New employment research suggests artificial intelligence may be changing how young professionals enter the workforce long before mass layoffs appear. The evidence is troubling—but not yet conclusive.
For years, entry-level jobs have served two purposes: completing necessary work and teaching inexperienced employees how to become experts. Artificial intelligence challenges that arrangement because many of the tasks easiest to automate—drafting routine documents, preparing basic analyses and writing straightforward code—are also assignments through which graduates learn.
The risk is not simply that AI replaces junior workers. It is that companies could gradually hire fewer beginners while continuing to depend on senior professionals whose expertise took years to develop.
Recent research suggests this possibility deserves attention. But it also shows why claims that AI has already caused a broad employment crisis are premature.
Young Workers Are Falling Behind in AI-Exposed Occupations
In an August 2026 research revision, Stanford Digital Economy Lab researchers Erik Brynjolfsson, Bharat Chandar and Ruyu Chen examined ADP payroll records covering millions of US workers through June.
They reported that employment among 22- to 25-year-olds in highly AI-exposed occupations was approximately 19% below where it would have stood if it had grown alongside employment for similarly aged workers in less-exposed occupations. That relative gap widened from 15% in July 2025.
Crucially, this does not mean 19% of young workers lost their jobs. The researchers found that reduced hiring, rather than increased separations, appeared to explain much of the divergence. They found no comparable gap among experienced workers and no evidence of widespread displacement across the economy.
The authors characterized their findings as descriptive, not causal. Changing recruitment practices, economic conditions and other structural forces could also contribute.
The Consequences May Begin Before the First Paycheck
A September 2026 US Census Bureau working paper by Cody Orr, Lee C. Tucker and Lawrence Warren examined employment outcomes for graduates from different college majors.
Its regression-adjusted estimates found that graduates from the most AI-exposed tenth of majors experienced a five-percentage-point decrease in their likelihood of initial employment and a 13% decline in initial full-quarter earnings following the arrival of widely available large language models.
Roughly half the earnings decrease reflected lower earnings within employing industries. The remainder resulted from graduates moving into lower-paying sectors, including restaurants and retail.
The association does not establish AI as the sole cause. Still, it raises a consequential possibility: some graduates may struggle to obtain the first occupation-specific role that enables their careers to develop.
Why the First Rung Is Especially Exposed
Stanford researchers distinguished between codified knowledge, documented in procedures, textbooks and structured tasks, and tacit knowledge, acquired through practical judgment, mentorship and experience.
Entry-level work frequently depends on codified knowledge. An AI system can generate an initial software function, summarize a contract or prepare a preliminary financial comparison. Experienced employees must still determine whether those outputs make sense in context.
That difference may help explain why automation affects younger workers differently. Employers can reduce routine assignments without eliminating senior review, client relationships or accountability.
But this creates a longer-term problem: if juniors rarely prepare initial drafts, debug basic mistakes or observe difficult decisions, how will they acquire the expertise needed to supervise AI later?
That remains an inference about possible organizational consequences, not an outcome already demonstrated by employment data.
Other Researchers Find a Less Dramatic Picture
The Yale Budget Lab’s October 14 analysis found no clear AI-related change in the occupational composition of US employment. Measures of AI usage showed no connection to aggregate employment or unemployment changes, while its statistical analysis found no unambiguous AI labor-market footprint.
Similarly, Anthropic’s March research identified no systematic rise in unemployment among highly exposed workers, although its findings suggested hiring might have slowed for younger entrants.
These studies examine different datasets and outcomes. Economy-wide stability can coexist with weaker hiring in particular occupations or age groups. Neither observation proves that AI alone caused the changes.
The International Labour Organization’s June evidence review likewise found limited large-scale displacement while highlighting uneven productivity gains, inequality and declining employment opportunities for younger workers.
AI Could Also Make New Workers Better
Automation is not the only possible outcome. In a peer-reviewed study of 5,172 customer-support agents, Brynjolfsson, Danielle Li and Lindsey Raymond found that AI assistance increased issues resolved per hour by 15% on average.
Less experienced and lower-skilled agents benefited most, with productivity improvements of approximately 30%, alongside evidence of learning and better service quality.
The distinction is between replacing beginner tasks and helping beginners perform them. A company using AI to coach employees may accelerate development; one using it primarily to avoid hiring could weaken its future expertise pipeline.
What Builders and Employers Need to Measure
For startup founders and engineering leaders, the immediate question is not simply whether an AI agent can complete a junior developer’s assignment. It is whether automating that assignment improves the business after accounting for review costs, errors, mentoring and future recruitment needs.
That suggests tracking junior hiring, skill development and promotion alongside automation savings. Apprenticeships, supervised code reviews and clearly defined ownership of AI-generated work could preserve learning opportunities while retaining productivity gains.
For investors, the concern extends to whether organizations can sustain experienced talent if they stop developing it internally. Short-term operating efficiency and long-term workforce resilience are not automatically aligned.
The available evidence does not justify predicting the disappearance of entry-level careers. It does justify examining whether AI adoption is quietly changing who gets to begin one—and whether technology companies are preparing the people who will eventually be expected to oversee the systems they deploy.
