Three studies published in 2026 now measure what AI is doing to the bottom of the career ladder, and they agree on something narrower than the headlines. Companies that adopt AI are adding senior people faster than junior ones, and young workers in the most exposed occupations are being hired less. None of the three found a wave of entry-level layoffs.
The biggest of the three is a Stanford Digital Economy Lab working paper by Bharat Chandar and Bouke Klein Teeselink, published September 21, 2026. It analyzes 1.25 billion job postings and 154 million employment records across 41 countries, and its abstract says the fall in the junior share of the workforce "primarily comes from growth in senior employment rather than from a fall in junior employment."
That distinction decides what an early-career candidate should do next, so it is worth reading the evidence slowly.
What did the 41-country Stanford study find?
It found that AI adopters grow top-heavy. Chandar and Klein Teeselink compare foreign affiliates of multinationals that adopted generative AI with matched affiliates that did not, in the same country and industry and of similar size before ChatGPT launched. In Chandar's own summary, adopting affiliates saw senior employment rise 6.7% relative to those matches, which "adds up to a two percentage point decline in the share of employees who are junior."
Overall employment at adopters rose about 3.3% relative to matched non-adopters, a result Chandar calls marginally statistically significant. The junior share fell in 23 out of 31 countries where the team could get precise estimates, and the decline was statistically significant in seven of them.
Two limits travel with the finding. The comparison is adopters against non-adopters, so it describes how AI reshapes a firm, and Chandar writes that this tells us little about total employment across the economy. Adoption is measured from job postings that mention generative AI, which the authors treat as "a lower bound on AI adoption."
The paper's abstract attributes the shrinking junior share primarily to growth in senior employment. Seniors are being added, and the junior slice shrinks as a share of the whole.
Are young workers in AI-exposed jobs being hired less?
In the US payroll data, yes, and the gap is widening. The Stanford Digital Economy Lab's August 2026 update of "Canaries in the Coal Mine?", by Erik Brynjolfsson, Chandar and Ruyu Chen, uses ADP payroll records. It finds employment among workers ages 22 to 25 in highly AI-exposed occupations now about 19% below where it would be had it kept pace with same-age workers in less-exposed occupations. At the July 2025 data vintage that shortfall was 15%.
Measured in levels, the same update reports that employment of workers ages 22 to 25 in the two most exposed quintiles fell about 11% between November 2022 and June 2026, while the same age group in the three least-exposed quintiles grew about 10%. "Experienced workers show no comparable gap," the authors write.
The mechanism matters most for a job seeker. The canaries update says the adjustment "appears to operate primarily through reduced hiring of young workers rather than increased separations." People already in these jobs are mostly keeping them. Fewer new ones are being let in.
The authors are careful about cause. They call these "descriptive patterns, not causal estimates" and note that the gaps between more- and less-exposed young workers "shrink when we account for education." Several alternative explanations, including interest rates and remote work, did not appear sufficient to account for the pattern in their tests.
Are there fewer entry-level job postings?
In the most exposed fields, entry-level roles are a much smaller share of what gets posted. Indeed Hiring Lab's September 2026 analysis of postings that advertise an annual salary found the entry-level share in the most AI-exposed occupations fell from 29% to 10% between 2021 and 2026.
Read that figure on its own basis. It covers salaried postings only, a population Indeed says skews senior (about 37% senior by 2026, against about 14% across all US postings), and it is a share of postings, never a count of jobs. The pay side of the same study, and why it points the other way from the fear, is covered in our breakdown of AI exposure and pay.
Four-Leaf's own index shows the same shape without any AI lens at all. Across the active postings in the June 2026 snapshot, classified by title, senior-and-above titles made up 26% against 9.0% for intern, junior, entry-level, new-grad and associate titles. That is a single snapshot, so it cannot say whether the tilt is new or what caused it. The full write-up of that index has the rest of the numbers.
Indeed's and Four-Leaf's figures each show the entry tier thinner than the senior tier, on different bases, so do not add them together.
Is AI taking entry-level jobs or changing who gets hired?
The evidence so far says AI is changing who gets hired more than it is removing people already hired. Chandar put his own reading plainly: "no widespread displacement so far, likely decently-sized negative impacts for exposed junior workers."
For a candidate, that difference is practical. A layoff wave would mean the jobs are gone. A hiring tilt means entry-level roles still exist but make up a smaller share of hiring in exposed fields. That is a targeting problem, and targeting is something a candidate controls.
The canaries update also splits occupations by the kind of knowledge they run on. It found employment falling among young workers in occupations that rely heavily on codified knowledge, the formal, documented kind taught in textbooks, and rising among experienced workers in occupations that lean on "tacit knowledge acquired through practice, mentorship, and repeated exposure to real situations." That describes occupations, not individual candidates, but it is a useful frame for what to put in front of a reviewer.
What is overrated
The advice to avoid AI-exposed fields entirely is overrated. The Stanford canaries update found the declines concentrated in occupations where AI tends to automate the work, while "in occupations where AI is used more to complement workers, employment is flat or rising, particularly among more experienced workers." Exposure alone does not tell you which of those a field is. Walking away from a field because a chart moved is a bigger bet than most early-career candidates realize.
Volume is overrated too. When the entry tier is a small share of postings, sending two hundred identical applications mostly means competing in the same crowded slots two hundred times.
What to do this week
- Search by level on purpose. In Four-Leaf's AI job search, a level word like "junior" or "entry" in the query gives titles that carry junior, entry or associate a ranking boost, so they rise, without dropping postings whose title carries no level word. When the search names a company, that company's roles are pinned first and sorted by role match instead. Many real starter roles never put a level in the title.
- Lead with work you have done. A project, an internship deliverable or a shipped fix is evidence of practice. Our read is that evidence of practice travels further than a list of courses.
- Write for the level above. Describe your strongest piece of work in terms of the decision it informed or the problem it closed, the way a hiring manager would describe a mid-level hire's work.
- Cut the list, deepen each application. Pick fewer roles, read each posting closely, and rewrite your bullets in that posting's language rather than sending one resume everywhere. Four-Leaf's resume builder tailors a resume to a specific posting.
- Keep applying to exposed fields where you have proof. Entry-level roles are a smaller share of postings there, and still present.
Where this is heading
The canaries authors say plainly that no single study settles this, and their dashboard will update monthly. The trend could accelerate, stall or reverse. What 2026 has already shown is where the pressure lands. It lands on the first hire. In exposed fields the entry-level job is still there, in smaller numbers, and our bet is that it goes to the applicant who shows what they have already done.
Originally published on the Four-Leaf blog.
Top comments (0)