Job Displacement in the Age of Artificial Intelligence

Divyansh Mishra¹*, Rekha Agarwal², Rajesh Kumar Mishra³

Abstract

Estimates of how much employment artificial intelligence will displace span nearly an order of magnitude, from roughly nine per cent of jobs at high risk to two-thirds of occupations exposed in some degree. This paper argues that the spread is not primarily a failure of measurement but a consequence of three distinct questions being conflated: how many tasks a technology can perform, how many jobs a firm chooses to eliminate, and how many workers the economy as a whole fails to reabsorb. Building on the task-based framework of automation and reinstatement, the paper develops a fourth and largely neglected mechanism, termed here the apprenticeship externality, whereby the tasks most readily automated are also the tasks through which occupational expertise has historically been formed. Synthesising the forecasting literature from 2013 to 2026 with the first generation of post-ChatGPT empirical studies, the paper reports a consistent pattern. Aggregate displacement remains undetectable in national labour-market statistics through mid-2026, while a sharp and widening effect has emerged at the entry level: employment of workers aged twenty-two to twenty-five in the most exposed occupations now stands approximately nineteen per cent below where it would be had it tracked their less-exposed peers, with no comparable gap among experienced workers, and the divergence operates through reduced hiring rather than increased separation. Firm-level attribution data move in the same direction, artificial intelligence having become the single most frequently cited reason for announced job cuts in the United States for five consecutive months in 2026. A case study of India’s information technology and business process management sector shows aggregate employment growth of roughly 135,000 positions in FY2026 coinciding with a twenty to twenty-five per cent contraction in entry-level intake, a pattern that the aggregate figure conceals entirely. The paper argues that the operative risk over the medium term is not mass unemployment but the erosion of the entry-level rung upon which occupational skill formation depends, together with a redistribution of returns from labour to capital and from outsourcing economies to the owners of frontier models. Because the affected cohorts are absent from employment statistics as individuals who never entered rather than as individuals who departed, the aggregate series that policymakers monitor are structurally incapable of detecting the mechanism until it is approximately a decade old. Policy attention should shift accordingly, from income replacement after displacement toward the deliberate preservation, subsidy and redesign of junior work, and toward the construction of flow-based measurement capable of observing the hiring margin in close to real time.

Keywords:

Artificial intelligence; Technological unemployment; Task-based models; Labour market exposure; Entry-level employment; Skill formation; Apprenticeship externality