AI and Entry-Level Jobs

AI and Entry-Level Jobs describes the emerging evidence that generative AI is reducing employment for young workers at the start of their careers, mostly because employers are hiring fewer of them into AI-exposed roles rather than laying off people already in those jobs. By October 2026, the effect is measurable in payroll data but still small at the level of the overall economy.

The Canaries in the Coal Mine

The most-cited research comes from Stanford's Digital Economy Lab. In "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," Erik Brynjolfsson, Bharat Chandar and Ruyu Chen use high-frequency payroll records from ADP covering millions of US workers to track employment by age and by how exposed an occupation is to AI. The idea behind the title is that early-career workers in exposed jobs would be the first to show any effect, the way canaries once warned miners of gas.

The revised version, published on August 12, 2026 with data through June 2026, finds that employment of 22- to 25-year-olds in the most AI-exposed occupations is 19% below what would otherwise be expected. A year earlier, the same measure stood at 15%, and the authors say the gap has widened steadily since August 2025. Other findings matter just as much. The paper documents no economy-wide job displacement. The declines concentrate in roles where AI substitutes for tasks, while roles where AI complements workers show stable or rising employment. And "adjustment is occurring through employment rather than base compensation," meaning wages have barely moved.

Hiring, Not Layoffs

The researchers find that the change is happening "primarily through reduced hiring of young workers rather than increased separations." That distinction is the most important point in this debate, and it explains why the effect has been easy to miss. Layoffs are visible: they make headlines and show up in unemployment claims. A hiring slowdown is invisible: no one is fired, but the junior positions that would once have been filled simply aren't opened. Experienced workers in the same occupations can keep their jobs, even benefit from AI tools, while the bottom rung of the ladder thins out.

The pattern fits how AI is used in practice. Tasks given to entry-level staff, such as drafting, summarizing, first-pass research, routine code and customer support scripts, are the tasks today's language models and work agents handle best. In software, AI coding tools and vibe coding let senior engineers do work that might once have been split with juniors. The long-term worry is a pipeline problem: if fewer people get entry-level jobs today, there may be fewer experienced people to promote later.

The Morgan Stanley Estimate

A Morgan Stanley analysis, reported in August 2026, estimates that AI has added about 15 basis points (0.15 percentage points) to the US unemployment rate, up from about 10 basis points in December 2025. The bank calls this a "modest effect." It finds workers aged 22 to 27 are hit hardest, consistent with the Stanford findings.

Caveats

The estimate should be read carefully. Isolating AI's effect on unemployment is difficult when other forces are moving the labor market at the same time. Tariffs, monetary policy and changes in immigration all affect hiring, and they confound the estimate. A rise of 15 basis points is also small. And because the dominant mechanism is fewer hires rather than job losses, unemployment figures may understate the effect: young people who never get a first job in an exposed field may go into other work, stay in school or not count as unemployed at all.

California's Worker Protections

Policy is starting to respond, though mostly to the more visible risks. On September 30, 2026, California Governor Gavin Newsom signed a package of laws protecting workers from AI-related harms. They:

  • ban inferring workers' emotions from biometric data;
  • require written notice when AI drives mass layoffs;
  • prohibit firing a worker based on an AI decision alone;
  • require risk assessments before chatbots are launched.

These laws address surveillance, layoffs and automated decision-making. None of them directly addresses the main effect the research finds, which is fewer entry-level hires. A notice requirement for AI-driven layoffs does nothing when no one is laid off and positions simply go unfilled. That gap is likely to shape the next round of AI regulation debates about work.

What to Watch

Three questions will decide how big this becomes. First, whether the Stanford gap keeps widening as agentic AI takes on longer, multi-step tasks. Second, whether employers create new kinds of entry-level roles built around supervising AI, in the occupations where AI complements rather than substitutes. Third, whether the effect spreads from young workers to mid-career ones. For now, the evidence supports a narrow conclusion: AI is not yet causing mass unemployment, but it is making the first job harder to get in the occupations it touches most. For the broader picture, see the future of work.

Further Reading