America’s wired belts, and where the first rust forms
On a map of American ambition, Washington and Boston glow for all the familiar reasons—policy and research, consulting and code, degrees stacked high enough to block the sun. Yesterday, Tufts University’s Digital Planet added a colder layer to that glow: a new index that doesn’t ask who is merely exposed to AI, but where and in which jobs the machines are most likely to actually push people out. For a country still talking about “copilots” and “productivity gains,” it reads less like hype and more like a weather report with a storm line arcing straight through the nation’s brain centers.
They call it the American AI Jobs Risk Index, and its headline is disarmingly specific. Under a mid-range adoption path over the next two to five years, about 9.3 million jobs sit in the displacement zone, attached to roughly $757 billion in annual wages. The plausible band is wide—2.7 million at the slow end, 19.5 million at the fast—but the point is not clairvoyance; it is calibration. This is not a tally of roles that might use new tools. It is a projection, by place and by occupation, of where the labor market loses seats as tasks pass to software.
Not just exposure—exit risk, priced in
What’s new here is the insistence on vulnerability rather than abstract exposure, and the choice to tie that vulnerability to income. Down to metro and non-metro areas, the index tracks where wages could vanish from payrolls as adoption climbs along a logistic curve. It is careful, too, about what it is not measuring. New job creation is out of scope for now; the authors want a clear look at downside risk before we start netting things out with the proverbial “jobs of tomorrow.” They also caution against the comfortable story that automation always pushes workers up the value chain. Their inference is sharper: for every one-point increase in task automation, jobs fall by about three-quarters of a point. Tasks don’t just move up; some of them disappear, and with them, headcount.
The aggregate number—roughly 6% economy‑wide vulnerability—masks steeper cliffs where digital work is the work. Information sits near 18%; Finance and Insurance around 16%; Professional, Scientific, and Technical Services also near 16%. The occupations most lopsidedly exposed are telling: Writers and Authors at 57% vulnerability; Computer Programmers at 55%; Web and Digital Interface Designers also at 55%. If you’ve felt the ground shift under keyboards faster than under cranes, the index quantifies that sensation.
Where the gravity is strongest
The geography is not heartland nostalgia. It is the corridor and the coast. In absolute terms, the places that pull in the largest income losses map to the familiar megahubs—New York, Los Angeles, Washington, San Francisco, Chicago, Dallas, Boston. By share of jobs at risk, the lights flash over Washington, Massachusetts, Virginia, Maryland, Washington state, and Colorado. Add up the payrolls most exposed, and California, Texas, New York, Florida, and Illinois account for about four in ten projected losses. Digital Planet brands these hotspots “wired belts,” innovation centers that, paradoxically, could rust first as AI scales.
That inversion reframes the politics. The states likely to take the hardest hits are already out front on AI bills and rulemaking, while safer states are legislating less. It is not hard to imagine a coming decade in which the places that built the tools demand stronger guardrails, richer disclosures, and bigger public cushions. Losing income in the country’s cognitive core is a very different bargaining problem than easing layoffs on a factory floor. Safety nets designed for cyclical shocks start to creak when the displaced are holding graduate degrees.
The edge of the curve
One of the index’s most unnerving ideas is not the headline number but the hinge. Digital Planet flags 4.9 million workers across 33 occupations who are at what they call a “tipping point”—roles that could lurch from under 10% displacement to more than 40% within the two‑to‑five‑year window, depending on how adoption bends. These are the jobs living on the steep part of the curve, where a new model release or a workflow integration flips from nice‑to‑have to default. It is a reminder that the timeline for adjustment is tight; the gap between exposure and exit can close quickly once usage crosses internal thresholds.
There is also a sobering distributional twist. Roughly 38% of workers are effectively “AI‑proof” in this framework, but they cluster in the lowest‑paid roles. That means the immediate pain, if the index proves right, concentrates higher up the wage ladder—precisely where household budgets, local tax bases, and political attention are thickest. The risk is not just job loss; it is a redistribution of security, with stability hardening at the bottom and volatility spreading through the professional ranks.
How they did the math
Under the hood, the team blends O*NET task definitions and federal employment and wage data with multiple views of AI’s reach—Eloundou et al.’s task‑time reduction, Brynjolfsson et al.’s “suitability for machine learning,” and Felten et al.’s mapping of AI abilities to occupations. They layer in observed usage and automation signals, from Anthropic’s Economic Index and Microsoft’s Copilot adoption, then run the adoption path on a logistic curve. Roles heavy in physical activity get adjusted downward—reality checks on where current models simply can’t go. The translation from task shifts to employment and income effects is presented as scenario‑based rather than deterministic, a set of futures to plan against rather than a prophecy to resign to.
The policy architecture they sketch is correspondingly practical. If half of projected losses come from just 26 occupations, targeted response becomes tractable. Safety nets need to modernize for faster reentry and partial‑earnings coverage. Companies should be pressed to disclose how AI reshapes their workforces. Public incentives for AI adoption ought to come tied to measurable reskilling commitments. None of this is presented as cure‑all; it is a triage plan for a labor market whose fault lines now cut across high‑income cognitive work.
Reading the index like an altimeter
For those of us working inside the professions with the highest percentages—yes, writers are at the top of this chart—the index lands like an instrument panel. It doesn’t tell you whether a new category of jobs will bloom two years from now. It does tell you where altitude is being lost today, and how fast, if adoption accelerates. The plausible range matters: if AI uptake stalls, the U.S. could shed closer to 2.7 million jobs; if it sprints, nearly 20 million. The difference between those worlds may hinge on decisions being made this quarter in IT budgets, product roadmaps, and procurement rules.
Watch how quickly copilots become the default interface at work. Track where task routing moves from suggestion to enforcement. Pay attention to the 33 occupations on the steep slope of the curve. And keep an eye on the places with the most to lose; when Washington, Boston, San Francisco, and New York feel their payrolls thin, the country’s policy conversation will shift from thought experiments to line items.
The public discussion of these findings is set for Friday, March 27, 2026. It will not be a comfortable conversation, and that is the point. The index is a mirror held up to the economy we built—one that digitized its core, concentrated its talent, and now must decide how to absorb the consequences of making cognition computable. If the first rust appears on the shiniest parts, it’s because those are the parts most exposed to the new weather. The question is whether we treat that as a diagnosis to act on, or a headline to scroll past until the curve catches up to us.