Is AI Taking Entry-Level Jobs? Where the Data Agrees and Conflicts

By AI Agent Hub Editorial Desk · Review method · Corrections

Two credible sources, published within months of each other, appear to say opposite things about entry-level work. Payroll records show young workers in AI-exposed jobs losing ground. Employer surveys show hiring intentions rising. Reporting that treats one as the truth and the other as noise gets the story wrong, because the two are measuring different things.

The payroll finding

The most-cited result comes from Stanford's Digital Economy Lab, in an August 2026 update to a paper titled "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence." Using anonymized ADP payroll records covering more than 26 million U.S. workers:

• Employment of workers aged 22 to 25 in the most AI-exposed occupations sits 19 percent below where it would have been had it tracked less-exposed peers, up from a 15 percent shortfall a year earlier.
• Since 2022, employment for that age group fell about 11 percent in the top 40 percent of AI-impacted jobs, while growing 10 percent in the remaining 60 percent.
• Across all ages, the study found little to no difference in overall employment between more- and less-exposed occupations. The effect is concentrated in young workers.
• The decline appears through lower hiring rates, not through layoffs or resignations, and shows up as fewer jobs rather than lower wages.

The lead researcher, Erik Brynjolfsson, summarized the concern as a labor market that "keeps its overall employment level while quietly closing the on-ramp for people starting their careers." That framing — an on-ramp problem rather than a layoff problem — is what makes the finding hard to see in headline unemployment numbers.

The surveys say something different

Employer-side data points the other way:

None of these measures the same quantity as the payroll study. Surveys capture what employers intend or believe. Job ads capture what employers ask for. Payroll captures what actually happened. A market can simultaneously post more openings, demand more senior skills in them, and hire fewer 22-year-olds — and that combination is roughly what the full picture shows.

Why causation is still contested

The Stanford authors state their result is descriptive, not causal. Other researchers have published specific reasons for caution:

• Researchers at the Federal Reserve Bank of New York, using real-time job-postings data, found it "difficult to attribute the recent slowdown in entry-level hiring to AI alone," and observed that junior and senior hiring within highly exposed occupations have moved broadly in parallel.
• A separate New York Fed study attributed 64 percent of the rise in unemployment among young college graduates to remote work rather than generative AI, noting the increase predates rapid AI diffusion.
• Harvard economist David Deming has raised timing problems with a simple AI explanation, and University of Chicago economist Anders Humlum has pointed to firm-level evidence that complicates it.

Underlying softness is nonetheless real. U.S. unemployment for workers aged 20 to 24 stood at 7.1 percent in August 2026 against 4.1 percent overall, and the New York Fed's recent-graduate series read 5.6 percent with underemployment near 42 percent in Q2 2026. Globally, the ILO reported youth unemployment at 12.4 percent in 2025, or 67 million people, and estimated 6.1 percent of jobs held by 15-to-29-year-olds sit in highly exposed occupations.

Codified versus tacit knowledge

The most actionable finding is not the headline number but the split it identifies. The Stanford study finds the effect concentrated in work built on codified knowledge — the kind that can be taught from textbooks or written procedures — and absent or reversed in work requiring tacit knowledge gained through practice and mentorship.

Anthropic's Economic Index draws the same line from usage data, separating tasks where AI is automative, substituting for the worker, from tasks where it is augmentative. Reported effects cluster in the automative group: accounting, auditing, reception, and clerical work, rather than augmentative roles such as nursing or executive management.

A published breakdown of programming occupations makes the point sharply. U.S. programmer employment fell 27.5 percent between 2023 and 2025, while the broader and more design-oriented software developer category fell only 0.3 percent. Same industry, same technology, opposite outcomes — the difference is how much of the role is routine execution versus judgment.

Education appears to buffer the effect

The Stanford analysis also reports that occupations with a higher share of college graduates showed more muted differences between exposed and less-exposed roles, while in occupations with few college graduates, the least-exposed roles grew and the most-exposed ones declined. Read cautiously, this suggests the effect is not uniform across the education distribution, though the mechanism is not established.

What is defensible as of September 2026

Putting the sources together, the honest statement is narrower than either the "AI is eliminating jobs" or "nothing is happening" version:

1. Entry-level employment in highly AI-exposed occupations has weakened measurably for workers aged 22 to 25. This is the most reproducible pattern across sources.
2. The mechanism is reduced hiring rather than layoffs, which is why aggregate unemployment looks stable.
3. The effect is concentrated in codified, routine, automative work rather than across the labor market.
4. Causal attribution to AI remains unresolved, with credible competing explanations including remote work and a broader hiring slowdown.
5. Employer intentions and realized hiring are diverging, and both are worth watching.

For anyone choosing what to learn rather than what to legislate, the practical signal is the codified-versus-tacit split: work that can be fully specified in writing is where substitution is showing up first.

Frequently asked questions

How much has entry-level employment fallen in AI-exposed jobs?

Stanford's August 2026 update, using ADP payroll records, found employment of 22-to-25-year-olds in the most AI-exposed occupations was 19 percent below the level implied by less-exposed peers, up from a 15 percent shortfall a year earlier.

Do employer surveys contradict that?

They point in a different direction: NACE projected a 5.6 percent increase in Class of 2026 graduate hiring and Strada found leaders 2.7 times more likely to expect AI to increase entry-level hiring. Surveys measure intention while payroll measures filled positions, so they can diverge without either being wrong.

Has AI been proven to cause entry-level job losses?

No. The Stanford authors describe the finding as descriptive rather than causal, New York Fed researchers found it difficult to attribute the slowdown to AI alone, and another New York Fed study attributed 64 percent of the rise in young-graduate unemployment to remote work.

Which jobs are most affected?

Roles built on codified knowledge teachable from textbooks or procedures — accounting, auditing, clerical and reception work — rather than roles requiring tacit knowledge gained through practice and mentorship.

Is programming employment falling?

Published figures show U.S. programmer employment fell 27.5 percent from 2023 to 2025 while the broader software developer category fell only 0.3 percent.

Primary sources

• Stanford Digital Economy Lab — "Canaries in the Coal Mine?" (August 2026 update)
• Federal Reserve Bank of New York — research on entry-level hiring attribution
• NACE Job Outlook 2026 — graduate hiring projections
• PwC 2026 Global AI Jobs Barometer — job advertisement analysis
• International Labour Organization — global youth unemployment statistics
• What recorded AI incidents actually show — the same evidence-first approach applied to AI harm