The research, in plain terms
The new analysis, released in January 2026 by researchers Sam Manning and Tomás Aguirre of the Centre for the Governance of AI, together with Brookings senior fellow Mark Muro, takes a various method than previous AI research studies. Instead of just cataloguing which jobs AI can theoretically carry out, it asks a harder question: if a worker loses their job to AI, how most likely are they to land on their feet?
To address that, the researchers constructed what they call an “adaptive capacity index” for 356 professions covering 95.9% of the U.S. workforce. The index makes use of 4 factors: an employee’s liquid cost savings, the transferability of their abilities to other roles, the density of job opportunities in their regional labor market, and their age. Integrate that index with basic AI direct exposure scores and a picture emerges that is both assuring in locations and deeply worrying in others.
Of the 37.1 million employees whose jobs fall in the leading quartile of AI exposure, about 26.5 million likewise have above-median adaptive capacity. Software application designers, financial experts, attorneys– yes, their tasks are extremely exposed to AI, but they tend to have cost savings, broad ability, and expert networks that make transitions workable. The system, for them, has some give.
For 6.1 million employees, it does not. These are individuals whose tasks are both extremely exposed to AI automation and who score in the bottom quartile for adaptive capability. They have restricted savings. Their abilities do not transfer easily. They are frequently older. And they regularly reside in smaller sized cities and towns where the next job just isn’t there. If AI takes their work, the research recommends, the repercussions will be lasting.
The professions focused in that vulnerable group read like a staffing directory site for a mid-sized home mortgage lending institution.