The Entry-Level Tech Squeeze: What Junior Developers Are Actually Up Against

Here is the pairing that explains everything, and almost nobody puts these two figures next to each other.

People already working in computer and mathematical occupations had an unemployment rate of 2.8% in July 2026 — better than the 4.3% national average, better than most of the economy.

Recent computer science graduates are at roughly 7%.

Tech jobs are not disappearing. Employment in computer systems design and related services sits at 2.37 million and has been essentially flat for two years — not collapsing, not growing. The industry is fine. The industry is simply not opening the door as often, and when it does, it is looking for someone who has already been inside.

The NY Fed puts recent graduate unemployment overall at 5.6% and underemployment at 42% as of Q2 2026. Underemployment is the quieter number and often the more painful one: it means working, in a job that doesn’t require the degree.

One honest caveat on the by-major figures, because they get abused constantly. The Economic Innovation Group ran the numbers behind that viral unemployment-by-major chart and found computer engineering’s 7.5% carried a 95% confidence interval of 4% to 11%. These are small subsamples of a census survey. So “CS unemployment is elevated and has drifted up” is supportable. “CS unemployment jumped 15% this year” is noise wearing a suit. EIG’s alternative framing is worth holding onto too: on an employment-to-population basis, about 90% of 22-to-27-year-olds with computer or information degrees were employed.

You have read that AI is eliminating entry-level tech jobs. The real picture is more specific and more interesting.

Three independent research teams, using three unrelated datasets, found the same age-skewed pattern:

• Stanford’s Digital Economy Lab (“Canaries in the Coal Mine,” August 2026 revision), using ADP payroll records, found employment for 22-to-25-year-olds in the most AI-exposed occupations running 19% below where it would be had they kept pace with less-exposed peers. Experienced workers showed no comparable gap.

• Harvard researchers Hosseini and Lichtinger, using LinkedIn-derived data on 284,974 firms, found junior employment falling roughly 9–10% in the six quarters after a firm adopted generative AI, with senior employment unchanged.

• The Dallas Fed, using Census CPS microdata, found a 13% employment decline for 22-to-25-year-olds in the most AI-exposed occupations since November 2022.

Every one of those studies found the same mechanism, and it matters enormously for how you job hunt: separations actually fell. Nobody is being fired. Firms stopped hiring juniors. Harvard’s paper measures it precisely — five fewer junior hires per quarter, with promotions unchanged. It’s a hiring freeze, not a purge.

Now the part most coverage skips.

The NY Fed examined job postings across 2018 to January 2026 and concluded that “labor demand for junior and senior roles within highly exposed occupations is moving broadly in parallel” — no meaningful divergence at all, and the relative declines in high-exposure occupations predate ChatGPT. The Federal Reserve Board found “no evidence of a reduction in job postings for industries or firms which have higher levels of AI adoption,” describing its results as “precisely-estimated null effects.” The St Louis Fed concluded that the general decline in labor demand “accounts for most of the deterioration in young workers’ outcomes,” with AI meaningful but secondary.

And every author of the age-skew papers says the same thing about their own work. Stanford’s, verbatim: “We interpret these facts as early, descriptive indicators — canaries in the coal mine — rather than causal estimates.”

So: the entry door narrowed, measurably and across datasets. Whether AI narrowed it is genuinely unresolved, and anyone selling you certainty in either direction is selling something else.

This is the finding that reframes the whole problem, and it’s counterintuitive enough that it’s worth sitting with.

Indeed’s Hiring Lab tracked experience requirements in tech postings. The share open to candidates with a year or less of experience went from 17% in Q2 2022 to 18% in Q2 2025. It went up.

What moved was the middle. Postings seeking 2–4 years fell from 46% to 40%. Postings seeking 5+ years rose from 37% to 42%. In non-tech occupations, senior requirements actually eased over the same period — this tightening is tech-specific.

So there are proportionally just as many entry-level tech postings as there were four years ago. The total pool is a quarter smaller, and the rung you used to climb to has thinned. The two-years-experience job that used to be your second move is the one that disappeared.

Most tech workers don’t work in tech. Only about 9% of software developers are employed by software publishers. Thirty percent are at computer systems design firms, 10% in finance and insurance, 8% in manufacturing. If your search is scoped to companies whose product is software, you are ignoring most of the market.

And most computer degree holders aren’t developers. Just 24% of people with a computer or IT bachelor’s work as software developers. The rest are in systems analysis, security, support, IT management, data roles. Median wage across the whole group: $100,000.

AI skills are in demand mostly outside tech. Lightcast analysed 1.3 billion postings and found 51% of AI-skill job postings sit outside IT and computer science occupations, carrying a roughly 28% salary premium — about $18,000 a year. The fastest growth in AI-skill requirements is in HR (+66%), marketing (+50%) and finance (+40%). The person who can do the domain job and work with these tools is currently rarer than either specialist.

Do not assume government is a refuge. Federal employment is down 327,000 — nearly 11% — since January 2025. That door is narrower, not wider.

Internships are the best-evidenced intervention available. The Burning Glass Institute, working from 60 million-plus career histories, found that doing an internship cuts the odds of being underemployed by 49%. For computer science specifically: 29% underemployment with an internship, 38% without.

And conversion has rebounded hard. NACE reports intern-to-full-time conversion at 63.1% for 2024–25 interns, the highest in five years. One detail worth planning around: in-person internships converted at a 72% offer rate against roughly 56% for hybrid.

Registered apprenticeships exist, at modest scale. The Department of Labor served more than 58,000 registered apprentices in technology, cybersecurity and AI-related occupations in 2025. Against 1.9 million software developers, that’s a real door but a narrow one.

There is no authoritative figure for how long it takes a computer science graduate to find their first job. I went looking. Everything claiming to know is either a job board’s self-reported panel survey or an SEO farm.

The closest honest benchmark is NACE’s First Destination Survey, which measures outcomes six months after graduation: for the Class of 2024, an 85.7% career outcomes rate overall, with 14.1% still seeking. Computer and information sciences saw its outcomes rate decline 1.9% year over year.

Worth knowing, since it complicates the doom narrative: NACE’s spring 2026 survey has employers projecting a 5.6% increase in Class of 2026 hiring, with the Information sector and engineering services growing while computer and electronics manufacturing declines.

The market is not rejecting you personally. It has narrowed one specific door — first jobs in a sector where postings are down about a quarter — while the inside of that sector stays healthy and the adjacent industries that employ most tech workers keep hiring.

Which means the tactics that work are unglamorous. Widen the aperture past software companies. Get the internship or apprenticeship if you’re anywhere near eligible, because it’s the only intervention here with a hard causal-ish number attached. Learn the AI tooling as a domain skill rather than a tech one, since that’s where half the demand actually sits. And prepare properly for the interviews you do get, because there are fewer of them and each one carries more weight.

On that last point, our interview question generator builds question sets by industry and role level — useful mostly because it forces you to practise for the job you applied for rather than the generic one.

Nothing about this is your fault. Some of it is still your problem. Both things are true.

Sources: Indeed software development postings, via FRED · Indeed total postings, via FRED · NY Fed, Labor Market for Recent College Graduates · EIG on the viral graduate unemployment chart · BLS CPS Table 25b · Stanford, Canaries in the Coal Mine · Hosseini & Lichtinger, SSRN · Dallas Fed · NY Fed Liberty Street on AI and job postings · Federal Reserve Board FEDS Notes · St Louis Fed · Indeed Hiring Lab on experience requirements · BLS OOH, Software Developers · Lightcast, Beyond the Buzz · Burning Glass Institute, Talent Disrupted · NACE intern conversion · NACE Job Outlook 2026 Spring Update · DOL Apprenticeship.gov