The Quiet Replacement: how AI is rewriting the first chapter of careers

The spread of generative artificial intelligence has moved beyond hype and headlines; it is now showing up in payroll data. A new study from the Stanford Digital Economy Lab, Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, traces the first measurable labour market ripples of the AI era. Using data from millions of workers across thousands of U.S. firms, the study paints a portrait of subtle but consequential shifts already underway in how companies hire and who gets hired.

The evidence points to a labour market that is beginning to bifurcate: overall job levels remain stable or rising, but younger workers in AI-exposed roles are being quietly squeezed out. The staffing industry, which acts as an early sensor for employment change, stands at the frontline of this adjustment.

Early signals of disruption

The Stanford researchers—Erik Brynjolfsson, Danielle Chandar, and Jiayi Chen—link detailed payroll records from ADP with occupational measures of “AI exposure,” estimating how susceptible each job is to automation or augmentation by generative tools. Their findings reveal six consistent facts.

Employment among younger workers, particularly those aged 22 to 25, has declined sharply in occupations that overlap heavily with generative AI capabilities. In high-exposure roles such as software development, marketing, and customer service, employment for that age group dropped by roughly 13 percent compared with less-exposed jobs between 2022 and 2025. For staffing markets, this represents an inflection point: the traditional entry-level hiring engine (the steady stream of graduates and early-career professionals) is losing momentum in precisely the segments most open to automation.

A mixed picture beneath aggregate growth

Despite these declines, overall employment continues to rise in many sectors. Older and more experienced workers in the same AI-exposed occupations are, in some cases, seeing modest gains. Employment for individuals aged 35 to 49 in those same roles increased by around 9 percent during the same period.

This dual movement (fewer young entrants, more retention of seasoned staff) suggests that companies are adjusting to AI not by wholesale layoffs, but by rebalancing their workforce composition. Experience and contextual judgment are becoming premium assets in occupations where machines now perform a growing share of routine tasks.

Automation versus augmentation

The study distinguishes between jobs where AI substitutes for human labour and those where it augments it. Employment declines among younger cohorts are concentrated in the substitution-heavy group (roles where AI performs large portions of work independently). In occupations where AI complements workers, productivity gains appear to be sustaining or even expanding employment.

This distinction underscores the need to think of “AI exposure” as multidimensional. The presence of AI technology does not automatically imply contraction; rather, the outcome depends on how the technology reshapes the workflow. In sectors where AI amplifies output without erasing the human role, demand for labour may persist or rise.

Structural, not cyclical, change

By controlling for firm-specific and time-specific effects, the authors show that the decline in young worker employment persists even after accounting for industry cycles or economic shocks. This robustness indicates a structural adjustment rather than a temporary downturn.

For the staffing ecosystem, that structural shift manifests as a gradual thinning of entry-level requisitions in AI-exposed functions, counterbalanced by sustained or growing demand for mid-career profiles. Recruiters and workforce planners are likely to observe longer time-to-fill for experienced roles, even as early-career hiring volumes stagnate.

Headcount over wages

Another notable feature of the transition is that adjustment appears to be happening through employment rather than pay. Wage growth for AI-exposed roles has not diverged significantly from that of less-exposed roles, suggesting that firms are responding to technological change by reducing hiring rather than lowering compensation.