The 'junior programmer' job is shrinking. The pipeline may not be.
Stanford's Canaries dashboard records a 19% ADP payroll drop for US software developers aged 22 25 since October 2022, while older cohorts grew roughly 14%.
Stanford's Canaries dashboard records a 19% ADP payroll drop for US software developers aged 22 25 since October 2022, while older cohorts grew roughly 14%.
The data says young programmers are disappearing from US payrolls. The reality may be stranger: they never stopped working. They just stopped calling themselves programmers.
That is the tension at the heart of the Stanford "Canaries" dashboard and the industry essay that has driven the conversation since August 2025. The headline number is real and specific: from an October 2022 peak, the Stanford Digital Economy Lab's "Canaries in the Coal Mine?" paper (Brynjolfsson, Chandar and Chen, 2025) records a roughly 19% decline in ADP payroll for US software developers aged 22-25, against a roughly 14% rise for the 41-49 cohort. After firm-level shock controls, the relative gap settles around 16 percentage points. Tier-1 outlets including Fortune, TIME and Observer treated the finding as the first clean empirical handle on AI's labor-market footprint.
Two things can be true at once. The cohort-divergence pattern is the chart that industry reads such as SoftwareSeni's analysis of the same ADP series call the cleanest signal so far on AI and white-collar employment. At the same time, the dashboard measures one specific quantity: how many people aged 22-25 sit in jobs tagged "software developer" in the ADP payroll file. That taxonomy was not built for a labor market in which a 23-year-old ships code that runs in production, builds agents in a notebook, and calls themselves a "builder," an "AI engineer," or simply "in tech."
Laurie Voss, the npm and Socket co-founder whose essay has been the most circulated industry take, reads the Stanford finding as a pipeline collapse plus a redefinition: "the long tail of new programmers has materialized, but with a big twist — they don't call themselves programmers." It is the most quoted line in the Hacker News thread on his post, where working developers describe their own companies pausing junior engineering hires over the past twelve months and several younger commenters describe doing the same work under a different title. Both data types are useful: the hiring pauses are real, but they are anecdotes, not measurements, and they fit either the demand-shock reading or the title-shock reading.
That distinction matters because the broader public reading has become "AI is killing entry-level white-collar work." That is the Stanford authors' own generalization, not a finding of the canary framework. The paper flags disproportionate early-career exposure across white-collar knowledge work as a hypothesis, not a confirmed mechanism. Inside software specifically, the 22-25 cohort is the canary; the rest is extrapolation. A more defensible reading is that AI is compressing the entry rung: fewer titled junior roles, more fluid work, and a title churn that ADP's Standard Industrial Classification codes do not pick up.
There is a counterforce the headline writers tend to skip. ADP payroll is private-sector data; it undercounts contractors, founders, and anyone paid outside the W-2 system, which is exactly where the youngest cohort is over-represented. The Stanford team controls for firm-level shocks but cannot fully account for title migration inside a firm. If ten people aged 22-25 stop being coded as "software developer" and start being coded as "AI engineer" or "machine learning engineer," both common rebrandings across 2024-2025, the dashboard records a -10 for software, not a -19, and a matching bump somewhere the canary is not watching.
The watch item worth tracking through 2026 is whether the next Canaries refresh shows the 22-25 cohort stabilizing, recovering, or simply reappearing under a different series. The Stanford team has signaled an annual update; the next read is the test of whether the 2022-2025 drop was a demand shock, a title shock, or both at once.