A new update from Stanford University economists suggests that the impact of AI on employment is becoming increasingly visible among young workers entering the job market.
According to the researchers, employment among workers aged 22 to 25 in occupations most exposed to AI is now 19% lower than in less AI-exposed fields. That gap has grown from 13% last year, indicating that the trend is not only continuing but widening.
The study, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” analyzes anonymized payroll data from ADP and combines it with measures of AI exposure, including data from Anthropic's Economic Index.
Entry-level workers are taking the biggest hit
Across the economy as a whole, researchers found relatively little difference in employment between occupations that are highly exposed to AI and those that are less affected.
The picture changes dramatically when looking specifically at workers aged 22 to 25.
Since 2022, employment among young workers in the 40% of occupations considered most exposed to AI has fallen by roughly 11%. Meanwhile, employment among young workers in the 60% of occupations least exposed to AI has increased by about 10%.
The researchers say the decline appears to be driven primarily by reduced hiring, rather than widespread layoffs or workers voluntarily leaving their jobs.
That distinction is important: AI may not yet be eliminating large numbers of existing positions, but companies appear to be hiring fewer inexperienced workers for some tasks that can increasingly be handled by AI.
Automation matters more than AI assistance
Not every use of AI has the same effect on employment.
Anthropic distinguishes between AI used for automation, where the technology performs tasks previously handled by humans, and augmentation, where AI helps people perform their existing work more effectively.
The Stanford researchers found that occupations with heavier AI automation are experiencing the sharpest declines in entry-level employment.
Accounting and auditing, for example, are among the occupations particularly susceptible to automation. By contrast, jobs such as nursing and executive positions tend to involve more AI augmentation, where the technology supports rather than replaces human workers.
The researchers argue that this difference helps explain why some AI-exposed occupations are shrinking while others remain stable or even grow.
AI may be closing the career ladder's first rung
The researchers also looked at the type of knowledge required for different occupations.
Jobs based heavily on codified knowledge—formal information that can be documented, standardized and taught through textbooks or written procedures—appear to be particularly vulnerable at the entry level.
AI systems are well suited to this kind of work because much of it can be described, standardized and reproduced digitally.
By contrast, occupations that depend more heavily on tacit knowledge—skills developed through experience, mentorship and repeated exposure to real-world situations—appear to provide greater advantages to experienced workers.
This creates a potentially important divide. Experienced employees may remain valuable because they possess judgment and practical knowledge that AI cannot easily reproduce, while inexperienced workers may struggle to get the opportunity to develop those same skills.
Education may provide some protection
The Stanford researchers also found evidence that higher education could reduce the employment gap between AI-exposed and less-exposed occupations.
Occupations with a larger proportion of college graduates showed smaller differences in employment outcomes between AI-exposed and less-exposed fields.
The researchers suggest that education may provide workers with broader skills that make them more adaptable as AI changes individual tasks and occupations.
However, the findings do not mean that a college degree guarantees protection from AI disruption. Rather, the data suggest that workers in occupations requiring more education may currently be somewhat less exposed to the employment effects being observed among younger workers.
The bigger concern: the disappearing on-ramp
The most significant implication of the research may not be mass unemployment. Instead, it could be the gradual disappearance of entry-level opportunities.
If companies use AI to perform tasks traditionally assigned to junior employees, organizations may need fewer people at the bottom of the career ladder.
That creates a difficult long-term problem. Young workers typically gain expertise by starting with relatively simple tasks, learning from experienced colleagues and gradually taking on more responsibility. If those initial jobs disappear, future workers may have fewer opportunities to acquire the experience needed for senior positions.
Stanford economist Erik Brynjolfsson therefore argues that the current situation deserves particular attention. Overall employment can remain relatively stable while the labor market quietly becomes harder to enter for people beginning their careers.
The emerging picture is not necessarily an immediate “jobs apocalypse.” It is potentially more subtle—and, for young workers, more consequential: AI may be changing who gets the opportunity to start a career in the first place.