Structured from 5 source documents across the AI impact and market landscape research lenses, with corroborating evidence from the historical precedent lens. Source-reviewed, fact-reviewed, and gap-reviewed before publication.
The junior developer hired in 2025 would be a mid-level engineer by 2030. Five to seven years of production experience — debugging real systems, learning a codebase’s scar tissue, building the judgment that comes only from shipping code and watching it break. That mid-level engineer would be approaching senior status right as today’s seniors begin retiring in larger numbers.
The industry is not hiring that junior. It is going to need that mid-level engineer.
This is not a thought experiment. It is a structural consequence of decisions being made right now, at scale, across the profession. And the evidence — from Harvard payroll analysis to Indeed hiring data to university enrollment figures to bootcamp closure announcements — converges on a finding that should concern anyone with a stake in the long-term health of the software industry: the entry-level pipeline is being dismantled faster than anyone is building a replacement.
The numbers
The evidence for the junior hiring collapse comes from multiple independent sources, each measuring a different dimension of the same contraction.
The strongest academic evidence is a Harvard study tracking 62 million workers across 285,000 U.S. firms from 2015 to 2025. The finding: junior employment at AI-adopting companies declined 9–10% within six quarters of implementation. Senior employment was virtually unchanged. The mechanism was not layoffs — it was a hiring freeze. After late 2022, AI-adopting firms hired five fewer junior workers per quarter. Graduates from mid-tier institutions were the worst affected.
The Stanford Digital Economy Lab corroborated the pattern independently. Analyzing ADP payroll records covering millions of workers, they found employment for software developers aged 22–25 declined nearly 20% from its late-2022 peak. Workers 35–49 in identical occupations at the same firms saw stable or growing employment.
Both studies controlled for firm-level shocks — which would capture most cyclical effects from the post-ZIRP correction. They still found significant AI-adoption effects on junior employment. That is evidence for a structural component, not just a hiring cycle working itself out.
The hiring data tells the same story from the employer side. Indeed Hiring Lab found junior-level tech postings down 34% from five years earlier, compared to 19% for senior-level titles — entry-level positions hit nearly twice as hard. SignalFire VC data, reported via Stack Overflow, found that fresh graduate hiring share at major tech companies collapsed from 32% to 7%. A Rezi analysis documented a 67% decrease in U.S. entry-level tech postings between 2023 and 2024.
The magnitudes vary by source and methodology — Rezi’s 67% versus Indeed’s 34% reflects partly different baselines and partly the difficulty of separating cyclical noise from structural signal. Indeed’s own researchers note that nearly half of the net decline in tech postings occurred before ChatGPT’s public release in late 2022, implicating the post-ZIRP macro correction and 2020–2022 overhiring as significant drivers alongside AI. The most defensible reading: both forces are operating. Cyclical correction compressed junior hiring, and AI adoption is amplifying and extending that compression in ways that look structural.
One number crystallizes the situation: computer science graduate unemployment now sits at 6–7% — nearly a full percentage point higher than liberal arts graduates and well above the 4.3% national average. A CS degree, for the first time in a generation, is not a guaranteed path to employment.
The pipeline is responding
The educational infrastructure that feeds the profession is contracting in real time, responding to market signals with its own withdrawal.
For the first time since the dot-com bust in the early 2000s, undergraduate computer science enrollment across the University of California system declined in 2025 — down 6% from 2024 and 9% over two years. Nationally, 62% of computing academic units reported declining enrollment for 2025–2026, according to the Computing Research Association’s CERP Pulse Survey. Graduate CS programs saw a 15% enrollment drop. Sixty-four percent of pessimistic CS majors cite generative AI as a factor in their concerns about the field.
Students are not abandoning technology entirely — they are migrating to specialized programs. Computer engineering, cybersecurity, and AI-specific degrees are seeing growth. USC, Columbia, Pace, and New Mexico State are launching dedicated AI degrees. Berkeley’s Data Science major continues to grow even as core CS applications plateau. The signal is not that young people have lost interest in technology. It is that they can read a hiring market.
The bootcamp industry — the alternative pipeline that opened software careers to career changers and non-traditional students — is in severe contraction. 2U shut down all university-partnership bootcamp programs in December 2024, with CEO acknowledging that “the long-form, intensive training that boot camps provide no longer aligns with what the market wants and needs.” 2U’s enrollment had already dropped 40%. Epicodus closed in early 2024 after enrollment fell amid tech layoffs. Turing School paused new enrollments as of April 2025, after training over 2,500 alumni across eleven years. SNHU shut its coding bootcamp in 2023 citing low-cost competition and AI adoption.
The internship pipeline — historically the primary on-ramp to developer careers — is contracting just as sharply. Handshake data shows technology internship postings down 30% between January 2023 and January 2025. Meanwhile, competition has intensified to absurd levels: average applications per tech internship rose from 43 in 2022–2023 to 273 in 2024–2025. Two hundred and seventy-three applications for a single internship position.
And even when entry-level roles are posted, the definition of “entry-level” has inflated beyond recognition. Sixty percent of “entry-level” tech jobs now require three or more years of experience. The door is labeled “entry” but the lock requires credentials that entry-level candidates do not possess.
The missing rung
The hiring collapse and the educational contraction are the visible symptoms. The structural problem runs deeper.
AI coding assistants are automating exactly the tasks that historically served as the training ground for new developers: boilerplate generation, basic CRUD operations, simple debugging, test writing, documentation. These were never just grunt work. They were the mechanism through which juniors built foundational understanding — of codebases, of systems, of what happens when your code meets production. A junior who spent six months writing boilerplate learned where the bodies were buried in the codebase. A junior who spent a year fixing small bugs learned how systems actually fail. Remove those tasks, and you remove the apprenticeship.
An Anthropic-published randomized controlled trial — 52 junior developers — measured this directly. Developers using AI assistance scored 17 percentage points lower on coding comprehension assessments than those working without it. Fifty percent versus 67% — equivalent to nearly two letter grades. The largest gap appeared in debugging and code comprehension. Developers who delegated code generation to AI scored below 40%. Those who used AI for conceptual inquiry — asking questions, requesting explanations — scored 65% or higher.
The distinction matters enormously. The problem is not that juniors are using AI. It is how they are using it. The Anthropic researchers found that the difference between skill development and skill erosion was whether the developer engaged with the generated code or simply accepted it. High-scoring participants used AI to generate code and then asked conceptual questions, requested explanations, and resolved errors independently. Low-scoring participants, in the researchers’ words, “wholly relied on AI to write code.”
This creates a vicious cycle that the industry has not yet reckoned with. Companies reduce junior hiring because they believe AI can handle junior-level work. The juniors who are hired develop weaker skills because AI handles the tasks they would have learned on. And the code that AI generates is not production-ready — CodeRabbit’s analysis of 470 open-source pull requests found AI-authored code carried nearly twice the issue density of human-written code, while Veracode’s 2025 security audit found 45% of AI-generated code introduces vulnerabilities including critical OWASP Top 10 flaws. The industry is simultaneously shrinking the pool of people trained to catch these problems and increasing the volume of code that needs catching. Fewer juniors, weaker-skilled juniors, and higher skill requirements at entry. The ladder is not broken. The first rung has been removed.
The talent vacuum
If junior hiring remains suppressed for three to five years, the arithmetic is straightforward. The mid-level engineers that companies need in 2030 should be junior developers now. The senior engineers that companies need in 2035 should be gaining their first production experience today. Every year the pipeline stays suppressed adds another year to the downstream talent shortage.
This is not a novel pattern. In “Sixteen Funerals,” I traced entry-level elimination as the single most universal finding across 150 years of workforce disruption. The longshoremen’s hiring register was frozen for thirteen years — an entire generation excluded from the profession. IT offshoring moved entry-level coding jobs offshore first. In photography, minilab positions disappeared entirely. In the COVID correction, entry-level tech postings dropped roughly 67%. The mechanism is always the same: entry-level positions sit closest to the commodity layer, so they are automated or eliminated first. And their elimination removes the career ladder that produces the experienced workers the industry needs later.
The 54% of engineering leaders who told LeadDev they expect permanent junior hiring declines are making a prediction about the next two to three years. They may be right about the near term. The question is whether they have considered the five-to-ten-year consequence of that prediction. A profession that stops training its replacements does not stay staffed with experienced practitioners forever. It runs down the existing supply.
BLS still projects 15% software developer employment growth through 2034 — roughly 129,000 annual openings. That demand does not disappear because the entry pipeline has constricted. It shifts upstream, intensifying competition for the experienced developers who remain, and creating a seller’s market for mid-level and senior talent that would be the logical result of today’s hiring decisions if you follow them forward five years.
The counterarguments
Not everyone in the industry is following the herd. Some of the strongest dissent comes from people with the scale to test their convictions.
In February 2026, IBM’s CHRO Nickle LaMoreaux announced the company would triple its U.S. entry-level hiring for software developer roles, explicitly warning that cutting entry-level hiring “creates long-term risks: without enough early-career hires, companies may struggle to develop future mid-level managers.” What makes IBM’s approach notable is that it is not simply maintaining traditional junior roles. It is restructuring them around AI augmentation: juniors handle cases where AI chatbots fall short, verify AI output, correct AI errors, and manage customer escalations. This is the most concrete company-scale example of what a restructured junior role in an AI-augmented environment actually looks like.
AWS CEO Matt Garman has been the most publicly forceful critic of the junior-replacement thesis. He called replacing junior developers with AI “one of the dumbest things I’ve ever heard,” offering three specific counterarguments: juniors are often more AI-fluent than seniors and extract more value from AI tools; juniors are the least expensive employees, so eliminating them delivers minimal cost savings while destroying the talent pipeline; and companies without a junior pipeline lose their source of fresh ideas and future mid-level talent.
Garman’s pipeline logic is actually consistent with the evidence in this piece — he agrees on the structural problem, he just disputes whether the employer response is rational. Cloudflare announced plans to hire 1,111 interns in 2026 specifically to “help train the next generation of technology leaders.” Tech apprenticeships grew 29% over four years.
These are real signals. They suggest some companies recognize the pipeline problem and are investing in alternatives. But they remain the minority response. The majority of the market — as measured by posting data, hiring share, and engineering leader surveys — is moving in the opposite direction. Salesforce CEO Marc Benioff announced the company would hire no new software engineers in 2025, citing 30% AI-driven productivity gains. One industry expert summarized the prevailing economic logic: “Why hire a junior for $90K when GitHub Copilot costs $10?”
The countertrend is real but narrow. IBM, AWS, and Cloudflare are making long-term bets. Most of the market is optimizing for the next quarter.
The higher first rung
The career ladder is not being removed. It is being rebuilt with a higher first rung — and whether that matters depends on who you are.
Hiring managers are explicit about what they want from the entry-level developers they do hire. HackerRank’s 2025 Developer Skills Report — based on 26 million developers and over 3 million assessments — identifies debugging as “the most important AI-age skill to assess” in hiring. Seventy percent of hiring managers now prioritize relevant experience and skills over formal degrees. Thirteen percent of entry-level jobs explicitly require AI skills. Juniors must prove systems-level thinking, cloud and AI fluency, and production experience. Portfolio projects built from tutorials no longer suffice.
What was previously a junior role — writing boilerplate, implementing straightforward features, fixing simple bugs — is being automated. The new entry point requires skills that were previously mid-level: understanding system architecture, reviewing AI-generated code, reasoning about security, and communicating with stakeholders about trade-offs.
This is painful for anyone entering the field right now, but it is not the same as the ladder being removed entirely. The Anthropic study’s key finding offers a path: the difference between juniors who maintained skills and those who lost them was not whether they used AI, but how. Use AI to generate code, then explain it to yourself. Debug without AI assistance. Ask conceptual questions rather than copying output. The 17-percentage-point skill gap between AI-dependent and AI-augmented learners is the difference between employable and not.
But the Harvard study’s finding — that graduates from mid-tier institutions are worst affected — raises a harder question. The new entry requirements — production-grade portfolio work, demonstrated AI fluency, codebase familiarity — may be more reachable by those who already have strong credentials: students at elite CS programs, candidates with industry networks, developers with access to meaningful open-source projects. Career changers, bootcamp graduates, and students without those advantages face a first rung that may be effectively out of reach. The ladder is being rebuilt for some entrants while remaining effectively removed for others.
There is one sector that is bucking the broader trend entirely. Healthcare entry-level tech postings rose 13 percentage points, and 75% of jobs added in late 2024–2025 were in healthcare, government, and leisure/hospitality. The junior developer decline is concentrated in traditional tech, not universal. For entry-level developers who are willing to look beyond the tech sector, the ground is harder but not gone.
Where this leaves the ladder
The missing rung problem cuts differently depending on where you stand on the ladder.
If you are entering the field: The path is harder than it was three years ago, and no one owes you an apology for that. The hiring market has changed; your strategy must change with it. Build debugging and review skills deliberately — the Anthropic study is the clearest available guide. Demonstrate production-grade judgment through open-source contributions and real-world project experience, not tutorial capstones. Target sectors still hiring juniors: healthcare tech, government and defense, cybersecurity, infrastructure. These sectors have regulatory requirements, security clearance needs, or domain complexity that makes them less susceptible to AI automation and more willing to invest in training. The Stack Overflow 2025 survey found 55.5% of early-career developers already use AI tools daily — fluency with these tools is a baseline expectation, not a differentiator.
If you manage engineers: The economic logic of cutting junior hiring is real in the near term and self-defeating on a five-year horizon. Every junior not hired today is a mid-level engineer you will not have in 2030. IBM’s restructured junior roles — AI exception handling, output verification, customer escalation — suggest what the new entry-level position looks like. It is not the old junior role. It is a role designed around the reality that AI generates the first draft and humans ensure it actually works. If you are going to maintain a junior pipeline, design it for what juniors actually need to learn now: not syntax, but judgment.
If you are a senior developer: The missing rung is not your immediate problem. It is your downstream problem. The profession that trained you is contracting the mechanism it used to do so. In five years, the mid-level engineers who would have handled the work that sits between your architectural decisions and production deployment may not exist in sufficient numbers. The review burden documented in “The Productivity Mirage” — AI generating more code that requires more human verification — lands disproportionately on experienced developers. That burden does not decrease when there are fewer junior and mid-level developers to share it. It concentrates.
The industry is making a bet: that AI will advance fast enough to fill the gap left by the juniors it chose not to hire. The evidence so far is not encouraging. Developers consistently overestimate how much AI accelerates their work — the METR randomized controlled trial found experienced developers were actually 19% slower with AI assistance, despite believing they were faster. The Anthropic study found a 17-point skill deficit in AI-dependent learners. Code quality metrics across multiple analyses show AI-generated output requiring substantially more review, more debugging, and more security remediation than the human-written code it replaces. The bet is not paying off yet.
It may pay off eventually. The tools are improving. But “eventually” is not a staffing plan. The junior not hired today is a mid-level engineer missing in 2030 regardless of how good Copilot gets. AI can generate code. It cannot generate the five years of production experience that turns a junior into someone you trust with your architecture.
The rung is missing. The climb still has to happen. The question is who rebuilds the ladder — and whether they start before the gap becomes unfillable.