AI is already making a dent in the job market, but not all kinds of jobs are being equally impacted.
A new working paper from Stanford University provides some of the first large-scale, high-frequency evidence that the generative AI revolution is beginning to have a significant and disproportionate impact on the American labor market, particularly for entry-level workers.
The paper, titled “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” was authored by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen and released on August 26, 2025. The researchers leveraged a massive, near real-time administrative dataset from ADP, the largest payroll provider in the United States, to analyze employment trends. Their methodology involved tracking millions of individual-level monthly payroll records through July 2025 and linking them with established measures of an occupation’s exposure to AI. This allowed for a granular analysis of how the workforce is shifting since the widespread adoption of generative AI tools like ChatGPT in late 2022.
The study presents six key facts that paint a clear picture of the initial labor market adjustments to AI.
The Six Key Findings
The study found a substantial decline in employment for early-career workers (ages 22-25) in occupations most exposed to AI, such as software development and customer service. For example, the study’s data shows that by July 2025, employment for software developers aged 22-25 had fallen by nearly 20% from its peak in late 2022. In stark contrast, employment for more experienced workers in the same fields, and for workers of all ages in less-exposed jobs like health aides, remained stable or continued to grow.
Additionally, while the overall U.S. job market remains robust, employment growth specifically for young workers has been stagnant since late 2022. The study demonstrates that this tepid growth is being driven by job declines in AI-exposed occupations. In roles with high AI exposure, employment for workers aged 22-25 fell by 6% between late 2022 and July 2025, while it grew by over 9% for workers aged 35-49 in the same fields.
The study also found differences between AI that automates versus AI that augments. The employment declines are concentrated in occupations where AI is more likely to automate or substitute for human labor. Conversely, in fields where AI primarily augments or complements human skills, the researchers found that employment for young workers has actually grown. This distinction was made using data from Anthropic’s AI model, Claude, which helped classify whether AI usage in a given task was substitutable or complementary.
The study also showed these trends are not simply the result of broader industry shocks, like a downturn in the tech sector or rising interest rates. Even after controlling for firm-specific events that would impact all employees, the researchers found a 13% relative decline in employment for young workers in the most AI-exposed roles compared to those in the least exposed roles. This indicates the effect is tied to the specific job function, not just the health of the company or industry.
For now, the labor market adjustments are appearing in hiring numbers rather than paychecks. The researchers found little difference in annual salary trends by age or AI exposure, suggesting that wages may be “sticky” and have not yet adjusted to these employment shifts. This implies that, at least initially, AI’s impact is being felt more through job availability than through compensation.
These findings were robust across various scenarios. The patterns hold even when technology-related firms and occupations are excluded from the data. They are also consistent in occupations that are not amenable to remote work, suggesting the trend is not solely driven by outsourcing. The researchers confirmed that these sharp, divergent trends began acutely in late 2022, coinciding with the explosion of generative AI tools.
Why Younger Workers?
The paper posits a compelling reason why AI is hitting entry-level workers the hardest. Younger workers’ primary advantage in the labor market is often their recently acquired “book-learning” or codified knowledge. Generative AI, by its very nature, excels at replicating this type of knowledge. In contrast, more experienced workers possess tacit knowledge—the idiosyncratic skills, intuition, and wisdom gained through years on the job—which AI is far less capable of replacing. This makes senior workers less vulnerable to substitution by AI tools, at least for now.
While the authors caution that other economic factors could be influencing these trends, their results are strongly consistent with the hypothesis that generative AI has begun to significantly reshape the job market for those just starting their careers.