of workers already employed — companies are opening fewer entry-level slots rather than cutting existing staff in large numbers.
Which entry-level tasks are most at risk?
"Codifiable" tasks that don't require much experience — retrieval, summarization, scheduling, and formatting — show up as the most affected in the study. Roles built around judgment, mentorship, or tasks where AI complements rather than replaces the work showed flat or rising employment.
How is this data different from other AI-jobs studies I've seen?
It's built on ADP payroll records covering 4.6 million workers and over 730 occupations, updated through mid-2026 — a larger, more current, and more granular dataset than the government survey data behind most competing studies, which is part of why it can isolate an age-specific effect that broader unemployment numbers miss.