Entry-level software engineering hiring has contracted. However, the specific figures dominating the conversation, a 67% reduction in postings and a 73% reduction in hires, are considerably less well supported than their circulation suggests, and in at least one case they conflict directly with figures published by outlets citing the same underlying claim.
This matters beyond pedantry. Hiring managers are setting requisition strategy on these numbers. Universities are advising students on them. Anyone building a workforce plan needs to know which figures survive scrutiny.
The conflicting claims
Read across the current coverage and you find the same claim rendered at wildly different magnitudes:
Those are descriptions of the same phenomenon differing by a factor of more than two. Both appear in posts published within months of each other. Neither traces to an identifiable primary source.
The 73% figure is also frequently misdated. One post published in March 2026 describes it as a "73% hiring drop in the past year," then in the same passage attributes it to a window running from October 2023 to November 2024. Those cannot both be true. The framing makes a two-year-old measurement sound like current-year data.
The pattern behind the 67% and 73% numbers, where it is described at all, is "industry data tracking US job boards." That is not a citation. Job board scrapes vary enormously by which boards are sampled, how "entry level" is classified, whether reposts are deduplicated, and whether the denominator is postings or requisitions. Without methodology, the number cannot be evaluated.
What holds up
Several measurements in this area come from named institutions with published methodology. These are the ones worth building an argument on.
Stanford Digital Economy Lab, using ADP payroll records. This is the strongest evidence available, because it measures employment rather than job postings. Postings measure employer intent and are trivially inflated by roles advertised but never filled. Payroll records measure people actually working. The finding is that entry-level employment for programmers aged 22 to 25 declined by approximately 20% from its late-2022 level, with the decline continuing into 2026.
Note the magnitude. The more defensible payroll-based figure is roughly one third the size of the widely circulated 67% and 73% claims.
IEEE Spectrum on the largest tech employers. Entry-level hiring at the fifteen largest tech firms fell 25% from 2023 to 2024. This is a narrow slice of the market, and it should be presented as such rather than generalized to all employers.
New York Fed labor market tracker for recent college graduates. Unemployment for recent graduates rose to approximately 5.7% in Q4 2025, with underemployment at 42.5%. The second figure is the more revealing one and is almost never reported. A graduate working outside their field is not captured in unemployment statistics, and underemployment above 40% describes a labor market where the entry problem is placement quality rather than placement absence.
Handshake, Class of 2026 survey. 70% of computer science majors report feeling at least somewhat pessimistic about their careers, and 42% use generative AI daily. Sentiment data, clearly labeled as such.
NACE Winter 2025 Salary Survey. Computer science tied for the top spot in employer demand among bachelor's degrees, with 67.1% of surveyed employers planning to hire those majors. Average projected pay for the Class of 2025 computer science graduates was $76,251, with software engineering majors projected at $82,536.
That last data point is the one that most cleanly contradicts the collapse narrative. Employer demand for developers remains at the top of the survey while entry into the field has become genuinely harder. Both conditions are real at once.
The Mechanism
The causal story generally offered is that AI tools absorb exactly the work that used to justify a junior hire: boilerplate, small well-scoped tickets, CRUD endpoint wiring, minor bug fixes. That work was also the training ground. Remove it and the first rung of the ladder goes with it.
That mechanism is plausible, but it is not established. A competing explanation, that young workers are absorbing the effects of a slow general hiring cycle, fits the same data. Training and mentorship budgets are also typically the first cut in a cost-reduction cycle, which biases hiring toward candidates who contribute immediately regardless of any AI effect. All three can be operating simultaneously, and the available evidence does not apportion between them.
Anyone presenting a single confident causal account of this is presenting speculation as finding.
The Counterweight nobody includes
Long-range projections point in the opposite direction from the near-term data. The Bureau of Labor Statistics projects 15% employment growth for software developers, QA analysts, and testers from 2024 to 2034, with roughly 129,200 openings per year, against a 3.1% average across all occupations.
There is also observable countercyclical behavior. IBM tripled its entry-level hiring in 2026 on the reasoning that AI handles many junior tasks while still requiring human oversight. Enterprise organizations, defense contractors, financial institutions, and healthcare technology firms continue hiring juniors at relative consistency, because their codebases and compliance requirements demand judgment that generated code does not supply.
That last point deserves emphasis for anyone in regulated or public sector work. Environments with authority-to-operate obligations, audit trails, and formal review gates (like we have here at Interclypse) need people who understand systems deeply rather than people who ship quickly. Those environments still have a reason to train. That’s why we continue to emphasize our hiring of junior level positions. Part of our dedication to having a positive and transformational impact on the IT industry is preparing rising professionals for the constantly evolving tech environment. AI code is powerful, but far from perfect. Our entry level staff will always be trained in the best practices so that when they see something wrong they know how to fix it.