Skip to content
Fritz Gerald ZephirinMenu

Social

The Entry-Level Problem

4 min read

A Stanford study published in 2025 found something that should worry anyone who plans to be alive in 2035: since late 2022, employment for workers aged 22 to 25 in AI-exposed occupations — software engineering, customer service — has declined about 13 percent relative to less-exposed peers. Older workers in the same occupations were largely fine. The damage is concentrated precisely at the entry point.

This is not a temporary hiring freeze. It is a structural change in what firms buy, and it has a flaw at its center that almost nobody owns.

Here is the mechanism. When a company hires a junior employee, it buys output — tickets resolved, code written, briefs drafted. But society gets something else as a by-product: skill formation. The junior person does a thousand low-stakes reps under supervision, makes survivable mistakes, and slowly converts feedback into judgment. Nobody pays for the judgment directly. It rides along with the output, a subsidy hidden inside a salary.

AI just unbundled that package. A model supplies the junior-grade output — a quarter of new code at Google, the majority of tier-one support tickets — without the apprenticeship attached. So firms, behaving rationally one quarter at a time, stop buying junior time. Amazon cut 14,000 corporate roles citing AI; Salesforce took support from 9,000 people to about 5,000; by mid-2026, over half of tech layoff events explicitly cited AI. Each decision is locally sensible. Collectively, they withdraw the subsidy that manufactures the next generation of senior people.

And senior judgment is not optional. It is the appreciating asset of the whole AI economy — the thing that reviews the model’s output, sets the bar, signs. Every company currently bidding up scarce senior engineers while declining to hire juniors is consuming a stock it has stopped producing. The economy is eating its seed corn and calling it efficiency.

We’ve seen smaller versions of this before. Typing pools disappeared and took a whole entry ramp into office work with them — secretarial career ladders simply ended, and nothing replaced them. Law clerkships and medical residency survive precisely because those professions institutionalized apprenticeship instead of leaving it as a by-product — they made someone own it. The professions that left skill formation implicit are the ones where AI is now quietly demolishing it. Even the informal learning commons are going: Stack Overflow — where two generations of programmers learned in public — lost double-digit traffic within months of ChatGPT and never recovered.

I should argue the other side, because it’s strong. The best field evidence we have says AI helps novices most: in the landmark support study, AI assistance raised novice productivity 34 percent while barely moving experts. A motivated 23-year-old with a frontier model has the best tutor in history. Maybe AI-native juniors will form judgment faster — compressed apprenticeships, always-on feedback. I genuinely hope so, and if longitudinal data shows it, this essay gets revised. But tutoring is not employment. Judgment forms under stakes — real customers, real deadlines, real consequences — and stakes are exactly what the entry-level market is no longer offering. A flight simulator helps; nobody becomes a captain in one.

I run the kind of operation this is happening to. Customer support has been a first formal job — often the first formal job — for millions of people, including a large share of the global workforce my companies hire from. An AI-first support floor with no junior tier isn’t just a cost structure; it’s a missing rung. Which is why I’ve come to think the answer is not nostalgia but ownership: apprenticeship has to become deliberate now that it is no longer free. In practice, in our own operations, that means keeping a junior tier whose explicit job description includes learning — reviewing AI-handled conversations, owning escalations under supervision, graduating into the judgment seats. It costs real money. It used to be free. That’s the change.

The competitive claim, stated plainly: judgment formation is becoming a capital asset that firms must build instead of harvest. The companies — and countries — that build the new apprenticeship will own the senior talent of 2035. Everyone else will be bidding for it.

Sources & further reading