The Evidence That AI Is Killing Jobs Is Thin. The Entry Door Still Slammed Shut

What Happened

Two results out of the same university read like opposites.

The Stanford Institute for Economic Policy Research brief is the calm one. Written by Mahoney, McEntarfer, and Wahal, it concludes that all the evidence available so far suggests AI’s impact on current labor market conditions is small. The supporting facts are modest and stubborn. US Census data shows only one in five companies uses AI in any business function at all. MIT Technology Review’s roundup of the same evidence base adds more that cuts against the panic: occupations most exposed to AI have lower unemployment than less exposed ones, there is no sign of people moving en masse from threatened jobs to supposedly safer ones, and employment for coders keeps growing even though its growth rate slowed by about 3% after ChatGPT. McEntarfer, a former commissioner of the Bureau of Labor Statistics and an author of the brief, frames it as disruption that has not arrived yet, with time left to plan.

Then there is the paper from Stanford’s Digital Economy Lab that hits one precise spot. Brynjolfsson, Chandar, and Chen worked through individual-level payroll records from ADP, the largest payroll provider in the US, and found that workers aged 22 to 25 in the most AI-exposed occupations experienced a 13% relative decline in employment. The effect survives controls for firm-level shocks. It concentrates in roles where AI automates tasks rather than augmenting them. And the mechanism is not layoffs. It is a hiring collapse: the openings young people used to walk through simply stopped being posted. Unemployment for recent college graduates sits at 5.6% in early 2026, up 1.6 points from three years earlier.

What This Means for Founders

The two findings do not fight. The aggregate has not moved and the entrance has closed. That distinction matters because the entrance is exactly where founders meet this market.

In the short term it favors small teams. While large employers freeze graduate hiring, people who two years ago would have gone to a FAANG new-grad program or a well-funded Series C are still available. And what an early team can offer happens to be the scarce thing: owning something end to end. That is not purchasable with salary, and right now almost nobody else is selling it.

The bill comes with it. What AI absorbed was not whole jobs but entry tasks, the repetitive research, the first drafts, the ticket triage, the boilerplate a junior used to do while learning the shape of the work. Removing that work removes the path people used to climb. The industry has no answer for where senior engineers come from in three years, and every company skipping junior hiring today is deferring that invoice, not cancelling it.

There is also a measurement trap here. Challenger, Gray & Christmas counted 101,743 announced cuts citing AI in the first half of 2026, roughly 23% of the total. Academic measurement finds no matching move in the aggregate. That gap says something specific: companies are choosing AI as the stated reason. Writing “AI transformation” reads better to investors than “we over-hired” or “demand softened.” Anyone reading layoff announcements as a market signal is reading a press release, not data.

What You Can Do Now

Open one junior slot. The same budget buys a better hire than it did three years ago. Just do not fill that slot with work AI already does, or the person leaves in eight months. Write the posting around what a human will still be doing five years out, and lay out the order in which they will learn it. When you use layoff announcements as market intelligence, drop the stated reason and pair headcount changes with revenue instead. The reason is marketing copy. The headcount is a fact.