You’ve seen the ad: “Graduate Software Engineer”, two internships required, a production portfolio, AI tool fluency. You’ve probably noticed the junior postings thinning out while firms riding the AI hiring boom add headcount. In the UK, entry-level hiring fell 14% in the year to April 2026. Hiring and retraining decisions are being made on this signal. Is this part of something bigger?
Why are entry-level tech roles being “seniorised”, and what skills do employers now demand from junior hires?
“Seniorised” means entry-level roles keep their junior titles but carry materially higher expectations: portfolio work, AI tool fluency, internships, judgement that used to be expected years into a career. PwC’s Global AI Jobs Barometer found seniorised entry-level roles grew 35% since 2019 while other entry-level roles shrank 10%, and postings in AI-exposed roles are seven times likelier to demand human-intensive skills like judgement and communication.
More than a third of US entry-level jobs now require AI skills, nearly triple the autumn 2025 share. AI is removing the routine work that once acted as an apprenticeship: the small bug fixes and reviewed pull requests that taught juniors the trade are the first work it absorbs, so new hires are expected production-ready from day one. PwC adds that new tasks in AI-exposed roles are 2.5 times likelier to rely on judgement, communication and customer context, skills that gain value as routine work shrinks.
In the UK, entry-level hiring was down 14% year-over-year by April 2026, with software engineering entry postings down 27%. The same snapshot shows near-zero overlap between the skills employers need and those candidates offer, sharp enough that how the UK market is bucking the trend deserves its own article.
Net effect: the entry-level mix is being re-weighted rather than shut down, the same workforce recomposition story. The open question is the pipeline: if the rungs keep getting further apart, who becomes the senior engineers of 2035?
Why do young workers in the most AI-exposed occupations see declines while AI-adopting firms add headcount?
The two streams look contradictory. Both can be true because they measure different things: firms versus occupations, entrants versus incumbents. Stanford’s payroll data shows US employment for 22 to 25-year-olds in the two most AI-exposed quintiles fell about 11% between November 2022 and June 2026 while the least-exposed grew about 10%, and Ramp and Revelio Labs found entry-level headcount at the largest US AI investors up 12%.
Stanford’s Canaries in the Coal Mine, built on about 25 million ADP payroll records, puts the August 2026 gap at 19% below the counterfactual. The age gradient runs on knowledge: AI overlaps with the codified kind graduates hold, less so the tacit kind built by practice.
The primary Ramp paper is hard to retrieve and its platform sample skews toward tech firms, so treat the finding as reported.
The streams do not contradict each other; they answer different questions. Adoption-linked growth among adopters and exposure-linked decline across occupations can both be true, and the Dallas Fed finds the declines flow through reduced hiring, not layoffs.
Anthropic’s task-level research puts computer programmers at the top of observed AI exposure, with AI touching about 75% of tasks. Displacement concentrates where AI completes tasks end-to-end rather than assisting people, which is why customer-facing, regulated, human-intensive work holds up.
David Deming notes the junior decline began about six months before ChatGPT released, keeping the remote-work hypothesis alive. The macro and tax confounders muddy the attribution: the ZIRP unwind, Section 174’s R&D tax change, and one tracker of 2025 layoff announcements counted only 4.5% of them attributed to AI.
The firm-level side is covered in why AI-adopting firms are adding headcount; the payroll-versus-postings discipline lives in what the layoff numbers can and cannot show. All of these are US series; the UK 14% figure is separate.
How should engineering leaders weigh junior hiring decisions against the seniorised entry-level trend?
Weigh it as a pipeline investment with a delayed return. Under-hiring juniors erodes the apprenticeship: no scar tissue, a hollowed-out career ladder, a missing senior bench by the early 2030s. Over-hiring into task sets AI now automates misallocates budget. Four factors decide it: pipeline economics over time, skill mix, mentorship capacity, onboarding throughput.
One widely shared practitioner argument is to cost junior hiring as R&D: cutting the pipeline saves money now and defers the bill to the future bench. The under-hiring risk is losing the judgement that only safe failure builds; skip the juniors and that scar tissue stays concentrated in the few engineers who already have it. Track how many engineers can debug each key system the way you track uptime.
The over-hiring risk is real too: BCG finds entry-level positions in divergent roles, where AI substitutes for tasks, are exposed first, so hiring juniors into automating task sets spends budget on shrinking work.
Leading firms run the full range. IBM is tripling entry-level hiring around customer contact; Salesforce hired zero engineers in the fiscal year; OpenAI is trialling a pairing of senior and junior engineers at the extremes. Microsoft is proposing a medical-style preceptorship, with senior engineers formally responsible for juniors’ judgement. And 54% of engineering leaders expect AI tooling to reduce junior hiring, even as adopters keep adding headcount.
Stress-test it: model the 2035 team assuming no junior hiring. If AI progress plateaus and attrition continues, the crunch arrives on schedule, unless new entry points built on different abstractions form first.
How should leaders frame retraining and upskilling decisions when firm-level evidence is still thin?
Plainly: firm-level causal evidence on retraining outcomes is thin, and announcements are not measured outcomes. Reskilling into adjacent roles with AI leverage and investing in the human-intensive skills AI cannot supply (judgement, communication, customer context) are the better-supported moves.
The programme landscape (preceptorship, structured apprenticeships, skills-based progression) is reported as advocacy and forecasts, with no measured results on retention, promotion speed or code quality, so each remains a hypothesis until results arrive.
The evidence supports a direction more than a recipe: PwC’s Global AI Jobs Barometer shows professionalised roles, where AI emphasises human judgement, growing twice as fast as democratised ones, with 42% faster wage growth since 2021. Reskilling toward the roles actually growing in this market is the nearest thing to a supported bet.
Cost it as one of three moves: upskill inside the current role, reskill toward an adjacent one, or replace, and watch your organisation’s lead indicators, such as internal mobility.
If you are weighing these numbers from Australia: every headline figure is US or UK, no local benchmark has surfaced, so read direction, not magnitude. Keep a watch list: the Stanford dashboard revisions, the Dallas Fed series, the PwC barometer, Anthropic releases, assessed through the credibility lens.
The entry level is being repriced and re-routed rather than removed. Seniorisation raised the price of entry; displacement is real but concentrated in routine, automatable tasks; both sit inside a recomposing workforce. New entry points are forming, and the apprenticeship must be rebuilt deliberately. Which leaves the question this article kept circling: who becomes the senior engineers of 2035? The full series overview connects the threads.
Frequently Asked Questions
Are entry-level tech jobs actually disappearing, or is this a repricing?
The evidence points to a repricing rather than a closure. PwC’s Global AI Jobs Barometer found seniorised entry-level roles grew 35% since 2019 while other entry-level roles declined 10%, and the pattern sits inside a broader workforce recomposition. The honest caveat is that nobody can predict how many entry points will exist in five years, so the useful frame is the changing mix, not a shutdown.
Why do the Ramp/Revelio and Stanford studies disagree about AI’s effect on employment?
They ask different questions of different samples, so the disagreement is mostly about scope. Ramp/Revelio’s firm-level study of more than 21,000 US firms found entry-level headcount grew 12% at the largest AI investors after adoption. Stanford’s occupation-level payroll data found US employment for 22 to 25-year-olds in the most AI-exposed occupations fell about 11% between November 2022 and June 2026. Adoption-linked growth among adopters and exposure-linked decline across occupations can both be true.
Are young workers really being replaced by AI, or is something else happening?
So far, the pattern looks more like reduced entry than replacement. The Dallas Fed found the declines flow through lower hiring rather than layoffs or separations, and Stanford’s exposure-linked gap is concentrated in occupations where AI automates tasks end-to-end, not where it assists workers. Whether that narrows into outright replacement depends on how far task automation progresses, which nobody can currently forecast.
The junior hiring decline started before ChatGPT. How much can we really blame AI?
AI is one factor among several, and attribution remains genuinely contested. The junior decline predates ChatGPT, which keeps the remote-work hypothesis alive (training juniors from a distance is harder), and macro and tax factors, including the ZIRP unwind and Section 174, a US R&D tax change, confound simple readings. Only about 4.5% of 2025’s announced layoffs were explicitly attributed to AI, so treat any single-cause story as too tidy.
Is the squeeze limited to software engineering, or does it affect other roles too?
It is broader than software and uneven within it. Anthropic’s task-level research put computer programmers at the top of observed AI exposure, at about 75% of tasks, but the Stanford payroll decline covers 22 to 25-year-olds across the most AI-exposed occupations. In the UK, entry-level hiring was down 14% year-over-year by April 2026, with software engineering postings down 27%, while PwC finds professionalised roles are growing twice as fast.
How do I know which roles in my team count as AI-exposed?
Exposure is a task-level measure, not a job-title label. The working method is to ask how much of a role’s work AI can already complete end-to-end, the lens Anthropic used when it found computer programmers the most exposed occupation at around 75% task coverage. Roles heavy in judgement, communication and customer context score lower on exposure, which is one reason employers keep demanding those skills.
The Stanford data tracks 22 to 25-year-olds. Does that mean older career-changers are safe?
Not necessarily, because the cohort is age-based and age is a proxy for career stage, not a guarantee. The mechanism the Stanford team describes is that AI overlaps with codified knowledge, the textbook-style knowledge new entrants hold, while tacit knowledge built through experience is harder to automate. Career-changers bring more of that tacit knowledge, so the pattern may differ for them, but nobody has clean evidence either way yet.
How do I tell a credible AI-and-jobs study from a noisy headline?
Check four things before acting on a number: the unit of analysis (firms or occupations), the sample and its source, the geography, and the date. A payroll study covering about 25 million US workers answers a different question from a firm-level adoption study, and a UK postings figure is not a US one. Once those labels are attached, most apparent contradictions dissolve into scope differences.
What is workforce recomposition, and what is replacing the traditional entry-level rung?
Recomposition means the mix of entry-level work is being re-weighted, not deleted. Some routine tasks are absorbed by AI while new entry points form around AI-leveraged building, deliberately rebuilt apprenticeships and human-intensive work. The “builder wave” is the clearest example, with AI-assisted builders shipping software, sometimes without the developer title, though these routes are early and uneven and not yet a substitute for the old junior rung at scale.
If AI absorbs the training tasks, how should we develop junior engineers instead?
The honest starting point is that no replacement model has published measured outcomes; preceptorship, pairing and structured apprenticeships are experiments, not templates. The common thread across them is putting juniors beside seniors on real work, creating safe opportunities for failure that build the scar tissue AI cannot supply, and measuring what matters, such as retention, promotion velocity and code quality. Until those results arrive, treat your design as a hypothesis under test.
Does this apply to smaller companies, or only to Big Tech?
The trade-offs apply at any size, but the inputs differ. The programme experiments that get quoted (IBM, Salesforce, OpenAI, Microsoft) skew large, and none has published measured outcomes even for them; what transfers to a 50 to 500-person company is the decision frame: pipeline economics over time, skill mix, mentorship capacity and onboarding throughput. For smaller firms, mentorship bandwidth is usually the binding constraint, which changes the calculus rather than the framework.
Does any of this apply to Australia, or is all the evidence from the US and UK?
Every headline figure is US or UK; no Australian benchmark surfaced in the research reviewed here. That leaves Australian leaders reading direction rather than magnitude: exposure-linked declines and seniorised postings are plausible here, but the local numbers remain unknown, and nobody should import the UK’s 14% decline as if it were ours. It is an open question, and one worth watching local graduate and tech hiring data for.
