Two Studies Asked If AI Is Killing Entry-Level Jobs and Disagreed
Stanford says young workers in AI-exposed jobs are 19% behind. Ramp says the biggest AI spenders grew entry-level headcount 12%. Both are right, and the reason matters.
Quick answer
Stanford reported in August 2026 that employment for 22 to 25 year olds in highly AI-exposed occupations sits about 19% below where it would otherwise be. Ramp, across 21,000 US firms, found entry-level headcount grew 12% at the largest AI investors. Stanford measured occupations and Ramp measured companies, so AI appears to relocate junior work rather than delete it.
Two credible research teams asked whether AI is eliminating entry-level jobs in 2026. They reached opposite conclusions, and both published their data.
Reconciling them turns out to be more useful than picking a side.
What Stanford found
The Stanford Digital Economy Lab, whose "Canaries in the Coal Mine?" research tracks entry-level employment using payroll data, published an update on August 12, 2026. Its title states the finding directly: "No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%."
Employment among workers aged 22 to 25 in highly AI-exposed occupations now sits roughly 19% below where it would be had it kept pace with less-exposed workers. A year earlier that gap was about 15%.
Two things deserve emphasis, because both are routinely lost in secondhand coverage. The gap is relative, measured against comparison groups, not an absolute collapse in headcount. And the researchers' own headline says there is no widespread displacement. The story is a widening gap for a specific group in specific roles, not a general jobs apocalypse.
What Ramp found
Ramp's economics lab took a different approach: firm-level corporate spending data joined to workforce records from Revelio Labs, covering more than 21,000 US firms.
- Companies that adopted AI grew headcount 10.2% over the two years following adoption.
- At companies making the largest AI investments, entry-level headcount grew 12%.
- Those gains came entirely from high-intensity adopters, defined as firms in the top third of AI spend per employee per month in their first three months.
- Low-intensity adopters showed no statistically significant change in either direction.
Ramp states its own main caveat clearly, and it should be carried along with the finding: AI adopters were already larger, more engineering-intensive, more likely to be venture-backed, and faster-growing before they adopted. Heavy AI spending may be a symptom of a healthy company as much as a cause of one.
Why do the two studies disagree?
The two studies are not measuring the same object.
| Stanford | Ramp | |
|---|---|---|
| Unit of analysis | Occupations | Companies |
| Question | What happened to people in AI-exposed roles? | What happened to firms that bought AI? |
| Headline | 19% employment gap for ages 22 to 25 | Entry-level headcount up 12% at heavy adopters |
An occupation can shrink while the companies investing hardest in that occupation's tooling expand. That happens when work moves between employers rather than disappearing from the economy.
Which is the most plausible reading of both datasets together: AI is not deleting a fixed quantity of junior jobs. It is relocating them, away from firms treating AI as a way to need fewer people, toward firms treating it as a way to take on more work.
The finding almost nobody quotes
The most actionable number in either study is the one about dabbling.
Low-intensity adopters, the firms that bought a seat or two and experimented, showed no statistically significant change at all. Not job losses. Not job gains. Nothing.
Ramp's high-intensity group was not defined by ambition or strategy decks. It was defined by spend per employee: firms using multiple models and the more advanced products, coding agents and APIs rather than a chat subscription. That is a measurable threshold, and beneath it nothing observable happened to headcount either way.
What this changes for a business deciding what to do
The strategic question is not whether to adopt AI. It is which of two questions you are pointing it at.
- "How do we run this with fewer people?" produces a cost program. It has a floor, and you eventually hit it.
- "How much more can we take on now?" produces a capacity program. Ramp's growth showed up in the second group.
These sound like the same initiative in a board deck. They produce opposite hiring decisions within two years.
The relocation reading is worth saying plainly, because it cuts against a lot of the coverage, including reporting we have done here on companies cutting the support layer to fund AI platforms, such as Dentsu's profit rising on flat organic growth. That reporting is accurate about those companies. Ramp's data is a reminder that those companies are not the whole distribution, and the evidence does not say they are the winning half.
Sources
- Stanford Digital Economy Lab, "No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%," August 12, 2026. Underlying paper: Brynjolfsson, Chandar and Chen, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence."
- Ramp Economics Lab, "A New Look at AI's Impact on Jobs," with Revelio Labs, June 30, 2026.
Frequently asked questions
- Is AI reducing entry-level hiring?
- It depends on what you measure. By occupation, Stanford's Digital Economy Lab found employment for 22 to 25 year olds in highly AI-exposed roles is about 19% below where it would have been. By company, Ramp found firms making the largest AI investments grew entry-level headcount 12% over two years.
- What did the Stanford Digital Economy Lab find in August 2026?
- Its August 12, 2026 update was titled 'No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%.' The gap for 22 to 25 year olds in highly AI-exposed occupations widened from about 15% a year earlier to 19%.
- Do companies that adopt AI hire more or fewer people?
- Ramp's data across more than 21,000 US firms found headcount grew 10.2% over the two years after adoption, but entirely among high-intensity adopters. Low-intensity adopters showed no statistically significant change either way.