Research update · 3 October 2026
More AI-exposed startups does not settle the jobs question
Fewer junior openings can coexist with new firms hiring. A CEPR working-paper abstract reports 20% more startup formation in industries with higher generative-AI task exposure than in less-exposed industries after the technology spread. It also reports employment and wage growth from new entry. Neither finding proves AI caused a net gain for all workers.
What the researchers compared
Bena, Bian and Giannetti compare industries by task-level exposure to generative AI before and after its diffusion. Their abstract says individual new firms were smaller, but greater entry generated industry-level employment and wage growth, including in highly exposed occupations. It also says more startups offered AI products and entry extended beyond established funding hubs.
What the abstract cannot establish
We inspected the CEPR abstract and an author summary, not the complete paid paper. The accessible abstract does not specify the geography, observation dates, confidence intervals, exact employment or wage estimates, or enough of the identification design to test alternative explanations. Task exposure is not proof each new firm used AI. The 20% figure concerns startup formation, not 20% more jobs.
New-firm growth also does not tell us whether incumbent employers cut positions, whether workers who lost jobs moved into the startups, or whether the gains persisted. This is a counter-signal to a one-way displacement story, not an all-clear for an exposed worker.
What would change the conclusion
The full paper's methods and linked employer–employee evidence would help test whether entry gains remain after incumbent losses and industry trends are counted. Until then, compare this result with the junior software vacancy warning as a different population and outcome, not a direct contradiction.