Stanford Tracked Millions of Paychecks. AI Isn't Firing People. It's Quietly Not Hiring Them.

Via Stanford Digital Economy Lab: Canaries in the Coal Mine, August 2026 update (paper, PDF)

The study everyone cites about AI and young workers

"Canaries in the Coal Mine" is the research from Stanford's Digital Economy Lab by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen. It uses payroll records from ADP covering millions of US workers to track employment by age and by how exposed a job is to AI. The latest version runs the data through June 2026.

It's worth reading carefully, because it gets quoted in two opposite ways: as proof AI is wiping out jobs, and as proof nothing is happening. It says neither.

What the update found

• No widespread displacement. The authors don't find broad job losses across the workforce.
• Young workers in exposed jobs are falling behind. For workers aged 22–25 in the most AI-exposed occupations, employment is about 19% below trend, up from about 15% in the July 2025 version of the analysis.
• It's happening through hiring. The gap comes from fewer people being hired, not more people being let go.
• It depends on how AI is used. The decline is concentrated in jobs where AI tends to automate tasks. Where AI mainly complements workers, employment is flat or rising.
• Headcount moves, pay doesn't. The adjustment shows up in employment, not in base pay.

The paper also splits jobs by the kind of knowledge they rely on. Employment fell for young workers in jobs built on codified knowledge, the kind you can write down in a manual or textbook, and rose for experienced workers in jobs built on tacit knowledge, the judgement you pick up by doing. In levels, employment in the two most exposed quintiles of occupations fell about 11% between November 2022 and June 2026, while the three least exposed quintiles grew about 10%.

What it doesn't prove

The authors are explicit that these patterns are descriptive, not causal. They line up with where AI is used, but the study doesn't prove AI caused them. The authors run checks such as controlling for occupations' interest-rate exposure and excluding tech-sector firms, but the design can't rule out every other cause, and they note the gaps shrink when they control for education. Anyone telling you this study shows "AI killed entry-level jobs" is claiming more than the researchers do.

Why an owner should care anyway

Even as a descriptive pattern, the study points at a real choice firms are making, mostly without discussing it: as AI handles the codified parts of junior work (first drafts, data entry, standard replies), many firms simply hire fewer juniors. That can look like savings this year. It also cuts off the route by which people pick up the tacit knowledge that, in the same data, is where experienced workers are doing well. A firm that stops training juniors is borrowing against its future senior staff.

The study's automate vs complement split is also a useful lens for your own plans. Using AI to remove a task entirely and using it to make a person faster at that task lead to different staffing outcomes, and different risks if the output is wrong.

Five questions to ask before you cut a junior role

1. Is this task codified or tacit? Codified work (a template, a checklist) is where AI tends to automate. Tacit work needs a person who has learned from experience. The AI use case scorer helps you sort tasks.
2. Who checks the output? Juniors often do the first review. Take them out and the checking moves to your most expensive people. Use the hallucination check to see how much review a task needs.
3. What does it really save? Work out cost per finished task, review time included, with the AI ROI calculator.
4. Where will your next seniors come from? If the answer is "we'll hire them", check that the market will still be producing them.
5. Can AI make the junior faster instead? That's the complement pattern, where employment held up in the study's data. Our AI literacy quiz is a quick way to see where your team is.

For the wider picture on what AI is and isn't delivering, read AI hype vs reality, the MIT AI report and McKinsey's State of AI 2025, or check where you stand with the AI readiness assessment.

Frequently asked questions

What is the Stanford Canaries in the Coal Mine study?

Research by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at Stanford's Digital Economy Lab that uses ADP payroll data to track employment by age and AI exposure. The August 2026 update runs through June 2026.

Is AI taking entry-level jobs?

The study finds workers aged 22–25 in highly AI-exposed jobs are about 19% below trend, mostly through reduced hiring. The authors stress the pattern is descriptive, not proof that AI caused it.

Does the Stanford study show widespread AI job losses?

No. The authors find no widespread displacement. The effect is concentrated among young workers in jobs where AI automates tasks; where AI complements workers, employment is flat or rising.

Are wages falling for AI-exposed jobs?

According to the study, the adjustment shows up in employment rather than base pay.