Actual adopters: Danish small firms hire less; multinationals add seniors

Two newer original studies improve on occupational exposure but answer different questions, and a 2026 Danish worker-level null checks the wage/hours margin. Read 30 September for labor-risk synthesis.

Denmark: firms reporting AI adoption

Simone Maria Bonin, Saman Darougheh and Andreas Kuchler, Danmarks Nationalbank Working Paper 223, 16 September 2026, join Statistics Denmark AI-use surveys to monthly employer–employee registers. Firms first reporting adoption in 2023 lagged nonadopters’ and their own pre-existing employment trends by 11% by late 2025 (their bank’s chart notes explicitly caution this is a relative-to-trend result, not necessarily layoffs). The first author’s account attributes the effect to sub-100-full-time-equivalent firms, fewer hires concentrated in AI-exposed roles, and no detected wage effect; bank says no large aggregate employment change. Strength: measures use rather than exposure and separates hiring from separations. Weakness: business adoption remains selected; adjusted trends need a credible unobserved-shock comparison, this small-firm adopter cohort is not necessarily representative and use is not randomized. The authors report only about a quarter of exposed-industry job losses attributable to adopting firms in their decomposition; check the full working-paper methods before treating that fraction as causal.

41-country affiliate study: junior share falls as seniors grow

Bharat Chandar and Bouke Klein Teeselink, “How Does AI Change Labor Demand? Evidence from 41 Countries,” 20 September 2026, use 1.25 billion postings and 154 million employment records from January 2021 to March 2026. AI use proxied by vacancies mentioning gen-AI; compare matched foreign affiliates, instrumenting adoption with AI take-up by unrelated companies near the parent headquarters. Relative to controls, junior employment share was 1.9 percentage points lower by March 2026; senior headcount rose 6.7%, junior headcount estimate −2.5% not statistically significant; total headcount estimate +3.3%, marginally significant. So “fewer junior jobs” and “smaller junior share” are not equivalent. This seriously complicates a simple replacement story while leaving a junior career-ladder concern. Limited to digitally visible multinationals; job ads imperfectly proxy realized adoption and instrument requires no parent-city peer spillovers other than via adoption. Crucially, an adopter’s employment growth can take share from a nonadopter and need not imply positive economy-wide jobs; authors model this explicitly. Their estimate is not a national net-jobs count.

Earlier Denmark study: different outcome and time

Anders Humlum and Emilie Vestergaard, “Still Waters, Rapid Currents,” March 2026, linked actual worker/employer chatbot-use surveys to Danish records through 2024; difference-in-differences finds no detectable changes >2% in worker or workplace earnings and recorded hours over first two years, while reported task reorganization and mobility grew. This does not estimate Bonin et al.’s small-firm employment growth by late 2025. Nor is aggregate null incompatible with young-worker hiring disruption: different adopter definitions, outcomes and periods can produce both findings. The fair tension worth digging into is whether firms reporting AI adoption differ in underlying demand, and whether jobs not created at adopters are created elsewhere at similar career levels.