The people most likely to be devastated by AI are not factory workers or call center agents. They are the Manhattan law partners, the tenured academic class, and every professional who spent decades mastering the art of following the rules. That is the central point in a keynote delivered by economist Tyler Cowen at the Sana AI Summit in New York, May this year – a talk that has since spread rapidly across X and through Sand Hill Road group chats, where its rebuke of conventional elite wisdom proved either galvanizing or uncomfortable depending on the viewer.
The talk’s viral reach reflects a genuine fault line in how VC investors are placing bets. If Cowen is right, the most defensible venture categories over the next decade are not productivity software for knowledge workers but infrastructure for a world remade by initiative, physicality, and the experimental testing of AI-generated ideas.
Cowen, the Holbert L. Harris Professor of Economics at George Mason University and chair of the Mercatus Center, argued that AI is about to execute what he called a “status remix” across society. The mechanism is simple and brutal: smart, credentialed professionals who played by institutional rules are the ones whose skills AI replicates most easily. “There was an older world we all grew up in, where if you were very smart, went to a good school, worked hard and followed the rules,” he told the audience. “Those are actually the people who might lose the most.”
The biggest losers in his framing are the highest earners in rule-governed professions. A partner at a white-shoe firm earning $2 million a year in Manhattan does not become unemployed; instead, they get commoditized and relocated. “They will be sent to Houston. At 37 they’ll be an energy company executive, earning $350k a year.” The academic class, in Cowen’s telling, faces a similar status deflation.
The categories receiving the most capital in 2025 and 2026 have disproportionately targeted enterprise knowledge work: legal tech, contract analysis, financial research automation. Anthropic’s $50 billion Series H announced in May 2026 compressed national VC statistics so severely that New York’s $2 billion in monthly deal flow represented only 3% of US activity that month, against an underlying share closer to 12% when Anthropic is stripped out. The concentration of capital into frontier model infrastructure is rational if AI commoditizes the high-skill professional class. But Cowen’s framework suggests the next layer of value creation sits elsewhere.
Two job categories, in his view, are badly underinvested. The first is experimental testing of AI-generated ideas, particularly in biomedicine. “AI will have many, many new ideas as it does already,” he said. “We will need to test those ideas. In the world today, only PhDs run experiments. That’s on the way out.” The implication for investors is a large market for platforms that democratize experimental infrastructure, from lab automation to clinical trial coordination for non-specialist operators.
The second category is data collection from the physical world. Most physical reality remains undigitized. Cowen framed this as a constraint on AI capability rather than a solved problem: “Most of the world is not currently expressed in the form of data. Insofar as we turn the world into data, AI can do much more for us.” Sensor networks, field data services, and physical-digital translation tools are the unglamorous plays this thesis generates.
On macroeconomics, Cowen positioned himself explicitly against Silicon Valley consensus. His projection for AI’s contribution to GDP growth is half a percentage point, from roughly 2% to 2.5%, a forecast that provokes derision from founders and investors who believe transformative technology should produce transformative growth. His counter-argument is: “The smarter the AI is, the harder it is for you to work with it, the harder it is to get organizations to adopt it, it is a Human Bottleneck problem.” He estimates that government, higher education, healthcare, and the nonprofit sector constitute at least 40% of US GDP, and that these sectors will be slow to restructure. The FDA’s average drug approval timeline of a decade is his anchor example.
The counterintuitive case for that half-point: it may be the only mechanism by which the US national debt, now above $38 trillion, stabilizes at a manageable ratio rather than compounding toward crisis. “If our economy can grow two and a half percent instead of two percent, that debt, rather than exploding and making us the next Greece, actually converges to a manageable level,” Cowen said. “For the US economy, AI is our plan A. There is no plan B.” He also points out that the Human Bottleneck problem is a factor that will stabilize AI adoption, which may not be a completely negative thing.
The biggest beneficiaries, in Cowen’s analysis, are the global poor and immigrants. Both groups gain access to expert-level advice in diagnostics, legal matters, and business strategy at near-zero marginal cost. Both groups, he argues, have structural incentives toward the initiative and creative problem-solving that AI rewards. “They’re people who are willing to take initiative, did not necessarily go to the best schools, but think creatively about how to use AI.” The investment implication is a large emerging-market AI opportunity that venture capital has systematically underweighted relative to enterprise SaaS in developed markets.
The premium Cowen predicts for in-person presence and human physicality is perhaps the claim with the widest implications for how the next generation of companies should be built. “Being in the world and being human and being physical will be more the thing,” he said. He noted his own behavioral shift as evidence: more mentoring, more public talks, less writing, more travel. “Charisma will matter more. How you look will matter more.” For founders, this is a warning that the remote-first, async-native company structure optimized for cognitive output may be the wrong default for the decade ahead. He also emphasizes that “going back to the physical world” will feel like a healthy correction, and return to the baseline for many.
The geopolitical frame Cowen closed with is the most immediate concern for investors with international exposure: countries that fail to develop AI capacity risk losing sovereignty, and the world is bifurcating into US and Chinese technological spheres that will force every government to choose an alignment. That binary creates a durable tailwind for AI infrastructure investment in nations currently on the fence, and a corresponding risk for companies whose supply chains or data operations straddle the divide.
The thesis that initiative beats intelligence may seem like a feel-good reframe, but only at a surface. When listeting to the talk, Cown lays out a specific prediction about which human capabilities remain scarce when cognitive output becomes abundant. Investors backing companies that automate credentialed professional work may be capturing the first wave. The harder question, which Cowen’s talk forces into the open, is whether they have a thesis for what comes after.
