AI Didn't Kill Junior Jobs. It Ended the Hidden Apprenticeship Subsidy

Workers aged 22 to 25 in the most AI-exposed occupations now run roughly 19 percent behind peers in less-exposed fields, according to Stanford's Digital Economy Lab. That single figure has fueled panic that AI is destroying entry-level white-collar work and that college degrees are no longer worth it. A former Federal Reserve official with a decade at the central bank says the panic misreads both the timing and the economics of what is happening.

Consider the timing first. When the FOMC began raising rates in March 2022, job postings for the occupations most exposed to AI peaked at the same time and began to fall. ChatGPT would not exist for another eight months.

Two researchers, Zanna Iscenko and Fabien Curto Millet, analyzed 238 million job postings and tied the posting decline to the tightening cycle. They noted that AI-exposed occupations tend to cluster in information, finance, and professional services — sectors especially sensitive to interest rates. The Economic Policy Institute adds a complicating data point: young workers without college degrees, whose occupations score negative on AI exposure, also saw their unemployment rate rise at a similar pace over the same period. If AI is the culprit, unexposed workers should have been spared. They were not.

The Stanford researchers themselves are more careful than the headlines. Their own paper describes the findings as "descriptive patterns, not causal estimates" and states they "do not see widespread, economy-wide job displacement associated with AI." They have also pushed back on the monetary-policy explanation, noting that the most AI-exposed jobs are not generally the most rate-sensitive, and that the employment gap for young workers in exposed occupations keeps widening even as rates have come down.

The honest answer is that nobody yet knows what happens to the labor market as AI spreads, because it is impossible to separate overlapping shocks in real time. But businesses still have to make expensive decisions about hiring, education, and regulation right now, as if they knew the answer.

One pattern in the data is clear, though. Firms are not firing their junior employees; they are hiring fewer of them. The decline concentrates where AI automates work rather than where it complements it — a fact that may itself answer the Stanford team's rebuttal.

Ask what firms were actually buying all those years when they hired juniors. Not output. A first-year associate's document was checked by a partner and frequently redone. The patient history a resident doctor took down at 2 AM often had to be retaken by the attending physician. By any honest accounting, that work was unproductive. Firms bought it anyway because that is how you turn a junior employee into a senior partner. The output was the byproduct; the formation of the employee was the point, and the work helped offset the cost.

Now AI can do the work of the junior analyst. Firms invest less in these hires, and the pipeline that turns juniors into seniors thins. Matt Beane, who documented this mechanism in his research, watched it in operating rooms years before ChatGPT, as surgical robots quietly cost residents the case time that made them surgeons. Employers still want experienced people. Many have simply stopped funding the process that produces experience.

The prescription for employers is to stop treating junior hiring as a cost line that automation just erased. It was never an operating expense; it was a capital investment mislabeled — the mechanism by which the firm manufactured its own future partners. AI may have taken some of the production value juniors created, but the value of training remains, and it must now be funded more deliberately: rethought training, redesigned job rotations, and real mentoring aimed at building the judgment of tomorrow's senior professionals.

Universities face the same problem from the other end. The Ph.D. is an apprenticeship funded by the productive value of apprentices' work, and the editors of Nature warned this spring that early-career researchers now face the danger that "tasks that are crucial to their training as scientists are done by a machine." Classroom evidence points the same way: students learn when the tool is constrained so effort cannot be skipped, and fail to learn when it hands over answers. As the author's old Texan boss put it: "no friction, no traction." Forming judgment requires friction — tough problems, failure, repetition. Getting the right answer was never the point. "Show your work," as the math teachers said.

There is a version of this transition in which AI does the routine work and an entire generation never gets the reps that turn talent into judgment. Nothing in the technology makes that outcome inevitable. It arrives only if employers keep booking formation as a cuttable cost and universities keep certifying work the machine did.

The postings data will recover when the hiring cycle turns; it always does. What will not recover on its own is the old bargain in which production quietly paid for formation. Rebuilding that bargain, deliberately and on someone's budget, is the real AI question facing firms and universities now.