A new Federal Reserve Board study adds evidence to a question now hanging over the software labor market: how much has generative AI changed demand for programmers? The answer is not a simple story of mass layoffs. It is a more specific shift in hiring patterns, concentrated in programming-heavy jobs and especially visible in parts of the economy that employ programmers at scale.
What the Study Measured
The study looks at monthly US employment data from a large household survey and compares it with an occupational database from the US Department of Labor. That database classifies jobs by their requirements and skills, allowing researchers to identify occupations with a high share of programming work.
Those programming-heavy occupations make up roughly 3.7 percent of all US workers. Before ChatGPT launched in November 2022, they were expanding at just under 5 percent a year, faster than the broader labor market.
Since then, the pace has dropped sharply. In sectors where programmers are especially common, including IT services and software development, growth has essentially flatlined.
The Slowdown Is Not Just a Tech Sector Story
A major question is whether the decline simply reflects broader pressure on technology companies. The source article notes several forces that hit the industry around the same time: interest rate hikes, the end of the Covid-era boom in online services, and the crypto crash.
To account for those broader forces, the researchers built a counterfactual employment curve. It estimated what programmer employment would have looked like if the share of programmers inside each industry had stayed constant while the size of those industries changed.
Even after that adjustment, programmer employment is still falling by about three percentage points a year. The study’s interpretation is that companies appear to be reducing the share of programmers in their workforces, not merely responding to shrinking industries.
The researchers also ran a control test using occupations barely touched by AI. That comparison did not show a similar dip, which strengthens the case that something specific is happening to programming-heavy work.
Why 500,000 Fewer Jobs Is Not the Same as 500,000 Lost Jobs
Over three years, the gap between expected and actual employment equals roughly 500,000 jobs that likely would have existed without the rise of large language models. But the authors warn against treating that figure as a direct count of job losses.
There are several reasons for caution:
- Some would-be programmers may have moved into adjacent fields.
- Programming tasks may be spreading into other job categories.
- The study does not capture broader macroeconomic feedback effects.
- If AI raises productivity enough, long-term labor demand could grow.
That distinction matters. The finding points to fewer filled programming-heavy jobs than expected, not necessarily a one-for-one displacement of workers from employment.
The wage data also complicates the picture. The researchers found no clear drop in wages. So far, the main effect appears in the number of jobs filled rather than in pay levels.
Job posting data from Indeed adds another layer. Software developer postings have been largely stable since 2024 and have ticked up slightly in recent months. Before that, postings had fallen by more than half in 2022 and 2023.
Where the Hiring Shift Is Most Visible
The typical programmer in this analysis is not necessarily working at a major tech company or startup. Around 40 percent of all US programmers work at IT service providers that build software on contract.
That contract software sector is the largest employer of programmers, and the slowdown is most pronounced there. This matters because a change in contract software hiring can show up as a broad labor-market signal, even if the public conversation focuses more on high-profile technology companies.
The timing is also important. According to the study, the gap between actual and expected employment did not open until mid-2024, roughly 1.5 years after ChatGPT launched.
If AI is responsible for the change, the delay suggests companies may have waited to see whether the models improved before changing hiring plans. The data does not show whether those decisions were driven by real productivity gains or by expectations that productivity would improve.
AI Is a Strong Candidate, but Not the Only One
The study does not claim that AI is the only possible explanation. A provision of the Tax Cuts and Jobs Act of 2017, which took effect in 2022, requires companies to amortize research expenses over several years instead of deducting them right away.
Because software development counts as research for tax purposes, that tax change may have cooled hiring. The source article notes that existing research on its actual impact is mixed, and that the study’s results hold up even in sectors where the tax change should matter less.
The authors also acknowledge that proving causality is difficult when several forces are moving at once. To test their method, they compare it with past disruptions. For bank tellers, the approach identifies ATMs as an occupation-specific shock even while banking grew overall. For seamstresses, it identifies an industry-wide shock from textile manufacturing moving overseas.
There is also a measurement problem across AI labor research. There is no standard method for deciding which occupations are most exposed to generative AI. Only about half of common approaches agree when identifying the most affected group.
Programmers stand out as an exception. More than 98 percent fall into the most-affected group across every measurement method. That matches the Anthropic Economic Index finding that programming-related queries account for more than a third of all interactions with the chatbot Claude.
The open questions remain large. The trend could reverse if cheaper programming services create new markets. Offshore workers may be affected differently. AI could also end up reshaping other occupations in ways that are not yet visible.
For now, the study frames its conclusion carefully. The authors describe the work as "only a first step" toward understanding how generative AI is changing employment. For programmers, though, the first step already points to a real hiring shift.