More startups did not mean more startup jobs in one U.S. study

Three studies count different things

Gupta, Qian, Simintzi and Sun compare U.S. startups by pre-ChatGPT task exposure. Their latest abstract says existing exposed startups reduced employment within two quarters, mainly in junior and implementation-heavy roles. More firms entered exposed markets, but each entrant was smaller. The authors report no change in aggregate startup employment and a shift toward senior roles.

Bena, Bian and Giannetti report 20% more startup formation in more exposed industries and industry-level employment and wage growth from new entry. An independent UK paper by Agarwal finds no statistically significant change in firm entry with higher industry LLM exposure, but does find smaller founding teams from 2025. A one-standard-deviation exposure increase was associated with 2.1% fewer founding directors and a 14.3% lower share of firms with at least three directors.

What this changes—and what it does not

The U.S. result blocks a simple reading of startup counts as replacement jobs for workers losing positions at existing firms. It does not directly refute the CEPR result: startup employment, broader industry employment, U.S. markets and UK incorporations are different samples and outcomes. A non-significant UK entry result is not proof that entry never rose. None of the three measures a worker's personal chance of finding a similar-paying role.

We inspected the papers' current abstracts and the CEPR author summary, not the full Gupta or Agarwal methods. Their confidence intervals, coverage and checks against concurrent industry trends need review. Task exposure is not observed AI adoption by each firm.

The next test

Compare new-firm payroll gains with incumbent startup losses in the same market and period, then follow displaced junior workers into later employment and pay. Recheck whether the result survives direct AI-adoption measures and comparable non-exposed firms. Until then, the honest answer is narrower: startup creation may rise without a net startup-jobs gain.