"The Contested Ledger: Artificial Intelligence, Automation, and the Labor Market" An Economic Review of Employment, Wage, and Productivity Effects, with Evidence from India's Services-Led Economy

Abstract

Three years after the public release of ChatGPT, the economic literature on artificial intelligence's effect on employment, wages, and productivity has grown rapidly but has not converged. This review synthesises that literature, organising it around the task-based framework developed by Autor, Levy, and Murnane (2003) and extended by Acemoglu and Restrepo's (2019) distinction between the displacement and reinstatement effects of automating technologies, and situates the evidence against India's services-led economy as a case study of concentrated exposure. On employment, the review finds the aggregate evidence genuinely contested: administrative payroll data analysed by Brynjolfsson, Chandar, and Chen (2025a) documents sizable relative employment declines in the most AI-exposed occupations following ChatGPT's release, while Hartley, Jolevski, Melo, and Moore's (2026) large-scale US worker survey finds small positive wage effects and no statistically significant decline in job openings or employment despite 35.9 percent of workers having adopted generative AI tools by December 2025 — a divergence this review argues partly reflects Iscenko and Millet's (2026) finding that AI-exposed occupations are disproportionately concentrated in interest-rate-sensitive sectors, confounding technology effects with macroeconomic conditions. Where the evidence is more consistent is at the margin: employment among 22-to-25-year-old workers in AI-exposed US technology occupations fell roughly 6 percent between late 2022 and mid-2025 even as older workers in the same occupations gained roughly 9 percent, a pattern echoed in India's own entry-level information-technology hiring, which EY estimates fell 20 to 25 percent over the same period. On wages, the review highlights Brynjolfsson, Li, and Raymond's (2025) Quarterly Journal of Economics finding that generative AI assistance raised customer-service agents' productivity by up to 34 percent for novice workers while leaving experienced workers' productivity essentially unchanged, a skill-levelling pattern in some tension with the skill-biased technical change literature that dominated the pre-generative-AI automation debate. On productivity, the review reads firm- and task-level gains against Brynjolfsson, Rock, and Syverson's (2019) productivity J-curve framework, arguing that the continued absence of a clear acceleration in aggregate productivity statistics is consistent with, rather than contrary to, the historical pattern following prior general-purpose technologies. India's IT-BPM sector, examined as a concentrated case study, shows early, tangible effects: the country's three largest IT services firms recorded a combined net headcount decline of approximately 65,000 in 2024, the first such combined decline in two decades. The review closes with policy recommendations addressing the specific evidentiary and distributional gaps this synthesis identifies.