A January 2026 Brookings Institution study, published as an NBER working paper by Sam Manning, Tomás Aguirre, Mark Muro, and Shriya Methkupally, introduces a two-dimensional framework for assessing AI's labor market impact that goes beyond simple exposure metrics. Rather than asking only how much of a job's tasks AI can perform, the research adds a second axis: workers' "adaptive capacity," defined by their financial savings, age, skill transferability, local labor market density, and union membership. The Washington Post visualized this research as an interactive feature, with the piece making clear that AI exposure alone does not predict hardship — a worker with financial buffers and transferable skills is far better positioned than one who lacks those resources.

The headline finding is that of approximately 37.1 million highly AI-exposed workers in the United States, roughly 70 percent — around 26.5 million — have sufficient adaptive capacity to weather potential displacement. The more pressing concern is a cohort of 6.1 million workers concentrated in clerical and administrative roles, such as secretaries and office support staff, who are both highly exposed and poorly equipped to adapt. The research's most striking finding is demographic: 86 percent of this most-vulnerable group are women, meaning AI-driven labor disruption, without targeted intervention, risks compounding existing economic inequalities. The contrast the researchers draw between web designers and secretaries makes the point concrete — both face high AI exposure, but web designers typically have stronger financial buffers and more transferable skills.

Geographically, the research finds that highly exposed, low-adaptive-capacity workers are disproportionately concentrated in college towns and state capitals across the Mountain West and Midwest — regions that may have fewer alternative employment pathways for displaced clerical workers. The policy implication the authors draw is direct: support programs should be targeted at the 6.1 million workers with the weakest adaptive capacity, as they face the highest welfare costs from displacement. Prior AI labor research has focused heavily on <a href="/news/2026-03-14-andrej-karpathy-scores-ai-exposure-of-342-us-occupations-using-gemini-flash-llm">which occupations are exposed</a>; this study's contribution is shifting attention to who, specifically, will bear the cost when exposure becomes displacement.