This is the condensed edition of the deep dive of the same name. The deep dive carries the full argument chain, every source, and evidence grading; this version keeps the main line in plain words. Key figures were independently verified (data through June 2026).
The profession isn't disappearing; the entrance is narrowing. Total US software developer employment hit an all-time high in 2025 — but payroll employment of 22-25-year-old developers has fallen nearly 20% since late 2022, and it's still falling. Half of "junior engineers are disappearing" is true — and the true half is more specific, and more worth taking seriously, than the rumor version.
Most of the shouting on this topic happens because people are reading different dashboards. Same job market, four gauges — different directions, all real:
Put the four gauges together and you get the opening sentence: the profession is fine; the door got narrower. Especially the door for the youngest.
One detail says a lot: computer science majors' unemployment rate is 7.0% — second-to-fourth worst of all college majors. But among CS grads who do find work, almost everyone gets a real degree-level job — only 19% end up in jobs that don't need the degree, nearly the lowest of any major (the average is about 40%). The door is hard to get through. The room behind it hasn't gotten worse.
Here's the strangest part of the whole story. You'd think the case for "AI replaces juniors" is that AI does junior work better than juniors. The experiments say the opposite.
The largest randomized experiment on AI coding (4,867 engineers at Microsoft, Accenture, and another large firm — half randomly given AI, half not, the same method medicine uses to test new drugs): people with AI completed 26% more tasks on average, and the shortest-tenured people gained the most — 27-39% for newer engineers versus 8-13% for veterans. A similarly rigorous study of customer service agents: AI boosted the least experienced workers by about 30% and did nothing for people past their first year.
So the question: if AI helps novices most, why are novice jobs the first to shrink?
The answer isn't in the technology — it's in the company's ledger. Experiments measure how much faster an already-hired person works with AI. Companies decide whether to hire one more person at all. AI raised the floor of what a beginner can produce — and at the same time lowered the cost of not hiring one, because a senior engineer with AI can now casually absorb the work that used to be a junior's training ground. The technology that helps beginners most is exactly what makes not-hiring-beginners feasible. Those two statements aren't a contradiction; they're two sides of one sentence.
The economists who ran the customer-service study said it themselves at the end of the paper: firms can respond to more productive novices "by hiring more of them or by seeking to develop more powerful AI systems that replace labor altogether." The experiment can't tell you which one firms will pick. The hiring data since 2023 says American software companies — for now — picked the second.
The honest answer: AI is at most an accomplice, and not yet convicted.
The defense has strong evidence. Half of the tech-hiring collapse happened before ChatGPT even launched — the culprits were 2022's interest-rate hikes, the payback for pandemic over-hiring, plus a peculiar US tax rule (Section 174, which forced companies to spread software salary deductions over five years — effectively a tax hike, estimated to have killed about 20,000 jobs). Yale's Budget Lab keeps re-testing and keeps finding no detectable AI disruption in the overall labor market. And there's a quieter suspect that's hardest to rebut: offshoring. While US junior hiring froze, India's tech centers have been expanding early-career hiring 18% a year. Junior jobs didn't vanish — some of them moved.
But the prosecution holds one card the defense can't answer: none of those forces care how old you are. Rate hikes, tax law, and layoffs hit whole companies — yet inside the very same companies, the only group falling is the youngest one in the most AI-exposed roles, while older engineers kept growing. And when that tax rule was repealed in July 2025, total hiring bounced back — but the junior share kept shrinking. Macro forces explain how far the water level fell. They don't explain why the water tilted.
So the fairest verdict today: the macro cocktail caused the collapse; AI is deciding who it lands on. How much blame AI ultimately takes is still being fought over — the best researchers in the field say this research is "still in the first inning."
The first half of 2025 was the proclamation era: Anthropic's CEO warned AI could wipe out half of entry-level white-collar jobs within a few years; Meta said AI would code like a mid-level engineer within the year; Salesforce announced it would hire no new engineers.
2026 is the boomerang era, and it's been humbling: 32% of hiring managers who cut a job because of AI have since re-hired the same role (Robert Half survey); 55% of executives who did AI layoffs now say it was a mistake (Orgvue survey); IBM announced it would triple entry-level hiring, with its HR chief spelling out the logic — companies that don't hire juniors now will be paying top dollar to poach everyone else's mid-level people in three to five years. The CEO of Amazon Web Services put it bluntest: replacing junior employees with AI is "one of the dumbest things I've ever heard" — juniors are the cheapest employees and the best AI users — and Amazon is hiring 11,000 interns and new grads this year anyway.
One piece of history is worth reading before believing any forecast. In 2016, AI pioneer Geoffrey Hinton said people should stop training radiologists, because AI would surpass them within five years. Ten years later, radiology residency slots hit an all-time record in 2026, average pay reached $571,000 — and Hinton admitted he got it wrong. But the story has a dark tail: even though the prediction failed, applications to radiology still fell 14% from their peak. A scary prediction doesn't have to come true to scare away a generation. Computer science enrollment just started dropping (new majors down 13% in 2025). The same script may be running again.
Bigger than "will juniors be replaced" is a question nobody is measuring: if beginners don't do the work, how do beginners become experts?
Engineers grow by doing — that's not a platitude, it's Nobel-grade economics (Arrow, 1962): skill forms only in the act of solving real problems. Two paths now both cut it off: companies don't hire beginners, so they get no work — or companies hire them and AI does all the formative work anyway. Education research has produced the first rigorous warning: high-schoolers given unrestricted GPT-4 for math practice scored 48% better while using it — and 17% worse than students who never had it, once it was taken away. But in the same study, a version with guardrails (an AI tutor that coached instead of answering) erased almost all the damage. Whether AI is poison or protein for growth depends on how it's used, not whether it's used.
And the most telling fact in the industry: everyone is arguing about whether the career ladder is broken, yet no company anywhere publishes whether junior-to-senior promotion time has gotten longer or shorter. The first one that starts measuring the ladder will be worth more than every opinion piece combined.
Students (or anyone deciding whether to study CS): don't make life decisions off headlines — the radiology prophecy went unfulfilled for a decade, and the applicants who fled simply handed their spots to those who stayed; CS enrollment is already shrinking, so competition may well be looser by the time you graduate. What actually helps is preparing for the new bar: entry jobs now want a "small senior" — real projects, internships, and the ability to read someone else's code and spot what's wrong. And one warning straight from a rigorous experiment: set rules for how you use AI while learning — for anything new, take your own attempt first (even just guessing what AI will output) before it reveals the answer, rather than asking for the full answer up front. People who copy answers end up worse than people who never used AI at all, once it's taken away.
New grads on the market: get the framing right — you hit the worst timing (record graduating classes × the narrowest entrance), not a verdict on your ability, and the wind is turning (postings recovering, companies re-hiring roles they cut). Tactically: don't fixate on Big Tech — that's exactly where the new-grad share collapsed hardest; entry opportunities hide in campus pipelines, intern conversions, referrals, and software roles at non-tech companies (explicit "junior" titles are only ~2% of public postings). In interviews, demonstrate that you can direct AI and catch its mistakes — that's closer to what employers actually lack than whiteboard algorithms.
Employers: skipping juniors saves money this year and overdraws the next three to five — by then you'll be paying premiums to poach someone else's mid-levels, or finding none to poach; a third of managers who cut roles for AI are already re-hiring them. The smarter move isn't deleting the entry role but redesigning it: make juniors the first line of review for AI output, with mentorship and the same quality gates — experiments show novices brought up this way reach six-month performance in two months, and the skills stick. And start measuring your career ladder (how long until a new hire owns something independently?) — nobody in the industry measures this today, and the first company that does will out-decide everyone else's slogans.
The whole argument reduces to seven claims, ordered from hardest to softest evidence:
Worth watching through 2027: whether the 26–30 bracket starts showing the same decline; whether junior/senior job postings diverge; whether IBM-style "hire juniors back" reversals go mainstream; whether any company starts publishing "years from junior to senior."