Still waters, rapid currents: Early labor market transformation under generative AI
S · Shapes the field Anders Humlum (University of Chicago), whose Danish survey-register studies are among the most cited evidence on generative AI and wages, with 25,000 workers linked to administrative records; it directly tests the entry-level displacement claim that drives the current debate.
Summary based on the full text·Original: about 24 minutes to read
In short
A Brookings article by Anders Humlum (University of Chicago) and Emilie Vestergaard (University of Copenhagen), published 6 October 2026, summarising their research on generative AI in the Danish labour market. They surveyed 25,000 workers in 11 AI-exposed occupations and linked answers to monthly administrative records through December 2024. The headline is a precise null: AI chatbots have not changed earnings or hours, for users or for adopting workplaces. Beneath that, work is being reorganised, adopters are switching into better-paid occupations, and adopting firms are not the cause of falling early-career employment.
Key numbers
- 2%largest average effect on earnings or hours that the estimates still allow, for workers and workplaces
- 93%workers who have used chatbots where employers combine encouragement, enterprise tools and training (41% where they do nothing)
- 12 ppfaster earnings growth for chatbot adopters who switched occupations, versus other Danish workers
- 1/3 pplargest difference in early-career worker share at adopting workplaces that the estimates allow
As reported by the source.
English·Web page·opens brookings.edu
What it reports
- 43% of workers in the 11 occupations work for employers that explicitly encourage chatbot use; 6% are prohibited. In encouraging workplaces, 61% have enterprise chatbots and 39% have had training.
- Where employers do nothing, about 41% of workers still use chatbots. Where encouragement, tools and training are combined, 93% have used them, 28% daily, and 19% save more than one hour a day.
- New AI tasks: about four in ten involve generating content, about a third involve overseeing AI (quality review, ethics, compliance), and about a quarter are integration work such as writing usage policies and building chatbots into workflows.
- 85% of users spend saved time on other job tasks. About 4% of non-users report new workload caused by the tools; among teachers who never used them, 10%.
- Earnings and hours: no differential change after ChatGPT's launch; confidence intervals rule out effects above 2%. 98% of adopters say their earnings were unaffected.
- Adopters earned about 9% more (encouraging employers) and 4% more (others) than non-adopters, but those gaps existed before ChatGPT.
- Adopters put in about 4% of a full-time equivalent more hours in their latest occupation, a sign of occupational switching; switchers' earnings growth runs 12 percentage points ahead of other Danish workers.
- Early-career shares at encouraging workplaces evolved no differently; differences larger than a third of a percentage point are ruled out.
How they got it
Two representative survey rounds (late 2023 and late 2024) run with Statistics Denmark, each inviting 115,000 workers through official digital mailboxes; the 2024 round produced about 25,000 complete responses across 7,000 workplaces. Answers were linked to monthly register data on earnings, hours and occupations, and adopters were compared with similar non-adopters in a difference-in-differences design around November 2022, with pre-trend checks.
Limits
- Data end in December 2024; effects that take longer to surface, or that follow agentic tools, are outside the window.
- Denmark has flexible but distinctive labour institutions; transfer to other countries is assumed, not tested.
- The mobility result rests on relatively few switchers, too few to move average adopter earnings.
Why it matters for companies
The adjustment so far is happening inside firms, through new tasks and reorganised roles, rather than in pay or headcount. Adoption roughly doubles when employers pair permission with enterprise tools and training, so the organisational investment matters more than licence counts. The evidence also cautions against blaming AI adoption for a hiring slowdown of junior staff without checking actual use. HR teams should track task change and internal mobility, since that is where the measurable effects currently appear.
Summary based on the full text. Read the original for the full argument and data.Read the original on brookings.edu