A recent Federal Reserve study has quantified a phenomenon that technology developers have been experiencing for two years. Fed economists Leland D. Crane and Paul E. Soto discovered that employment growth among U.S. programmers declined by approximately 50% following the launch of ChatGPT in November 2022. Previously, jobs requiring programming skills were increasing at about 5% annually, significantly outpacing overall labor market trends.
Post-launch, these growth rates have plummeted, particularly in sectors like IT services and software development, where they have nearly stagnated. This study is the first by the Federal Reserve to link AI adoption directly with a significant decrease in developer hiring, suggesting that AI has caused an occupation-specific shock.
In 2022, the tech sector faced challenges from rising interest rates, the end of the pandemic’s digital surge, and a cryptocurrency downturn. Skeptics have attributed the slowdown in developer employment solely to these factors. However, Crane and Soto accounted for this by creating a counterfactual scenario where programmer shares within industries remained constant. They found that programmer employment would still decline by about 3% annually even without those influences. Conversely, occupations not exposed to AI showed no similar downturn.
Over three years, the difference is estimated at around 500,000 jobs that might have existed absent large language models’ rise. The authors caution against interpreting this figure as a direct job loss count, noting many affected workers likely transitioned to related roles and broader economic feedback wasn’t captured. Nonetheless, the trend is evident.
The employment gap didn’t manifest until mid-2024, approximately 18 months after ChatGPT’s debut. The researchers propose that companies needed time to observe sufficient improvements in LLM capabilities before reducing their workforce. Whether these changes reflect actual productivity gains or merely anticipated ones remains unclear from the data.
Programmers are identified as the most AI-exposed occupational group in the U.S., consistent with usage statistics. For instance, Anthropic’s Economic Index indicates that tasks related to coding, debugging, and software architecture constitute about a third of all Claude.ai conversations and nearly half of enterprise API traffic.
Concerns also extend to future job pipelines. A Decrypt report noted accelerating AI-driven layoffs across white-collar sectors last year, with Anthropic CEO Dario Amodei warning up to 50% of entry-level roles might vanish within five years.
The Fed study lends institutional credibility to previously anecdotal evidence: a Harvard analysis of 62 million payroll workers found junior developer positions drop by about 9-10% within six quarters after companies adopt generative AI, while senior roles remain largely unaffected.
“If A.I. disproportionately impacts junior positions, it could have lasting effects on the college wage premium, upward mobility, and income disparities,” noted Harvard researchers.
Beyond the Fed, analysts are expressing concerns about tech job slowdowns due to AI-driven automation and replacement. A recent survey involving 69 economists, 52 AI experts, and 38 superforecasters showed widespread agreement that rapid AI advancements could reduce labor force participation, even among those who previously believed in ‘augmenting’ rather than replacing human roles.
The Fed researchers do not see the findings as disastrous. Programmer wages remain stable; the impact is evident in reduced headcount rather than salary decreases. Job postings have stabilized since 2024 and are slightly increasing. The authors suggest that AI-assisted programming could lower costs, potentially opening new markets and boosting overall demand for developer labor over time.
Crane and Soto regard their study as “only a first step.” Published as a preliminary designation, it has yet to undergo full Fed review. Nevertheless, it marks the Federal Reserve’s initial foray into this research area with substantial methodological rigor and institutional backing.