Closing of the entry rung

This is the labour-market effect with the most observed evidence and the one most abused when it is cited. The finding comes from administrative payroll data —a panel of between 3.5 and 5 million employees per month, from January 2021 to June 2026— and says that employment of workers aged 22 to 25 in AI-exposed occupations is 19% below where it would be had it kept pace with their less exposed peers ^[164]Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial IntelligenceBrynjolfsson, Erik; Chandar, Bharat; Chen, Ruyu · 2026 · preprintView source ↗Accessed on 9 September 2026.

That 19% is not a 19% fall. It is a relative divergence: employment for the group in the two most exposed quintiles fell ~11% while that of the three least exposed quintiles grew ~10%. Total employment for the group came out at −1.9%. Nor is the channel the one assumed: separation rates fell in both groups —the opposite of what displacement would imply—, so what is narrowing is hiring ^[164]Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial IntelligenceBrynjolfsson, Erik; Chandar, Bharat; Chen, Ruyu · 2026 · preprintView source ↗Accessed on 9 September 2026. The declines are concentrated where the predominant use is automation, not augmentation ^[991]Canaries Dashboard — AI Economic IndicatorsStanford Digital Economy Lab; ADP Research · 2026 · datasetView source ↗Accessed on 21 September 2026.

What this does not demonstrate. The authors do not claim causality: they read the data as early descriptive indicators, and list three weaknesses —the effect attenuates when controlling for education, there are diverging trends predating generative AI, and the pattern is more pronounced in their sample than in the national surveys— ^[164]Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial IntelligenceBrynjolfsson, Erik; Chandar, Bharat; Chen, Ruyu · 2026 · preprintView source ↗Accessed on 9 September 2026. The gap in the American Community Survey is −0.022 [−0.055, +0.011]: indistinguishable from zero. Zanna Iscenko and Fabien Curto Millet —Google economists, and that has to be said— argue that the break fits better with the most aggressive rate cycle in forty years, and that job postings in those occupations also fell more in 2020, when generative AI could not even have been the explanation ^[566]Looking for the Ladder: Is AI Impacting Entry-Level Jobs?Iscenko, Zanna; Curto Millet, Fabien · 2026 · reportView source ↗Accessed on 9 September 2026.