The Infinite Intern

Somewhere on LinkedIn right now, there is a job posting for a junior software developer. It requires three to five years of experience with AI coding tools. GitHub Copilot became generally available in June 2022. ChatGPT launched in November 2022. It is March 2026. The maximum possible experience anyone on Earth could have with these tools is three years and nine months — and that assumes they started on day one, which means they were already employed as a developer, which means they were not junior. The posting is asking for a person who cannot exist. And that, in miniature, is the entire problem.

The tech industry has spent the last two years replacing its entry-level workforce with AI agents. Companies simply stopped hiring. Positions were quietly eliminated. Headcount was "optimized." The juniors didn't lose their jobs. The jobs lost their juniors.

This is a story about what happens next. Not to the AI — the AI will be fine. To the profession. To the pipeline that turns confused twenty-two-year-olds into the senior engineers that every company says it desperately needs. To the career ladder when someone removes the bottom rung and then wonders why nobody is climbing.

The Metaphor That Said Too Much

In July 2023, the tech analyst Benedict Evans published an essay called "AI and the automation of work." In it, he offered a metaphor that would become industry scripture: "The last wave of machine learning gave you infinite interns who could read anything for you, but you had to check, and now we have infinite interns that can write anything for you, but you have to check."¹

"Infinite interns." The phrase spread through tech culture like a virus. Developers picked it up. Managers picked it up. VCs built pitch decks around it. Within months, the metaphor had calcified into conventional wisdom: AI is like having an unlimited supply of eager, somewhat careless junior workers who can crank out boilerplate code all day, provided someone experienced checks their output.

The metaphor is useful. It is also, in the way of the best metaphors, far more revealing than its creator intended. When a CEO says "AI is like having infinite interns," they are telling you something important about how they always valued interns. Not as future professionals being developed. Not as investments in a pipeline that would eventually produce the senior talent that actually runs the company. They valued interns as fungible labor for grunt work — warm bodies who do the tedious tasks that experienced engineers don't want to do.

The "infinite" part is the quiet part out loud. They wanted interns to be disposable all along. Now they literally are. An AI agent never asks for a return offer. It needs no desk, mentor, or performance review, and it never becomes senior. And in the boardroom, that last part sounds like a feature — not a catastrophic design flaw in your talent strategy.

The practical adoption of the metaphor followed its implications to their logical end. Developers began describing prompt engineering as "talking to a really dimwitted intern" — spell everything out, never assume context, check every output. Companies restructured their teams around the assumption that AI could absorb the work that juniors used to do, while seniors supervised the machine the way they once supervised people. The only difference: you don't have to train the machine. You don't have to grow it. You don't have to worry about its career path.

A Nature Communications Psychology study surveying over 12,000 Americans found that people increasingly view AI as "warm and human-like," which makes the intern framing feel natural rather than monstrous.² The replacement gets anthropomorphized; the replaced get dehumanized.

The Numbers

The data on what happened to actual human interns and junior developers is not subtle. It is a demolition.

Entry-level tech job postings dropped 60% between 2022 and 2024.³ Tech-specific internship postings declined 30% since 2023, according to Handshake data.³ Overall software developer listings are down ~35% from pre-2020 levels and ~70% from the 2022 peak. Entry-level tech hiring decreased a further 25% year-over-year in 2024.⁴

The Indeed Hiring Lab put the experience inflation in stark terms: only 18% of US tech postings in Q2 2025 were open to candidates with one year or less of experience. Eighty-two percent required at least two years. The share demanding five or more years rose from 37% to 42% between 2022 and 2025.⁵ The entry level is disappearing not because the positions were eliminated on paper, but because the experience requirements were inflated until no entry-level candidate could meet them. Over 60% of positions labeled "entry-level" in software and IT now require three or more years of prior experience.⁵

Software developers aged 22 to 25 saw employment drop nearly 20% from late 2022 to July 2025.³ Young adult unemployment in tech (ages 22-27) hit 7.4%, nearly double the national average of 4.2%. CS graduate unemployment stands at 6.1%. Computer engineering graduates: 7.5%.⁴ Those numbers come from Rezi, a resume-building company — not a neutral research institution — so treat them with appropriate caution. But even if the exact figures are off by a point or two, the direction is unmistakable: the golden ticket of the last two decades is losing its shine.

And then there is the Harvard study. Researchers Hosseini Maasoum and Lichtinger analyzed 62 million workers across 285,000 US firms between 2015 and 2025. Their finding: in companies that adopted generative AI, junior employment fell 7.7% relative to non-adopting firms within just six quarters of adoption (starting Q1 2023). Senior employment remained largely unchanged.⁶

Their term for the phenomenon is precise and damning: "seniority-biased technological change." And their finding about the mechanism is crucial — the decline is driven by slower hiring, not layoffs. Companies are not firing their juniors. They are simply not replacing them when they leave, and not hiring new ones to begin with. Nobody gets a pink slip. No one gets marched out of the building with a cardboard box. Positions just quietly vanish from the careers page, and a generation of would-be developers never gets its foot in the door.

This is why there has been no visible outcry. You can't protest a job that was never posted.

A 2025 LeadDev survey found that 54% of engineering leaders plan to hire fewer juniors, explicitly citing AI copilots that enable seniors to handle more work.³ Seventy percent of hiring managers believe AI can perform intern-level work. Fifty-seven percent say they trust AI output more than the work of interns or recent graduates. And 37% of employers say they now prefer "hiring" AI over hiring a recent graduate.⁴

One anecdote that captures the emotional weight: a 2023 CS graduate reportedly applied to 5,762 tech jobs without securing an offer.⁴ We can't independently verify the number — it comes from a single industry report — but it's consistent with the broader data on application volumes in a collapsing entry-level market. The infinite intern replaced by the infinitely rejected applicant.

The Pipeline Problem

Here is the question that the "infinite intern" metaphor obscures, the question that no AI agent will ever answer, because it concerns a future the model was not trained on: where do senior engineers come from?

They come from junior engineers. Every principal architect, every staff engineer, every CTO who ever debugged a production outage at 3 AM started as someone who didn't know what they were doing. They wrote bad code. They broke the build. They sat in code reviews and learned — not just syntax, but judgment. When to refactor and when to leave it alone. When the clever solution is actually the dangerous one. When the spec is wrong and you need to push back. This knowledge cannot be downloaded. It is accumulated through years of doing the work badly, getting feedback, and doing it slightly less badly the next time.

The typical progression from junior to senior developer takes 5 to 7 years, and up to 10 for specialized roles.⁷ This is not a number that can be optimized. It is not a pipeline that can be parallelized. It is the time it takes for a human brain to develop the pattern recognition, institutional memory, and professional judgment that distinguishes a senior engineer from someone who can write code.

The Octopus Deploy AI Pulse Report laid out a projected crisis in three phases — speculative forecasting, not established fact, but consistent with the hiring data:⁷

Phase 1 (2025-2027): Junior hiring freeze. This is happening now. 73% of organizations have reduced junior developer headcount over the past two years.

Phase 2 (2027-2030): Mid-level developer shortage emerges. The juniors who were never hired in Phase 1 never became mid-levels. Companies begin competing for a shrinking pool of experienced talent. Wage inflation accelerates.

Phase 3 (2030-2035): Senior developer crisis. The pipeline has been dry for five to ten years. The seniors who retired were never replaced. The widening skills gap becomes structural and self-reinforcing.

The math is straightforward and merciless. Saving $150K/year by not hiring a junior today will cost $2-3 million in increased senior compensation and recruiting costs by 2030-2035.⁸ The bargain looks good on this quarter's balance sheet. It looks catastrophic on the ten-year horizon. But quarterly earnings reports do not have a line item for "talent pipeline we destroyed."

Meanwhile, the existing senior talent is not getting any younger. 18% of senior developers born between 1970 and 1980 plan to retire before 2027.⁷ And 44% of mid-career developers are somewhat or strongly considering a career change.⁷ The pipeline is being squeezed from both ends — nobody entering at the bottom, experienced people leaving from the middle and the top.

The Stack Overflow Developer Survey is becoming evidence of the problem in real time. The 2025 survey (49,000+ respondents across 177 countries) showed a striking decline in the proportion of 18-to-24-year-old developers compared to 2022.⁹ The survey now skews heavily toward experienced professionals — 76% are professional developers with years of experience. As junior developers disappear from the profession, they disappear from the survey. The "developer community" is aging in place, and the data that's supposed to describe the field is documenting an extinction event.

This Has Happened Before

The tech industry likes to believe it is unprecedented. It is not. Other professions have faced the exact same question — what happens when you automate the junior role? — and their answers are instructive, if not entirely reassuring.

Law after Westlaw. Before digital legal databases, a junior associate's life was consumed by "Shepardizing" — the painstaking process of cross-referencing legal citations in physical volumes. It could consume entire workdays. Lexis launched in 1980. Westlaw followed around 1990. Research that once took hours could suddenly be done in minutes.¹⁰

The demand for junior lawyers did not plummet. Instead, as technology made associates more efficient, firms raised their expectations. The associates were not doing less work — they were doing more, faster. More research memos. More briefs. More analysis. Paralegals transformed from administrative support into "strategic, tech-savvy professionals." The legal profession absorbed the automation by raising the bar, not cutting headcount.¹⁰

But there is an important structural distinction: law firms bill by the hour and maintain a partnership track. The economics of legal practice mean that junior associates are revenue generators from day one, not cost centers. Software companies have a very different economic relationship with their entry-level workers.

Accounting after spreadsheets. The accounting profession has been through wave after wave of automation — mechanical calculators, digital calculators, Excel, enterprise resource planning systems — and each wave eliminated routine tasks while creating demand for advisory and strategic skills. Junior accountants who accept AI output at face value show smaller performance gains than skeptical seniors, according to Stanford GSB research.¹¹ The accounting talent shortage persists despite decades of automation. The junior positions got leaner, but the pipeline survived because firms recognized that you need humans who understand the fundamentals before they can exercise judgment about the output of the machines.

Architecture after CAD. The Bureau of Labor Statistics projects declining employment for many types of drafters as CAD and BIM software increase productivity — engineers and architects can now perform many tasks that previously required dedicated drafters.¹² But junior drafting positions adapted rather than vanished. CAD proficiency became table stakes. Design thinking became the differentiator. This is the optimistic parallel: the junior role transforms, the entry point shifts, but the ladder remains.

The common thread across these professions is that the ones that survived automation of the junior role did so by maintaining the pipeline, even when the nature of junior work changed. The junior accountant still exists; they just do different junior work. The junior architect still exists; they just use different tools. The question for software engineering is whether the industry will follow the same pattern — or whether, intoxicated by the speed and cheapness of AI output, it will do something the other professions were smart enough not to do: eliminate the entry point entirely and assume the seniors will just last forever.

The Smartest Man at AWS

In December 2025, AWS CEO Matt Garman said the thing that nobody else at his level was willing to say. In an interview with Fortune, he called the idea of replacing junior developers with AI "one of the dumbest things I've ever heard."¹³

His reasoning was specific and structural. First, entry-level workers are paid the least — so the supposed cost savings are minimal compared to the damage. Second, junior employees bring fresh ideas and energy that organizations need. And third — the critical point — the pipeline dies. "At some point," Garman said, "that whole thing explodes on itself."¹³

Garman's framing is notable because it comes from inside the machine. This is not an academic making a theoretical argument or a journalist writing a worried think piece. This is the CEO of the world's largest cloud computing platform — a company that sells AI services — telling his industry that its approach to junior talent is suicidal. He is describing a system that will, given enough time, destroy itself.

The counterpoint came from his peers. Marc Benioff announced that Salesforce was "not adding any more software engineers next year" — though the company continued hiring in sales and other roles.¹⁴ Google and Meta cut new-graduate hiring by approximately 50% compared to 2021.¹⁴ Google CEO Sundar Pichai frames AI as "making engineers more productive by eliminating tedious tasks" — a productivity booster, not a replacement. But the hiring numbers tell a different story than the press release.

The BLS, meanwhile, still projects 15% growth for software developers over the 2024-2034 decade — "much faster than average" — with roughly 129,200 openings projected per year.¹⁵ The 2024 median annual wage remains a robust $133,080. These projections are real, but they carry a critical blind spot: the 129,200 annual openings figure does not break down by seniority. The growth is real, but it is concentrated in senior and specialized roles. The BLS projections predate the full impact of AI coding agents on entry-level positions, and they describe a world that may no longer exist by the time the decade is over.

Note also: the BLS separately tracks "computer programmers" as a distinct category, and that category is projected to decline — continuing a long trend of the "programmer" role being absorbed into "developer" and "engineer" titles.¹⁵ The job isn't disappearing. The version of it that a junior could do is disappearing.

The Debugging Crisis

Even the juniors who do manage to get hired are entering the profession in a fundamentally different way than any previous generation. 97% of CS students now use AI tools for their education. 75% say they would continue using them even if their programs banned it.¹⁶ Students show a 10% improvement in exam performance when using AI — but the researchers flag a troubling possibility: these students may lack the foundational mental models that come from struggling through problems without assistance.

The DEV Community, a major developer forum, ran a widely-shared piece in early 2026 titled "The Junior Developer Crisis of 2026: AI Is Creating Developers Who Can't Debug."¹⁶ The thesis is simple: if you learn to code by generating code with AI and tweaking the output, you never develop the deep understanding of how systems actually work. You can produce working code. You cannot explain why it works. And when it breaks in a way the AI hasn't seen before, you are helpless.

This is the learned helplessness problem wearing a different hat. The Stanford GSB finding applies across professions: junior workers who accept AI output at face value see smaller performance gains than skeptical seniors.¹¹ Skepticism is a skill. It is earned through experience — specifically, through the experience of being wrong, understanding why, and developing the instinct to question plausible-looking output. If you never go through that process because the AI always gave you an answer, you never develop the instinct.

Bootcamp curricula are scrambling to adapt. General Assembly now teaches students to leverage AI tools for productivity. Codesmith launched an "AI engineering" track. The hot skills additions are AI/ML, cloud computing, data analytics.¹⁷ The shift in what "entry-level" means is tectonic: the portfolio that gets you hired in 2026 is not a todo app — it is documented judgment. Systems thinking. Architectural intent. The ability to explain not just what you built, but why you built it that way and not another way.

Bootcamp placement rates remain nominally high — 70-80% on average, with top programs above 90%.¹⁷ But these numbers may mask longer job search timelines, lower starting salaries, and the growing requirement for AI-specific portfolio work just to get in the door. The surface metric looks healthy. The lived experience of graduates does not.

AI tool adoption among developers reached 84% in 2025.⁹ The centaur model — human judgment paired with AI capability — is becoming near-universal. But the centaur model assumes the human half has skills worth pairing. If the human never developed those skills because the entry-level job that would have taught them no longer exists, you don't have a centaur. You have a rider who never learned to ride, sitting on top of a horse that doesn't know where it's going.

Eating the Seed Corn

There is an agricultural metaphor for what the tech industry is doing, and it is older than software, older than computers, older than electricity. It is called eating your seed corn.

Seed corn is the portion of the harvest you set aside to plant next year's crop. It is not surplus. It is not waste. It is the future, stored in the present. Every bushel of seed corn you eat this winter is a field that lies fallow next spring. The math is not complicated. It is just uncomfortable.

Junior developers are the seed corn of the software profession. They are not a cost center. They are not an inefficiency to be optimized away. They are the mechanism by which the profession reproduces itself. Every junior who doesn't get hired is a senior who won't exist in seven years. Every intern program that gets replaced by an AI agent is a pipeline that stops flowing. And the pipeline, once drained, cannot be refilled quickly — because the senior engineers who would do the mentoring are themselves burning out, leaving the field, or being laid off in the next round of "AI-motivated headcount reductions."

The Octopus Deploy report estimates that 74% of companies struggle to scale AI value — which, paradoxically, may accelerate the hiring freeze.⁷ These are projections from an industry report, not peer-reviewed findings, but the directional logic is sound: the AI isn't delivering the promised productivity gains, but instead of concluding that they need human judgment to make it work, companies double down on the bet. More AI. Fewer humans. Surely this time it will work.

Meanwhile, the humans who remain are stretched thinner. Senior developers earn 1.4 to 2.1 times more than juniors.⁷ When the shortage hits — and it will hit, with the mathematical certainty of a crop that was never planted — the wage inflation will be extraordinary. Companies that saved $150,000 a year by not hiring a junior in 2025 will spend millions competing for the seniors that junior would have become.

And here is the deepest irony: the companies that will be most desperate for senior engineers in 2032 are the same companies that are most aggressively eliminating junior positions in 2026. They are, right now, in this fiscal quarter, with their current hiring freeze, creating the exact crisis they will spend the next decade trying to solve. They know this. The data is in front of them. Garman said it out loud. And they are doing it anyway, because the quarterly earnings report is due in three weeks and the talent crisis is due in seven years, and no one in the C-suite will still be in the same job by then.

The machines write code. The humans review code. But nobody is growing the next generation of humans who will know enough to review anything at all. The infinite intern never gets promoted. It never develops judgment. It never pushes back on a bad architecture decision because it was there three years ago when the same mistake was made and saw what happened.

It generates infinitely, confidently, without understanding.

And somewhere, a twenty-two-year-old with a computer science degree and $87,000 in student loans is submitting application number 5,763. The posting says "entry-level." The requirements say "3-5 years experience." The hiring manager says they prefer AI.

The profession is eating its seed corn. And when the harvest fails, everyone will act surprised.