The AI employment gap for young workers has widened sharply. Employment among Americans aged 22 to 25 in occupations most exposed to AI is now 19% lower than in less-exposed fields, according to revised findings from Stanford’s Digital Economy Lab. A year ago, that gap was 13%.
The research, published in the August 2026 update of Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, draws on anonymised US payroll data from ADP covering millions of workers through June 2026. The authors — Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen — describe these as “early, descriptive indicators” rather than causal estimates.
What is the Stanford Canaries study? It is a recurring analysis by Stanford economists that tracks how generative AI is reshaping employment across age groups and occupations. The name borrows the coal-mine metaphor: young workers in entry-level roles are the first to show effects, serving as an early warning system for broader labour market changes.
The gap is getting bigger, not smaller
The headline number has moved steadily in one direction. When the researchers first documented the pattern in August 2025, the gap was smaller. By February 2026, it had grown. Now, with data through June, it stands at 19% — meaning employment for 22-to-25-year-olds in AI-exposed occupations would need to grow nearly a quarter just to catch up with their peers in less-exposed fields.
Since 2022, employment for this age group in the 40% of occupations most affected by AI has fallen about 11%. Over the same period, employment for young workers in the 60% of least-affected occupations rose 10%.
The divergence, the researchers note, persists even when excluding technology firms and computer occupations, controlling for interest-rate increases and remote work patterns, and across alternative measures of AI exposure. That robustness matters: it means the pattern is not simply a tech-sector restructuring story or a consequence of remote work normalisation.
Hiring is the mechanism, not layoffs
One of the study’s most counterintuitive findings is how the gap opens. It is driven primarily by weaker hiring of young workers — not by increased redundancies or people quitting.
Companies are not firing entry-level employees en masse. They are simply not bringing as many on board. That distinction matters enormously for anyone entering the workforce. If AI were triggering mass layoffs, the policy response would focus on severance, retraining, and displacement assistance. But if the mechanism is hiring suppression, the challenge is different: how do you get your first job when employers are filling entry-level roles with AI tools instead of people?
The study also finds that adjustment is occurring through employment levels rather than base compensation. Wages for young workers in AI-exposed roles have not cratered — they just can’t get through the door in the first place.
Substitution versus augmentation
The study distinguishes between occupations where AI replaces human work and those where it helps people perform their jobs better. That distinction turns out to be critical.
Roles like accountants, auditors, receptionists, and information clerks — where AI usage primarily substitutes for human tasks — show the weakest employment outcomes for entry-level workers. By contrast, occupations such as chief executives and registered nurses, where AI tends to complement human expertise, show flat or rising employment, especially for experienced workers.
The researchers also examined whether the type of knowledge a job requires affects its vulnerability. Occupations relying heavily on “codified” knowledge — formal, textbook-teachable information — had slower employment growth among entry-level workers. Jobs requiring “tacit” knowledge, developed through experience and mentoring, showed stronger growth among mid-career and senior workers.
The implication for anyone starting out: roles that reward judgement, relationships, and on-the-ground experience appear more defensible than roles built around processing standardised information.
The broader labour market is not collapsing
Here is where the study offers a corrective to alarmist narratives. The first of the six findings is explicit: “We find no evidence of widespread, economy-wide job displacement.” Across the workforce as a whole, there is little difference in employment between highly AI-exposed occupations and less-affected ones.
Challenger, Gray & Christmas reported in August that US employers announced 33,429 job cuts in July 2026 — the lowest monthly total in two years. AI led all reasons for the fifth consecutive month at 10,970 cuts, but announced hiring plans rose to 16,095, the highest July total since 2022. Through July, companies have announced plans to hire 107,500 workers, up 25% from the same period in 2025.
The picture, then, is not one of AI dismantling the labour market. It is one of AI quietly rerouting the entry ramp — making it harder for young people to get their first foothold in certain occupations while leaving experienced workers and the broader jobs market largely intact.
What this means for New Zealand
Stanford’s data is US-specific, but the pattern has obvious relevance for New Zealand’s labour market. NZ’s tech sector has been vocal about skills shortages, yet young graduates increasingly report difficulty landing first roles. Stats NZ’s most recent labour force data shows youth unemployment running above the national average — a trend that predates AI but could be amplified by it.
New Zealand’s heavy reliance on imported AI tools — most companies here consume American or European models rather than building their own — means the hiring-suppression pattern documented at Stanford could arrive here through the same channel. If US firms reduce entry-level hiring because AI handles routine tasks, NZ firms using those same tools may follow suit, even if the local labour market context differs.
The career advice that flows from this is straightforward. As we noted in our earlier coverage of entry-level hiring trends, young workers should seek roles where AI augments rather than replaces — and where tacit knowledge compounds with experience.
❓ FAQ
Is AI causing mass unemployment? No. The Stanford study finds no evidence of economy-wide job displacement. The effect is concentrated in entry-level hiring for specific occupations where AI substitutes for human tasks.
Which jobs are most affected for young workers? Roles involving codified, standardised knowledge — accountants, auditors, receptionists, information clerks. Jobs requiring tacit knowledge and human judgement show stronger employment growth.
What should new graduates do? Seek roles where AI complements human expertise rather than replacing it. Prioritise positions that build tacit knowledge — judgement, relationships, real-world experience — which appears to compound with time and is harder for AI to replicate.
Does this affect New Zealand? The data is US-specific, but the mechanism — companies reducing entry-level hiring as AI tools handle routine tasks — could replicate wherever those same tools are adopted, including New Zealand.
🔍 THE BOTTOM LINE
The Stanford findings are not a forecast of AI-driven mass unemployment. They are something more specific and arguably more insidious: evidence that AI is quietly closing the door on entry-level hiring in certain occupations, widening a gap that has grown from 13% to 19% in a single year. The mechanism is hiring suppression, not layoffs — which means the damage accumulates silently, one unbumped starter at a time. For anyone entering the workforce, the lesson is to aim for roles where experience compounds and AI serves as a tool rather than a replacement.
