Fed: AI tied to 500,000 missing U.S. programmer jobs

A Federal Reserve study by economists Leland D. Crane and Paul E. Soto found that the launch of ChatGPT and subsequent adoption of large language models corresponded with a roughly 50% drop in U.S. programmer job growth, producing an estimated shortfall of about 500,000 developer roles that emerged around mid-2024.

The authors compared actual employment in programming-intensive occupations with a counterfactual scenario in which the share of programmers in each industry remained constant. Before November 2022, programming jobs grew at about 5% a year. After ChatGPT’s release, growth in those occupations fell sharply. In industries with high concentrations of developers, such as software development and IT services, hiring largely flatlined.

Stretched over three years, the gap between actual and counterfactual employment amounts to roughly 500,000 positions that likely would have existed without the rise of large language models. The researchers caution that the figure is not a literal count of lost jobs, since some workers may have moved into adjacent roles and the analysis does not capture broader macroeconomic feedback.

The employment shortfall did not appear immediately. The study finds the gap opened in mid-2024, about 18 months after ChatGPT’s debut, which the authors interpret as firms waiting to see improvements in model capabilities before altering hiring plans. The data do not determine whether the change reflects actual productivity gains from AI tools or employer expectations about future productivity.

The study tested for other explanations, including broader tech-sector weakness after the 2022 market downturn, higher interest rates and the end of the pandemic tech boom. After accounting for those factors, programmer employment still fell by about 3% per year relative to the counterfactual. Occupations judged to be less exposed to AI did not show a similar pattern.

The decline shows up in hiring rather than pay. Programmer wages have not fallen, and job postings stabilized in 2024 and have inched up since then. The authors write that cheaper, AI-assisted programming may eventually create new markets and raise long-term demand for developers even as it reduces hiring in some roles today.

Enterprise usage data cited in the paper show that coding, debugging and software-architecture tasks make up a large share of developer interactions with language models and application programming interfaces, consistent with programmers being among the most AI-exposed occupational groups.

A separate academic study of payroll records found junior developer employment falls roughly 9 to 10% within six quarters after firms adopt generative AI, while senior developer headcount changes little. The researchers noted potential implications for entry-level opportunities and career mobility.

The authors describe the paper as “only a first step.” The study carries a preliminary designation and has not completed the Federal Reserve’s full review process.