Stanford Pushes Back on AI Jobs Apocalypse, But Finds Entry-Level Workers Are the Canary in the Coal Mine

The aggregate data was supposed to settle the debate. It settled the wrong question.

A new policy brief from Stanford's Institute for Economic Policy Research finds little evidence that AI is causing significant job losses right now. Unemployment among workers in the most AI-exposed occupations rose 0.77 percentage points since 2022. Among the least-exposed workers, it rose 0.85 points 1. The jobs AI was supposed to destroy first are weathering the labor market slightly better than everyone else. Online job postings for software developers, a highly AI-exposed occupation, are growing faster than for other roles over the past year 1. Among firms that adopted enterprise AI, employment grew 10 percent over two years, concentrated in companies with the highest per-capita AI spending 1.

Both sides are reading the wrong line.

Recent graduates are facing their toughest job market in years. Unemployment for new graduates reached 5.6 percent in early 2026, up 1.6 percentage points from three years earlier 1. That alone could reflect a softening economy. Researchers Brynjolfsson, Chandar, and Chen, in a study cited by the SIEPR brief, documented a notable decline in employment among early-career workers in AI-exposed occupations, particularly software developers and customer service representatives, since ChatGPT's public launch in November 2022. Employment among older workers in those same occupations remained stable or continued to grow 1. Other researchers have since found similar patterns among young AI-exposed workers in both the U.S. and the U.K. 1.

Junior roles are heavy on the exact tasks large language models handle well: routine research, data analysis, first-draft writing. HR executives told the SIEPR researchers that AI's effects are showing up less in layoffs than in role consolidation and hiring avoidance for positions where AI can automate many tasks 1. The position isn't disappearing from the org chart. It's simply never getting posted.

Causation is messier than the pattern suggests. The Federal Reserve began hiking interest rates in March 2022, eight months before ChatGPT's debut. Two studies cited in the brief found that hiring in AI-exposed occupations started declining after the Fed's policy shift but before ChatGPT appeared 1. Remote work may also be thinning the junior pipeline: when on-the-job learning slows because nobody is sitting next to a senior colleague, employers have less reason to hire inexperienced workers 1. When Brynjolfsson and coauthors added controls for these confounding factors, the employment decline among entry-level workers was not statistically notable until 2024, by which point both AI adoption and model capabilities had advanced enough to make a direct effect plausible 1.

AI is not eliminating professions, at least not yet. It appears to be eliminating the first job that launches one. The authors of the entry-level study liken these young workers to canaries in a coal mine 1. The metaphor is precise in a way the headline numbers are not. A canary doesn't tell you the whole mine is collapsing. It tells you the air is toxic at the bottom, where nobody is checking, and that the damage spreads upward if nothing changes.

Entry-level jobs are not just paychecks. They are where workers develop the judgment, institutional knowledge, and professional instincts that make them valuable at senior levels five and ten years later. If companies stop hiring junior analysts because AI can produce a passable first draft, the senior analyst of 2031 never gets trained. That cost does not surface in any quarterly report. It surfaces as a missing generation of experienced workers in exactly the fields where AI adoption is most aggressive. The disruption already underway is smaller than the mass-layoff forecasts and more specific than the aggregate data suggests. It is the first job, not the last one, that is quietly disappearing.