Shaduf.Research preview
Will AI Take My Job?/Junior software ads fell relative to senior ads. The AI share is unknown.

Research update · 1 October 2026

Junior software ads fell relative to senior ads. The AI share of that fall is unknown.

A June working paper found roughly 14–15% fewer U.S. junior software-developer vacancies relative to senior vacancies in the year after ChatGPT’s November 2022 release. It measures online job ads, not people hired or laid off.

What the paper measured

Westby, Modestino and Cheng used Lightcast U.S. postings for software developers. Their main comparison covers 12 months before and 12 months after November 2022. A junior ad requires at most three years of experience; a senior ad requires four or more. Across 575,397 underlying ads in the main regressions, junior vacancies fell roughly 14–15% relative to senior vacancies. Both groups' demand cooled after the pandemic hiring surge. The result is a relative gap, not a claim that exactly 15% of all junior jobs disappeared.

The bar also moved inside job titles

The authors find that required experience rose within some nominally similar software titles. Remaining junior ads shifted more toward problem solving, communication and attention to detail than toward AI-specific skills. Comparisons with other technical occupations and a mechanical-engineering placebo make a software-specific shift more credible. They do not reveal how many postings became hires.

What the result cannot prove

The public release of ChatGPT is a time marker, not a verified treatment for each employer. It overlapped a wider software hiring slowdown. The authors' comparisons reduce some rival explanations but cannot isolate AI adoption from every concurrent change, including remote-work and market adjustments. The main analysis excludes ads without a stated experience requirement; a classification check finds a similar pattern but cannot make that missing information certain. The paper cannot tell a worker their odds of losing a job, or show a national unemployment effect.

Other evidence remains mixed: U.S. graduate and payroll studies report narrow early-career warnings, while a broad U.S. tracker has not found a clear economy-wide AI jobs footprint. Those studies measure different people and outcomes. A junior vacancy is not a hire; a national unemployment rate can miss a narrow entry barrier.

A useful next question

Before acting on “AI killed junior jobs,” ask whether the claim counts advertised openings or actual hires, which employer and period it covers, and whether AI adoption was observed. For your own workplace, ask whether tasks, experience requirements or hiring plans have changed. Do not buy training or change career plans on this estimate alone.

Search published pools, pages, reports, and evidence.