The first rung of the career ladder is disappearing, and it is not a projection. Using ADP payroll data covering millions of US workers, Brynjolfsson, Chandar and Chen found that early-career workers aged 22 to 25 in the occupations most exposed to AI have seen a relative employment decline of about 16%, while experienced workers in the same occupations, and young workers in less-exposed fields, held steady or grew. The pattern survives the obvious objections: it holds even excluding tech firms and remote-friendly occupations. And it concentrates precisely where AI automates work rather than augmenting it. The canary, as the authors put it, is already coughing.
The comfortable reading is that this is a hiring-cycle blip. The uncomfortable reading, and the better-supported one, is structural: the tasks companies used to buy from juniors, first drafts, basic research, routine analysis, standard code, are exactly what AI now does cheaply. When the junior's output is replaceable, the junior's job goes. What does not go, but quietly breaks, is everything that job secretly delivered: the apprenticeship through which juniors watched real decisions, made small recoverable mistakes, and became the seniors the firm assumed it would always have.
The ladder was doing two jobs
Entry-level work always had a dual function: production and formation. The production half is what shows up in headcount planning, and it is the half AI eats. The formation half never appeared in any budget line, which is why it is being destroyed by accident rather than by decision.
Here is the twist the pessimists miss: the same technology dismantling the old apprenticeship is unusually good at part of the new one. Brynjolfsson, Li and Raymond, studying 5,172 customer-support agents given a generative AI assistant, found average productivity up about 15%, but with the gains heavily concentrated among novices and lower-skilled workers, up to roughly 34%, while the most experienced barely moved. The mechanism matters more than the number: the AI had, in effect, absorbed the tacit patterns of the best performers and was transferring them to beginners. That is apprenticeship, mechanized, for the codifiable part of the craft.
So the situation is stranger than "AI kills entry-level jobs." AI removes the tasks juniors learned from, while simultaneously offering the fastest skill-transfer channel ever built. Whether a given firm's young people end up formed or stunted depends entirely on which half of that bargain it engineers.
What deliberate apprenticeship looks like
The scale of the reskilling demand is not in doubt; the WEF's Future of Jobs Report 2025 estimates that 59 of every 100 workers worldwide will need reskilling or upskilling by 2030. But the deeper design question is what juniors should now practice, given that task repetition no longer pays their way. The honest answer: the parts of seniority that were never in the tasks. Framing problems. Deciding what "good" looks like. Catching the plausible-but-wrong answer, a skill that becomes more valuable, not less, when machine output is fluent. None of these are learned by watching; all are learned by holding real, bounded responsibility early.
Which is where an older, pre-AI heresy becomes newly practical. At Semco, as Ricardo Semler has told it, factories were run by engineers without degrees, while MBAs from excellent schools could not run a meeting; one maintenance technician without a diploma led 200 people because everyone trusted him. Semler never looked at a diploma before hiring: trust, he said, does not come with a gown. Strip the provocation and a design principle remains: seniority is demonstrated judgment plus earned trust, and both can be grown in people the credential system would never have surfaced, if the organization gives them real decisions to make and colleagues who watch them make them.
That is the replacement ladder. Not "two years of grunt work, then responsibility," but responsibility from the start, scoped small, with the grunt work delegated to machines and the judgment deliberately practiced: juniors owning a customer, a process, a small P&L, with a senior human (not just a model) reviewing the decisions rather than the drafts.
The bill arrives in a decade
The trap in the current moment is that destroying the ladder is profitable this quarter and catastrophic in ten years, when the missing cohort fails to appear in the senior ranks, and every firm that skipped formation goes shopping for the same scarce mid-career talent at once. The 16% decline is not just a labor-market statistic; it is a forecast of that shortage, visible early.
Companies did not decide to end apprenticeship. They decided to automate tasks, and apprenticeship was living inside the tasks. The decision that remains is whether to rebuild it on purpose. The firms that do will spend more per junior and get seniors nobody else can hire. The firms that don't will discover, around 2035, that the entry ladder was also the only factory they had for the people at the top of it.
Sources
- Brynjolfsson, Chandar & Chen (2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI. Stanford Digital Economy Lab. ~16% relative employment decline for workers 22-25 in AI-exposed occupations; losses concentrate where AI automates rather than augments. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
- Brynjolfsson, Li & Raymond (2025). Generative AI at Work. Quarterly Journal of Economics. 5,172 support agents: ~15% average productivity gain, concentrated among novices (up to ~34%); AI diffuses top performers' practices to beginners. https://academic.oup.com/qje/article/140/2/889/7990658
- World Economic Forum (2025). Future of Jobs Report 2025. 59 of every 100 workers will need reskilling or upskilling by 2030 (employer survey, 1,000+ firms, 55 economies). https://www.weforum.org/publications/the-future-of-jobs-report-2025/