Mark Muro and his colleagues at the Brookings Institution published what looked like a dry geographic analysis earlier this month — a deep dive into which American counties have the most workers exposed to artificial intelligence. The headline finding was almost an afterthought: 62 of the 100 most AI-exposed counties in the United States voted Democratic in 2024. The authors noted the correlation carefully, distanced themselves from causal claims, and moved on.

But read against the backdrop of what just happened in New York City this week, the paper looks less like a policy brief and more like a forensic explanation of a political earthquake nobody in the Democratic establishment saw coming.

The map that explains everything

On Tuesday night, New York City Mayor Zohran Mamdani — the democratic socialist who shocked the Democratic Party by routing Andrew Cuomo in last year’s mayoral race — completed one of the most striking primary sweeps in recent memory. Mamdani-endorsed candidates defeated entrenched Democratic incumbents across multiple congressional districts, with at least a dozen candidates aligned with the Democratic Socialists of America winning statewide. The Democratic Party’s establishment wing, still struggling to explain Mamdani’s rise, is now scrambling to explain his reach.

The Brookings research from Muro, Todd Jones and Shriya Methkupally offers a structural answer they may not like.

New York County — Manhattan — registers among the highest AI automation exposure scores in the nation, according to the report. Between 14% and 19% of workers there are employed in occupations where AI is not just theoretically capable of handling their tasks, but is already doing so — automating rather than merely assisting.

The methodology, which draws on actual usage data from Anthropic’s Claude models, counts full automation at twice the weight of human-AI collaboration, making it a deliberate measure of displacement risk rather than productivity gain.

That is the electorate that made Zohran Mamdani mayor. And this week, it voted his way again.

Bougie and scared

The uncomfortable truth embedded in the Brookings analysis is that the workers with the most rational reason to fear AI disruption are not factory hands in Ohio. The cognitive, nonroutine, information-based tasks that define white-collar urban work — conducting research, writing code, drafting presentations, preparing analyses, creating marketing content — are precisely the task categories where AI models perform best and are most actively deployed.

The more educated and “office-based” the job, the paper finds, the more involved it is with AI. And those jobs concentrate overwhelmingly in Democratic-leaning metros: the 62 most AI-exposed blue counties account for 75% of the population of the top 100 most AI-exposed counties in the country.

These are not working-class voters in the traditional sense. Many of them have college degrees, decent salaries, and strong opinions about their economic futures. They are, by most political taxonomies, exactly the kind of “professional class” Democrats that the party has spent decades courting. But as a separate academic study cited by the Brookings authors found, Democrats are both more likely to use AI and more likely to hold jobs with higher AI exposure than Republicans. Being the party of AI adoption and the party of AI anxiety are turning out to be the same thing.

Mamdani understood this intuitively before the economists wrote it down.

Naming the fear

“The Democratic Party has lost its focus on working people,” Mamdani said earlier this month — a formulation his critics dismissed as socialist boilerplate. But strip away the ideological branding and what he is describing is something the Brookings data validates empirically: a Democratic-leaning professional class whose economic anxiety has been met with platitudes about “retraining” and “the future of work,” and which is now voting its frustration. Critics heard “working people” and thought of the Rust Belt, but the “Wired Belt,” to paraphrase Tufts’ Bhaskar Chakravorti, were the ones who were really listening.

High-exposure blue counties flagged explicitly in the Brookings report as potential political flashpoints include Jefferson County in the Denver area, Hennepin County in Minneapolis, and King County in the Seattle area. At the state level, Massachusetts, New York, California, and Washington, D.C., all register AI exposure levels between 13% and 17% of workers — and all are trending toward the kind of working-class-coalition politics Mamdani is now actively exporting beyond New York.

The Brookings authors reach back to economist Jed Kolko’s research on industrial automation to make the same point in more measured academic language. A decade ago, Kolko showed how manufacturing job anxiety reshaped the politics of red counties; Muro and his co-authors warn that AI exposure “may turn out to be a source of economic and social concern especially in blue counties and states going forward.” They conclude with a line that reads, in retrospect, like a forecast of Tuesday’s results: “America’s bluest counties may become hotbeds of some of the AI era’s most agitated voters.”

They published that on June 3. The votes came in on June 24, with a Mamdani clean sweep.

A warning for November

None of this means democratic socialism is about to sweep the country (although it is sweeping exactly the types of blue cities highlighted by Muro and company). The Brookings authors are careful to note that AI exposure does not automatically produce job losses, and that separate research suggests AI adoption can actually generate employment and attract educated workers who lean Democratic. The causality is genuinely contested.

But in politics, perception has a way of outrunning data. CBS News reported just this week that support for democratic socialism is measurably rising across the U.S., with Mamdani’s election cited as a leading indicator. House Democrats, meanwhile, are increasingly anxious about the leftward pull of Mamdani’s coalition on the broader party. Most notably, swing states Arizona and Georgia — both ranking among the 15 most AI-exposed states in the country, per Brookings — are heading into the midterms with electorates that share the same structural anxieties as Mamdani’s New York base.

The most AI-exposed, most economically anxious, most politically restless voters in America live in blue cities — and they are already telling you exactly how they plan to vote. You’ve just been missing the reason for it.