Key takeaways
- The disruption is concentrated, not broad: a Stanford/ADP study found a 13% relative decline in employment for workers aged 22-25 in the most AI-exposed occupations since late 2022, while older workers in the same roles held steady or grew 6-9%.
- In software engineering and customer service, entry-level employment fell roughly 20%; large models compete most directly with the codified, book-learned knowledge a new graduate brings.
- AI also runs the hiring funnel: NYC's Local Law 144 requires bias audits, public summaries, and candidate notice — but the city's own Comptroller found enforcement ineffective; the teeth are private litigation (Mobley v. Workday) and the EU AI Act's high-risk classification.
- The entrance is compressed from both sides: the first job is rarer (AI does the work it used to contain) and harder to get (AI screens for it).
- There is an upside: where AI augments rather than replaces, employment grew, and entry-level AI-role starting pay rose about 12% in 2024-25 — the advantage is shifting to workers who use AI rather than compete with it.
The conversation about AI and work runs on a fear of mass unemployment, a single great wave of layoffs. The data says something narrower and more unsettling. The disruption so far is concentrated at the entry level, in a handful of exposed occupations, and it shows up not as firings but as jobs that quietly never get posted. Meanwhile the AI doing the most consequential work in HR is not replacing anyone. It is screening the candidates, under a transparency law that its own city found almost no one is enforcing.
The clearest casualty is the bottom rung of the ladder. Keep it in view.
The number that reframes the panic
Stanford economists, using ADP payroll records covering millions of workers, found that early-career employees aged 22 to 25 in the most AI-exposed occupations have seen a 13 percent relative decline in employment since late 2022, while older workers in the same roles held steady or grew between 6 and 9 percent. In software engineering and customer service, entry-level employment fell by roughly 20 percent. The effect held even after controlling for firm-level conditions, and it showed up in headcount rather than pay.
The mechanism is specific. Large models are trained on exactly the codified, book-learned knowledge a new graduate brings, so they compete most directly with the least experienced workers. Tacit, hard-won judgment is the buffer, which is why senior people are largely untouched. Corroborating signals are piling up: Bank of America noted that the unemployment rate for recent graduates has begun to exceed the overall rate for the first time in recent memory.
This is not the wave everyone braced for. It is erosion, and it starts at the on-ramp.
| Category | Entry-level employment decline (22-25) |
|---|---|
| Exposed roles | 13% |
| Software & CS | 20% |
Cite or embed this
Free to reuse with a credit link back to The Counter Brief.
The other half: AI is running the hiring funnel
While AI thins entry-level roles, it is also deciding who gets the ones that remain. Resume screeners, candidate rankers, and interview analyzers now sit between applicants and recruiters. New York City was first to regulate them: Local Law 144 requires an annual independent bias audit, a published summary, and advance notice to candidates with an opt-out, enforced by the Department of Consumer and Worker Protection.
Then comes the honest part. The city’s own Comptroller audited that enforcement in late 2025 and found it ineffective, citing misrouted complaints and superficial reviews of the bias audits employers posted. The first-of-its-kind law to make AI hiring accountable was barely being checked. The pressure that does have teeth is private litigation, like the Mobley v. Workday case over vendor liability, and the broader regimes closing in: the EU AI Act treats employment decisions as high-risk, and US disparate-impact law under Title VII still applies to an algorithm the way it applies to a manager.
The altitude shift
From the screening score to the career. A resume ranked below a model’s threshold is a single rejection. Multiplied across an industry that has automated the bottom rung, it becomes a structural problem: the first job is both harder to get, because AI screens for it, and rarer to begin with, because AI does the work it used to contain. The same technology compresses the entrance from both sides.
There is a real upside on the ledger, and it is worth naming. The harm concentrates where AI automates; where it augments, employment grew. Workers who use the tools are gaining, and starting salaries for entry-level AI roles rose about 12 percent from 2024 to 2025, with 23 percent of employees already using generative AI weekly. The ladder is not gone. Its bottom rungs are being rebuilt around using AI rather than substituting for it. The sales development rep, an archetypal first job, is already being rebuilt this way.
What to do with this
The instinct to ask whether AI will take the jobs is the wrong frame, because the answer is already more precise than that. It is taking the first job in exposed fields, leaving the experienced largely alone, and screening whoever competes for what remains. For anyone running a team, the practical move is not to brace for a layoff wave that the data does not show. It is to notice that the apprenticeship pipeline is thinning at the bottom, and to decide, deliberately, how a new hire is supposed to gain the judgment that AI cannot yet fake when the entry-level work that used to teach it has been automated away.
The Counter Brief — one email, every Monday.
The week's AI-for-revenue moves in a 5-minute read: which tools are worth the budget and which to skip, plus what to do this week. Source-checked, no vendor decks.
Edited by Aditya Marin Gasga
Free. One click to unsubscribe.
Frequently asked questions
Is AI causing mass layoffs?
The evidence so far points to concentration, not a broad wave. A Stanford study found a 13 percent relative decline in employment for workers aged 22 to 25 in the most AI-exposed occupations since late 2022, while older workers in the same roles were stable or grew. The effect appears in hiring and headcount more than in firings or pay.
Which jobs are most affected?
Entry-level roles in fields like software engineering and customer service, where early-career employment fell by roughly 20 percent. These are occupations heavy in codified knowledge that overlaps with what large language models do well.
Is it legal to use AI to screen job candidates?
It is regulated, not banned. New York City's Local Law 144 requires annual bias audits, public summaries, and candidate notice, the EU AI Act treats employment decisions as high-risk, and US anti-discrimination law still applies. Enforcement varies, and New York City's own audit found its law was being enforced ineffectively.
Is there any upside for workers?
Yes. Employment grew in roles where AI augments rather than replaces, workers who use the tools are gaining, and starting pay for entry-level AI roles rose about 12 percent from 2024 to 2025. The advantage is shifting toward people who can use AI well rather than compete with it.