The class of 2026 walked off the graduation stage into a job market that keeps producing two contradictory headlines, and both come with charts. One says AI has quietly deleted the bottom rung of the career ladder. The other says the data does not actually show that. The uncomfortable part is that neither side is lying.
The bad half of the ledger
Start with the numbers that scare people, because they are real. Entry-level job postings in the U.S. are down roughly 35% since early 2023, per Washington Monthly, and the tasks that used to fill a junior’s first year — data entry, basic coding, first-draft admin, the grunt work nobody enjoyed but everybody learned from — are exactly the tasks generative AI does for free. Unemployment for recent grads aged 22 to 27 sits around 5.6%, meaningfully above the 4.3% national rate, in a reversal of the usual pattern where a fresh degree buys you better odds, not worse. And per the New York Fed, about 43% of young grads are underemployed, working jobs that never required the degree they are still paying for.
The anecdotes rhyme with the data. Stanford computer science graduates — a group that three years ago could auction themselves to the highest bidder — are now struggling to land entry-level roles, which feels less like a soft market and more like a structural rug-pull. If the safest degree in the safest field is wobbling, the intuitive story writes itself: AI came for the interns first.
The Fed reads the same room and shrugs
Then there is the other pile of paper. In March, the Federal Reserve published a FEDS Note titled AI Adoption and Firms’ Job-Posting Behavior, built on data from more than a million firms. Its finding, in the driest possible language, was a set of “precisely-estimated null effects.” Translated: they looked hard for AI’s fingerprints on hiring and could not find them. The Yale Budget Lab, tracking the same question month after month, keeps landing in the same place — no meaningful aggregate connection between AI usage and employment or unemployment. Fortune summarized the Budget Lab’s earlier work with the phrase everyone in HR quietly fears: if AI were truly roiling the market, the data isn’t showing it, which raises the possibility of “AI-washing” — companies dressing up ordinary cost-cutting in a fashionable excuse.
And there are counter-numbers even on the hiring side. NACE says employer demand for the class of 2026 is up 5.6%. Internship postings on ZipRecruiter are up 32% year over year. Healthcare, cybersecurity, and the skilled trades are hiring new grads as fast as they can find them. That is not the shape of an extinction event.
So which chart is lying
Neither, which is the least satisfying answer and the correct one. The Fed’s null effect is an aggregate — it averages the whole economy, and averages are very good at hiding a small, concentrated wound. Both the Budget Lab and the Fed note the same caveat when pressed: the pressure that does show up is clustered in entry-level segments of the most AI-exposed occupations, among the youngest workers. In other words, “no effect across the economy” and “a real effect on 22-year-olds writing their first lines of code” can both be true at once, because the second group is a rounding error in the first.
What that means for anyone about to graduate is annoyingly specific rather than apocalyptic. The ladder didn’t get sawn in half; its first rung got narrower and moved. The work that AI hands back for free is precisely the work that used to justify hiring someone with no experience, so the on-ramp now demands you show up already merged onto the highway — internships, AI fluency, a portfolio, a network. That is a real and unfair shift in the terms. It is just not the same thing as “the jobs are gone,” and the difference matters, because the fixes for a narrower on-ramp and the fixes for a vanished profession are not the same fixes. Anyone selling you the cleaner, scarier version of this story is, at minimum, not reading the Fed’s footnotes.
This is a developing labor-market story and the data is genuinely contested; treat any single number — including the reassuring ones — as a snapshot, not a verdict.
