Stanford just published one of the largest studies yet on artificial intelligence and jobs. Three researchers read payroll records covering millions of American workers, running through June 2026. The headline runs against the loudest predictions. No wave of AI layoffs. What they found is narrower, and more useful if you own a P&L.
The paper is called "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence." Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen wrote it, dated August 2026, out of the Stanford Digital Economy Lab. Their line from the abstract: "We find no evidence of widespread, economy-wide job displacement."
So where is the pressure? On the youngest workers, in the jobs AI can already do.
What did the Stanford study actually find?
Employment for workers ages 22 to 25 in AI-exposed occupations now sits 19% below where it would be had it kept pace with their less-exposed peers. Experienced workers show no comparable shortfall. The finding comes from anonymized ADP payroll data covering millions of U.S. workers through June 2026.
How far below trend employment for 22 to 25-year-olds in AI-exposed occupations now sits, versus their less-exposed peers.
Source: Stanford Digital Economy Lab, 2026That 19% needs a caveat, because it is easy to misread. It is a "kept-pace" shortfall rather than a count of jobs destroyed. In levels, employment of 22 to 25-year-olds in the two most AI-exposed quintiles fell about 11% between November 2022 and June 2026, while the same age group in the three least-exposed quintiles grew about 10%. ADP pays more than 26 million U.S. workers, so the sample is deep. The authors describe the 21 percentage-point divergence between those groups as 19% relative to growth for the bottom three quintiles.
The shortfall is also widening. It stood at 15% in the July 2025 data and reached 19% by June 2026. Under a year, four points wider.
One more number keeps this honest. Total employment for all 22 to 25-year-olds is roughly flat, about 2% below its November 2022 level, once you count young people moving into less-exposed jobs. They are still getting hired. Fewer of them are getting hired into the exposed roles.
The authors are careful here, and any leader reading this should be too. They frame these results as early, descriptive indicators, not causal estimates. The pattern is also stronger in the ADP sample than in national benchmarks, so 19% is a sample-specific figure, not an economy-wide reading.
Why is entry-level hiring the pressure point?
The change runs through hiring, not firing. The authors write that it "operates primarily through reduced hiring of young workers rather than increased separations." Pay is not the lever either. Their words: "Adjustment is occurring through employment rather than base compensation."
Read that together and a picture forms for anyone who signs offer letters. Companies are not cutting junior salaries or purging junior staff. They are opening fewer entry-level seats in the roles where AI covers the task. Entry-level work has long concentrated the routine, teachable pieces of a job. Those are exactly the pieces a capable model now handles.
What separates the roles that are growing?
One line in the study does the heavy lifting: "Declines are concentrated in occupations where AI usage primarily substitutes for human tasks; where usage primarily complements workers, employment is flat or rising, especially for experienced workers." Older cohorts kept growing. Workers aged 35 to 40 saw employment rise as much as 11% over the same window.
They direct the tool instead of competing with it.
So the dividing line is substitution versus complementation. Where AI replaces the task, seats shrink. Where AI amplifies a skilled person, seats hold or grow. The people on the growing side share a trait. They direct the tool instead of competing with it.
That trait is learnable, and it has a shape. The 7 Levels of AI Proficiency describes the climb from Level 1: The Cadet (AI Aware), someone who has tried AI, up to Level 7: Mission Director (AI Orchestrator), someone who runs AI systems and leads the people around them. Complementation lives in the middle and upper rungs. A Level 3: Lieutenant (Critical Thinker) pushes back on weak answers instead of accepting the first draft. A Level 4: Commander (Context Engineer) feeds the model the right context so the output fits the business. Those are the workers the study shows employers keep hiring.
What should you do with this if you lead a team?
Start with the hiring question, because that is where the study points. Cutting entry-level roles to save money is the reflex. The more durable choice is to redesign those roles around complementation, so a new hire spends their first year directing AI on actual projects rather than doing the tasks a model already covers. That protects your future bench. Firms that stop hiring juniors entirely may win a budget cycle and lose a decade of talent development.
Second, treat capability as an asset you can measure. You track revenue per employee and pipeline coverage. AI proficiency deserves the same rigor. Knowing where your people sit on The 7 Levels of AI Proficiency tells you where to place training dollars and which teams can take on more ambitious projects.
Third, keep the caveats in the room when you brief the board. The researchers documented a divergence, and a widening one, not a proven cause. Some divergent trends predate generative AI, and the effect softens when they control for education. This is an early warning, and they built a public set of AI Economic Indicators to track it going forward. Use it as a reason to prepare, not a reason to panic.
A first step is simple. Find out where your team actually sits on The 7 Levels of AI Proficiency today, then decide where the training dollars go. The trend widened from 15% to 19% in under a year, so preparing now beats reacting later. What would your entry-level roles look like if every new hire spent year one directing AI instead of racing it?
Related reading: Level 4: The Commander (Context Engineer).
Sources
- No Widespread Displacement, but the AI Employment Effect for Young Workers Has Widened (Stanford Digital Economy Lab announcement)
- Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence (working paper PDF, August 2026)
- Canaries in the Coal Mine (publication page)
- Canaries in the Coal Mine (SIEPR working paper page)
- Stanford Digital Economy Lab: AI Economic Indicators
Frequently Asked Questions
Does this prove AI is destroying jobs?
No. The authors are explicit that these are early, descriptive indicators rather than causal estimates. They found a divergence in the data, not a proven cause, and they note some trends that predate generative AI.
Is the 19% figure an economy-wide number?
No. It comes from one payroll dataset and shows up more strongly there than in national benchmarks. Read it as a directional early warning for exposed entry-level roles, not a national statistic.
Which workers look most insulated?
Experienced workers, and anyone whose work AI complements rather than replaces. Employment for older cohorts held or grew, especially where AI amplified their judgment.
Bring us the work that keeps coming back to you.
We will trace the workflow, identify where a system can create capacity, and decide what should remain with your people.