Generative artificial-intelligence tools are already reshaping the U.S. labour market, according to a new paper from Stanford University’s Digital Economy Lab. Analysing anonymised monthly payroll records for about 25 million workers supplied by ADP, three Stanford economists found a clear early-career divergence: employment for 22- to 25-year-olds in occupations most exposed to AI fell 13 percent between late 2022—when ChatGPT debuted—and July 2025. The losses are heaviest in roles where AI systems can fully automate routine cognitive tasks. Entry-level software developers saw head-count drop nearly 20 percent, while customer-service and accounting jobs registered similar double-digit declines. By contrast, overall employment in those occupations rose for workers aged 26 to 50, in some cases by as much as 9 percent, suggesting firms are retaining experienced staff while curbing new hiring. Researchers Erik Brynjolfsson, Bharat Chandar and Ruyu Chen distinguish between jobs in which AI primarily replaces humans and those where it augments them. Occupations that rely on AI as a co-pilot—such as managerial roles and marketing strategy—showed stable or growing early-career employment. Wages across the dataset have remained broadly unchanged, indicating companies are trimming entry-level positions rather than cutting pay. The study, titled "Canaries in the Coal Mine?", offers one of the first large-scale empirical looks at AI’s labour impact using real-time corporate data. The authors caution that the technology’s reach could expand beyond junior roles if firms continue to favour automation over augmentation, and they urge policymakers and employers to monitor workforce composition and invest in skills that complement, rather than compete with, rapidly improving AI systems.