Stanford Study: Entry-Level Employment Gap Tied to AI Exposure Widened to 19% in 2026

College graduates keep saying the same thing: applying for jobs feels like screaming into a void. Now there's harder data behind that complaint.

Stanford University economists led by Erik Brynjolfsson released an updated version of their paper, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," in August 2026. The finding: workers aged 22 to 25 in the most AI-exposed occupations now have employment levels 19 percent below their peers in less-exposed fields, according to Ars Technica. Last year that gap was 13 percent.

How they measured it

The researchers used anonymized, high-frequency payroll data from ADP, then rated occupations for AI exposure using two methods: a labor-market impact gauge from earlier research and the Anthropic Economic Index, which tracks real-world Claude usage by occupation, according to Ars Technica and Digital Today.

Zoom out to the whole economy and the effect nearly disappears. Zoom into workers 22 to 25, and it doesn't. Since 2022, employment for that age group fell about 11 percent in the top 40 percent of AI-exposed jobs. In the bottom 60 percent, least exposed to AI, employment for the same age group grew 10 percent, per Ars Technica.

Employers are not laying off young people in large numbers. They're simply not hiring them in the first place, according to Digital Today's summary of the report.

Automation versus complementary use

The study draws a line between jobs where AI replaces tasks outright and jobs where it just makes existing workers faster. Accountants, auditors, receptionists and information clerks fall into the automation-heavy category. Chief executives and nurses fall into the complementary category, where AI use is linked to stagnation or even employment growth rather than losses, Digital Today reported, quoting the researchers directly: "Automation-focused AI use is consistent with labor replacement, and complementary use is linked to employment stagnation or growth."

The researchers also tested whether jobs built on "codified knowledge" — the stuff you learn from textbooks and formal procedures — are more vulnerable than jobs built on "tacit knowledge," the kind you only pick up through mentorship and repetition, according to Business Story. The pattern held: occupations heavy on codified knowledge saw slower entry-level growth, while occupations built on tacit knowledge saw faster growth for mid-career and senior workers.

College degree share mattered too. Occupations with more college graduates showed smaller gaps between AI-exposed and non-exposed jobs. In fields with fewer college grads, the divide was stark: least-exposed jobs grew, most-exposed jobs shrank, per Business Story.

Brynjolfsson told The Washington Post the trend worries him more now than it did a year ago. "The entry-level effects we're measuring are real, persistent and widening," he said, "and I'm more worried than I was about a labor market that keeps its overall employment level while quietly closing the on-ramp for people starting their careers."

Not every economist is convinced

Harvard economist David Deming pushes back on the AI-causation story entirely. Speaking to NPR (aired on Delaware Public Media and news.wjct), Deming said the decline in junior hiring predates the technology getting blamed for it. "If you look very carefully at the timing, it looks like the decline in junior hiring actually started, like, six months before ChatGPT was released," he said. "What that tells me is it's something else. I think it's more like remote work."

Deming's argument: remote work, which took off during the pandemic, made it harder to train junior employees from a distance, so employers shifted toward hiring more experienced people who need less hand-holding. "The value proposition of hiring a junior person when you know they're going to be at home is just not as great," he said. "And conversely, the value proposition of hiring a more senior person is greater."

That's a real, competing explanation, not a dismissal. It deserves to be taken seriously rather than waved off as economists protecting AI companies from blame. If junior hiring started softening before ChatGPT existed, that's a timing problem for any theory that puts AI at the center of the story.

Brynjolfsson isn't blind to that critique. He told NPR AI "is not the whole story, but it's part of the story, and the evidence is building." His own numbers, cited in the NPR piece, show early-career workers aged 22 to 25 in fields like software development and marketing saw a 16 percent relative employment decline since late 2022, while older workers in those same exposed fields stayed stable or grew.

What graduates are actually experiencing

Georgia Tech engineering graduate Irene Chang, 21, told NPR she has submitted around 450 job applications since last September and landed about 19 interviews with zero offers. Florida State graduate Jacqueline Kline, 25, said she has applied to more than 500 entry-level jobs since December. Both suspect AI is eating the tasks they were counting on to get a foot in the door.

They're not alone in that suspicion. A ZipRecruiter survey cited by NPR found 47 percent of recent graduates say AI has already affected hiring in their field. The Federal Reserve Bank of New York put the unemployment rate for recent grads at 5.7 percent as of June, versus 4.1 percent for all workers.

What's unresolved

Nobody in this reporting claims AI is destroying the labor market broadly. Overall employment numbers stay roughly stable even in heavily AI-exposed fields, according to Ars Technica's summary of the Stanford data. The contraction is narrow, concentrated on people ages 22 to 25 trying to get their first real job.

The open question is whether this is a temporary bottleneck that resolves as the economy adjusts, or a structural shift where the traditional apprenticeship path into skilled professions is closing for good. Brynjolfsson leans toward the latter. Deming isn't sold that AI is even the primary cause. Value Add Pulse's analysis flags a related worry: a Goldman Sachs partner has separately warned that AI tools may be eroding the reasoning skills of the junior employees who do get hired, meaning the jobs that remain increasingly involve supervising AI output rather than doing the work that used to build expertise.

Neither the Stanford team nor Deming has an answer yet for where the next generation of senior professionals comes from if the entry-level roles that used to train them keep shrinking.