AI Eliminates Entry-Level Jobs While Sparing Senior Workers

The career ladder is breaking at the bottom

Artificial intelligence is not eliminating jobs uniformly across the workforce. Instead, it is systematically removing the entry points that young workers have historically used to begin their careers, according to continuously updated data from a collaboration between Stanford economist Erik Brynjolfsson and payroll provider ADP.

The Canaries Dashboard, launched this month by Brynjolfsson's Stanford Digital Economy Lab and ADP Research, tracks employment patterns across 4.6 million workers in more than 730 occupations. The data covers roughly one in six American workers and reveals a widening gap between how AI affects workers at different career stages.

For workers ages 22 to 25 in highly AI-exposed occupations, employment is now contracting at 3.8% annually as of April 2026. That decline has accelerated over time—from a 2.8% decrease through April 2024 to more than 4% per year since then. Meanwhile, the least AI-exposed jobs in that same age group are growing at 2% annually.

The aggregate numbers mask this divergence. Across all workers, the most AI-exposed occupations contracted just 0.2% year over year, while the least-exposed roles grew 0.1%. These headline figures have fueled skepticism about AI's labor market impact, with critics attributing employment shifts to interest rates, tech-sector overhiring, or pandemic distortions.

Brynjolfsson has systematically tested those alternative explanations. He removed the entire tech sector from the analysis. He isolated remote-work effects and controlled for interest rate sensitivity. The pattern persisted in every case, he told Fortune.

Why junior workers are vulnerable

The mechanism behind this age-based divergence is straightforward. AI absorbs tasks before it eliminates entire jobs, and the tasks it handles first are those requiring the least experience: information retrieval, summarization, scheduling, formatting, and basic data assembly. These responsibilities fall disproportionately to early-career workers.

Senior workers have accumulated job-specific skills and contextual knowledge that remain difficult to automate. Junior workers have not yet built those defenses.

ADP chief economist Nela Richardson, Brynjolfsson's research partner, frames the distinction as automation versus augmentation. Occupations where AI augments human capabilities show sustained employment growth. Those where AI automates tasks outright show contraction. Early-career workers concentrate in the automatable layer.

Mid-career workers ages 31 to 34 also show contraction, down 1.7% year-over-year. Workers ages 35 to 40, by contrast, are growing at 2%.

Why it matters

The disappearance of entry-level positions has implications beyond immediate employment statistics. These roles have traditionally served as training grounds where workers develop the judgment, relationships, and tacit knowledge that define mid-career competence. If AI eliminates that on-ramp, organizations may face a future shortage of experienced workers—not because AI replaced them, but because they never entered the pipeline. The effect also raises questions about economic mobility and whether traditional career progression models remain viable in an AI-augmented economy.

The productivity debate continues

Brynjolfsson's findings have intensified his ongoing debate with MIT economist and Nobel laureate Daron Acemoglu, who produces far lower AI productivity estimates. The two have been exchanging arguments for months, seeking common ground on how AI should complement rather than replace workers while disagreeing sharply on magnitude and timeline.

Acemoglu recently told Fortune that much of the AI productivity discourse is speculative to the point of being "brainless." Brynjolfsson counters that the data increasingly supports his position. He has a public 10-year wager with Northwestern economist Bob Gordon predicting significantly higher productivity by decade's end.

The dashboard extends Brynjolfsson's original research, published last August, which drew on ADP's administrative dataset to document AI's early labor market effects. That paper sparked immediate pushback from Google economists and others who attributed the patterns to non-AI factors. The new dashboard, updated through April 2026, shows the effect has not reversed. It has grown by roughly half a percentage point per month.

Brynjolfsson compares the current disruption to the Industrial Revolution—the last time humanity built machines that fundamentally changed work and productivity. The difference, he argues, is speed. This transformation will be larger and ten times faster.

These findings were first reported by Fortune.