Junior Developer Hiring Is Down 20%, 25%, 50%, 67% or 73%. Each Number Measures Something Different.

Search for junior developer hiring and the first page offers a decline of 20%, 25%, 50%, 67% or 73%, depending on the article. They are all quoted as if they estimate the same thing. They do not, and the gap between them is large enough to change a headcount decision.

This piece takes each number back to where it came from, says what it counts, and flags the two that could not be traced. The pattern behind them is real. The magnitude is what people get wrong.

What each junior developer hiring number counts

The 20% figure comes from a Stanford Digital Economy Lab paper by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, titled "Canaries in the Coal Mine?" It uses payroll records from ADP and finds that employment of software developers aged 22 to 25 was nearly 20% lower in July 2025 than in October 2022. That is a stock: people on payroll at a point in time. It says nothing directly about how many offers were made, and it uses age as a stand-in for seniority.

The 25% and "over 50%" figures come from the SignalFire State of Tech Talent report. It tracks new-graduate hiring at the 15 largest tech companies. Hiring fell about 25% from 2023 to 2024, and by more than 50% against 2019. Same dataset, same population, two baselines. The report also found that new graduates made up 7% of Big Tech hires, down from 15% before the pandemic. That last one is a share, not a count. A share falls when senior hiring grows faster than junior hiring, even if the number of juniors hired stays flat.

The 67% and 73% figures are different. One widely shared article attributes its 67% to an analysis of 118 companies in its own directory, with no published method. Another reports a 73% drop in actual hiring and sets it against a 47% rise in job postings labelled entry-level over a different window. Neither links to a primary dataset. They may be right, but nobody outside the authors can check.

Why the baseline year moves the answer

A decline is a comparison, and the starting point does most of the work. Late 2022 sits close to the top of the post-pandemic hiring run, so measuring from there captures the correction as well as anything caused by AI tools. Measuring from 2019 asks a different question: are firms hiring fewer new graduates than before the boom? Both are fair questions. They are not the same question.

This is why two true statements, "down 25%" and "down over 50%", can come from one report. A reader who sees only one of them walks away with a different picture of how fast things are moving.

What the Stanford paper says that the headlines drop

The paper makes a narrower claim than most coverage of it. Young workers in the occupations most exposed to AI saw a relative employment decline of about 13% compared with young workers in the least exposed ones. Older workers in the same exposed occupations kept growing. The authors also report that the pattern appears where AI is used to automate tasks and not where it is used to augment them, and that it holds up against firm-level shocks, which makes a pure interest-rate explanation harder to sustain.

That distinction matters for planning. A pattern tells you where to look. It does not give you a number to plug into a hiring forecast.

"Junior" is also defined three different ways

Before any of these numbers can be compared, someone has to decide who counts as junior. The Stanford paper uses age, 22 to 25. SignalFire uses new graduates. Job boards use the word "entry-level" in a title, and recruiters use years of experience. A 27-year-old with two years of experience is junior in a hiring manager's head and invisible in the age-based series.

Titles drift as well. Postings labelled entry-level reportedly kept rising while actual hiring into those roles fell, which suggests the label stopped matching the job. Some "entry-level" postings now ask for production experience with AI-assisted tooling, a bar that used to belong to mid-level hires. If the definition moves while you count, the trend line moves with it.

Commentary on the decline often says AI now handles the well-bounded tasks that once trained new engineers: small bug fixes, boilerplate, test scaffolding. That is plausible and some of the strongest evidence points the same way. It is also a mechanism, and mechanisms are harder to measure than headcount. None of the five numbers above tests it directly.

How to use these numbers when you are deciding on junior headcount

If you run an engineering team, you are not hiring at the 15 largest tech firms, and your own pipeline is the data that matters most. Three questions turn a statistic into something you can use.

• Is it a stock or a flow? Payroll headcount moves slowly and lags decisions. Hiring flow moves first.
• Whose population is it? Big Tech new-grad intake tells a ten-person startup little about its own candidate pool.
• What is the baseline? A decline from a peak and a decline from a trend line are different findings.

There is also a longer-run question that no current statistic answers. If fewer people start as juniors in 2026, the pool of mid-level engineers in 2029 is smaller. Teams that stop hiring juniors are borrowing from a future they will still have to staff.

What to watch next

The most useful development would be boring: more datasets that report flow, by company size, with stated baselines. Stanford maintains a dashboard that tracks its payroll series, and a second independent source covering mid-sized employers would do more for the debate than another round of round-number headlines.