AI Jobs Research Shows Junior Hiring Falls While Total Jobs Hold

The labor market is changing at the bottom first

The early predictions about generative AI and jobs have run ahead of the evidence. Studies published in 2026 now give a clearer picture than the predictions did. Total employment has held up. Underneath that, hiring has shifted. Young workers are being hired less in a narrow set of jobs, and many existing jobs are changing as AI tools move into everyday office work. Fewer people are losing jobs outright.

A Harvard Business School working paper gave one of the earliest large-scale readings of this shift. The researchers looked at nearly every U.S. job posting from 2019 through March 2025. After ChatGPT's launch, postings for jobs built on structured, repetitive tasks fell 13%, while demand for analytical, technical and creative roles rose 20%1. The biggest drops were in finance and technology, and automation-prone jobs listed 7% fewer skills in their postings1. Co-author Suraj Srinivasan argues that generative AI is creating new demand in roles where it assists people, not just removing jobs1.

The paper reads the data as transformation, not collapse. Most later research supports that reading. The exception is the entry level, which is where most of the current debate is focused.

The "canaries" keep getting louder

The most cited evidence comes from Stanford's Digital Economy Lab. Its August 2026 update of "Canaries in the Coal Mine?" draws on ADP payroll records. It finds that employment for 22- to 25-year-olds in the most AI-exposed jobs is now about 19% below where it would be if it had kept pace with peers in less-exposed jobs11. In the 2025 version, that gap was 15%18. In raw numbers, young-worker employment in the two most exposed fifths of jobs fell about 11% between November 2022 and June 2026, while the same age group grew about 10% in the least-exposed jobs18.

The details matter. The decline comes mostly from fewer hires, not from more layoffs or quits, and it shows up in jobs, not pay11. It is concentrated in jobs where AI does the work instead of assisting with it, such as accountants and receptionists. Jobs where AI mostly assists show flat or rising employment11. Erik Brynjolfsson, the lead researcher, has said he is now more concerned about a labor market that keeps total employment steady while quietly cutting off entry for new workers11.

Other data points the same way. A Census-based study found that hiring of 22- to 24-year-olds in the most AI-exposed fifth of industries fell about 9% after ChatGPT's release and has not recovered2. A companion study followed graduates from the most exposed tenth of college majors, mostly computer science. They were 5 percentage points less likely to be employed right after graduation, and their starting pay fell about 13%, similar to graduating into a recession2. Indeed Hiring Lab found that entry-level roles fell from 29% to 10% of salaried postings in the most AI-exposed jobs between 2021 and 202618.

The Dallas Fed offers a likely explanation. AI can replace codified knowledge, the textbook skills that new graduates bring, while complementing the tacit knowledge experienced workers build on the job15. Wages are rising in AI-exposed jobs that reward experience, while exposure is linked to slower wage growth in jobs that reward experience less15. If that pattern holds, the usual white-collar career path, where juniors do routine work while they learn, becomes harder for employers to justify15.

Where the evidence diverges

The research does not all agree, and the disagreements are substantive. Studies of entire national workforces in Finland and Norway found no significant AI effect on young workers' employment. Norwegian data suggest junior employment in exposed jobs began falling before ChatGPT2. A U.S. study using unemployment-insurance records found unemployment risk in AI-exposed jobs started rising in early 2022, also before ChatGPT2. Brookings, reviewing the field, notes that the overall declines may reflect older trends and the rise of remote work as much as AI3.

A second disagreement is about which way the effect runs. A 41-country Stanford study of multinational affiliates found that the drop in the junior share of employment at AI adopters comes mainly from growth in senior jobs, not cuts to junior ones18. Senior employment at adopters rose 6.7% compared with similar non-adopters18. A U.S. study of firm-level AI spending found that companies investing heavily in AI grew employment by 10%, and entry-level employment by 12%, compared with firms that adopted later2.

Scale matters too. The Dallas Fed estimates that even if all of the lost employment among young, highly exposed workers had become unemployment, it would add only 0.1 percentage point to the national unemployment rate since November 20226. Anthropic's own study found no detectable increase in overall unemployment among highly exposed workers7.

The best overall reading, in this analysis, is that both sides are partly right. Total employment is stable. Displacement is real but concentrated, mostly in the U.S. and in fields where AI automates tasks, and it shows up mainly as fewer new hires. Pre-ChatGPT trends probably explain part of it, but they do not explain why the gap keeps widening in the most recent payroll data.

Layoff announcements and what firms actually do

Announced layoffs tell a different story. Challenger, Gray & Christmas reported that employers cited AI in 120,136 planned U.S. job cuts through September 2026, about 21% of all announced cuts. That made AI the most-cited reason10. Yet a Federal Reserve Bank of Atlanta paper surveyed nearly 6,000 executives in four countries, and more than 80% said AI had not affected employment or productivity over the past three years10.

One way to explain the gap is that "AI" has become a convenient reason to give for cuts that companies would have made anyway. That explanation is speculative, but it fits both sets of numbers. Business-side surveys also lean optimistic. In a Strada survey, 2.7 times as many senior talent leaders expect AI to increase entry-level hiring in 2026 as expect it to decrease it13. Employers also report that AI is giving junior staff more analytical work and less routine admin work13. Entry-level jobs are not disappearing in that picture, but they are getting harder20.

Adoption is broad but shallow

AI use at work has spread quickly but not deeply. Gallup's May 2026 survey found that 52% of U.S. employees use AI in their role. Only 30% use it a few times a week or more, and 15% use it daily29. The share who say their employer has formally integrated AI jumped from 41% to 47% in one quarter29. Only 25% say their organization has communicated a clear AI plan28.

The productivity data shows the same split between tasks and organizations. About two-thirds of employees at AI-adopting organizations say it has improved their productivity. Only 14% strongly agree it has changed how work gets done28. The strongest productivity ratings, 77%, come from people using AI for coding help and process automation, though only 16% of users use it for each29. Most people use AI for writing and research, which earn lower ratings29.

The gains also come with costs. SHRM found that workers save about six hours a week with AI but spend about four hours fixing its output27. In the same survey, 44% of AI users described their output as "AI slop"22. BCG reports larger savings: 42% of regular frontline users say they save eight hours a week. BCG also says most organizations have not worked out how to turn that time into value24. Meanwhile, the share of BCG respondents whose organizations have built AI agents into workflows rose from 13% to 30% in a year24.

In Germany, the Konstanz study found that work use of AI rose only from 35% to 38%. Just 55% of users said their most-used tool came from their employer23. PwC's global survey describes a growing divide between workers. Nearly two-thirds of workers used AI in the past year. More than half of a small group of "front-runners" use generative AI daily, compared with just 11% of the core workforce26.

Why it matters

The research points to one conclusion: the outcome depends on how employers use the technology, not on AI itself. The Harvard researchers argue firms should treat AI as a way to assist workers, not only to cut costs, and should invest in reskilling1. A review of the academic literature finds the same, linking AI adoption to rising skill demands and wider wage gaps between workers with and without AI skills4. Brookings points to occupational mobility as the place where AI's effects already show up in official records, and moving into better jobs is a core source of wage growth3.

The main risk is not mass unemployment. It is that fewer people get the early-career experience that turns new hires into the experienced workers firms still value. Companies that are cutting junior hiring while their senior staff become more productive should consider who will fill senior roles in ten years. The current data does not answer that question.