A new study from Stanford University provides some of the first large-scale evidence that the rise of artificial intelligence is having a tangible and disproportionate impact on the American labor market, with entry-level workers bearing the initial brunt of the shift. The research, which analyzed millions of U.S. payroll records, reveals that workers between the ages of 22 and 25 have experienced a significant 13% relative decline in employment since 2022, but only within occupations most exposed to AI technologies.
The report identified specific roles that are highly susceptible to AI-driven disruption, including customer service representatives, accountants, and software developers. In stark contrast, employment levels for more experienced workers in these same fields have remained steady or even grown. The study also found that jobs for workers of all ages in less-exposed occupations, such as nursing aides and front-line production supervisors, have shown resilience and continued growth, indicating that the AI revolution’s effects are far from uniform across the economy.
Researchers took meticulous steps to rule out other potential factors that could have skewed the data, including educational attainment, the rise of remote work, job outsourcing, and broader economic trends.
The findings suggest that the unique vulnerability of young workers may stem from AI’s current capacity to automate tasks that rely on “codified knowledge” or “book-learning” typically emphasized in formal education and foundational for entry-level roles. Conversely, the intangible expertise and nuanced decision-making skills gained through years of on-the-job experience appear to be more difficult for AI to replicate in the near term.
The study further clarifies that not all AI integration leads to job loss. In roles where the technology is used to augment and complement human skills, thereby improving efficiency, the changes in employment rates have been far more muted.
The significant adjustments in the labor market are primarily occurring through employment numbers rather than changes in wages or compensation. According to the researchers, these trends help explain why national employment growth for young workers has stagnated even as the overall U.S. job market has remained robust following the global pandemic.
Of critical importance, “The study makes a clear distinction between different applications of artificial intelligence, noting that not all uses are associated with declines in employment. The researchers directly state that entry-level employment has specifically declined in applications where AI is used to automate work, but not in those where its primary function is to augment human capabilities. This crucial difference was measured empirically by analyzing whether queries to a large language model, like Claude, were likely to substitute for or complement the core tasks of a given occupation. The findings confirm that while young workers face employment declines in roles where AI automates their work, they can actually experience employment growth in occupations where AI use is primarily augmentative.”
“Furthermore, the researchers rigorously tested to ensure these trends were not being driven by broader economic factors. A key concern was that the patterns could be explained by industry- or firm-level shocks—such as changes in interest rates—that might coincidentally correlate with a firm’s age demographics and its level of AI exposure.”
“To rule this out, the study controlled for firm-time effects, thereby isolating and absorbing any aggregate shocks that would impact all workers within a specific company, regardless of their role’s exposure to AI. Even after accounting for these firm-wide influences, the data showed a pronounced 12 log-point decline in relative employment for workers aged 22-25 in the most AI-exposed roles. This large and statistically significant effect stands in stark contrast to the much smaller and statistically insignificant estimates for all other age groups, providing strong evidence that the employment trends are a direct result of AI exposure and not merely a byproduct of broader company-level struggles.” Source: Stanford University
