Harvard: AI is cutting junior hiring, and in three years that could affect the entire industry

Generative AI is already changing hiring structures in IT: companies are cutting entry-level positions, and experienced developers increasingly rely on AI tools. If both trends persist, in a few years the market may face not a surplus, but a deficit of people capable of maintaining and fixing accumulated code.

What Harvard Found

Harvard's research covered nearly 62 million workers and 285,000 companies in the US. The key finding is simple: after implementing generative AI, companies begin hiring juniors noticeably less frequently. Hiring decline is estimated at 9–10% over six quarters, while demand for more senior roles remains almost unchanged. The authors separately highlight an important detail: this is not about mass layoffs, but about slower entry into the profession. The lower rungs of the career ladder are shrinking first.

This picture is reinforced by other signals gathered in the article:

• Employment of 22–25-year-old developers, according to Stanford Digital Economy Study, fell 20% from late 2022 peak to July 2025
• Number of entry-level vacancies in the US, according to Revelio Labs, dropped 35% from January 2023 to June 2025
• In the UK, decline in such vacancies at Revelio Labs reaches 46%
• Share of juniors in IT hiring, according to Stack Overflow, declined from 15% to 7% over three years

Individually, these figures could still be attributed to market cooling after the overheated 2021–2022 period. But together they show a more concerning shift: companies are already behaving as if AI has taken on a significant portion of tasks that used to be given to newcomers. Particularly important is that this is not a one-time spike, but a series of consistent signals across different samples and countries. If juniors aren't entering the system now, in a few years there will simply be nowhere to draw mids from.

Dependence on Assistants

The second alarming signal came from METR research. The organization tested 16 experienced open-source developers on real tasks in projects familiar to them. Participants' expectations were optimistic: they assumed AI would speed up work by about 20%. In practice, the opposite happened — when AI assistants were allowed, tasks were completed on average 19% slower. That is, subjectively the tools seemed useful, but in fact they added time to prompts, waiting, code verification, and corrections.

"It's like walking across the city when you're used to taking

Uber."

At the next stage, METR encountered not just a productivity question, but a habit question. According to data cited in the article, 30–50% of developers refused to send part of tasks to the experiment if they had to do them without AI. Researchers directly pointed to a selection effect: people who depend most on assistants are exactly those dropping out of the sample. This does not prove that AI is useless. But it shows how quickly it becomes not just an accelerator, but a crutch that many find hard to live without now.

Where the Risk Emerges

The problem doesn't come down to newcomers finding it harder to land their first job. The very logic of growing engineers within teams disappears. Previously, a junior took on routine work, made mistakes, got reviews, and gradually became a mid-level engineer. Now that routine is often handed to models, and a senior assumes three roles at once: their main work, oversight of AI, and final verification of results. This creates a talent doom cycle: today you save on junior positions, but in 3–5 years you discover a lack of people at the next level.

Meanwhile, the cost of errors is rising. The article cites estimates that AI code more often carries vulnerabilities, increases duplication, and reduces the share of refactoring. Even if you dispute specific percentages, the mechanics are clear: writing draft code has become easier, but sorting through its consequences has not. So part of the speed gain transforms into deferred technical debt. If juniors are hired less, and seniors spend more time reviewing machine output, then in a few years the industry may get not an army of super-productive developers, but a narrow bottleneck of overloaded specialists and bloated code bases.

What This Means

Right now, the IT market is testing a model in which AI closes the lower tier of engineering work. In the short term, this looks like savings, but in the long term, it may break the talent pipeline. If companies don't return to meaningful junior hiring and training, there simply won't be enough people to sort through the accumulated AI technical debt in a few years.