The Counter-Thesis: Ezra Klein and the Relational Sector

Why does AI make human connection more economically valuable, not less?

Because as cognition becomes abundant, the bottleneck moves to the human element: the parts of an exchange that depend on someone actually understanding you. The economist Alex Imas calls this the relational sector, and Ezra Klein has argued that AI productivity gains will expand it rather than collapse it. Rodin sits one step downstream of that argument; it builds the locator for the people you would otherwise never know to call.

In early May, Ezra Klein published a long piece in the New York Times under the title “Why the AI Job Apocalypse (Probably) Won’t Happen.” It opens, almost theatrically, with a billboard in San Francisco that reads “Stop Hiring Humans,” and the easy reading of that billboard, the reading the labour-displacement discourse has been training us into for two years, is the one Klein resists. The piece is structured around a counter-argument he borrows from Alex Imas at Chicago, the historian Robert Allen on the Industrial Revolution, and Ethan Mollick’s working test for whether a given AI tool is actually worth using, which is whether it is better than the best human you can call.

That last formulation is the one I want to keep returning to. It is small and it sounds modest, but it is doing more than it looks like it is doing. The best human you can call is not the best human in the world, and it is not the best human in your field; it is the one who is, at the moment you need help, inside your callable set. The argument of this post is that Klein names the demand-side hunger of the relational sector with great precision, and that the assumption hiding inside Mollick’s test — that your callable set is given — is the one place his account stops short.

The scarcity sequence

Imas’s framework, as Klein presents it, is a kind of historical recursion. Every wave of productivity has dissolved one bottleneck and exposed another. Agriculture made calories abundant and turned the bottleneck into goods; the Industrial Revolution made goods abundant and turned the bottleneck into expertise; the long professional century after that made expertise itself the scarce thing, which is why doctors, lawyers, designers and analysts have, for as long as anyone alive can remember, been the people who got paid. The thesis Imas advances, and that Klein finds persuasive, is that AI is now collapsing the price of expertise in the same way the steam engine collapsed the price of cloth, and that what becomes scarce in the next phase, and therefore valuable, is what AI cannot in fact do, which is the human element.

The empirical argument behind this is the income elasticity finding that Imas points to: as people get richer, they spend a disproportionate share of the additional income on services where the human element matters. They want the therapist who knows them, the tutor who knows their child, the doctor who has actually been listening, food with provenance, clothes with a story, restaurants where someone behind the counter recognises them. The expansion is not sentimental; it is, in Imas’s reading, the basic econometric shape of post-scarcity demand. The relational sector is what people choose to want once the prior bottlenecks have eased.

The Jevons recoil

Klein’s second move, and the one that makes the piece more than just a productivity argument, is to apply the Jevons paradox to cognition itself. Jevons noticed in the 1860s that as steam engines became more efficient, coal consumption rose rather than fell, because the cheaper resource opened up uses for it that had been uneconomical before. The same recoil, Klein argues, has been the historical pattern with cognitive tools. VisiCalc did not eliminate accountants; it quadrupled them, because suddenly every business that had been doing without an accountant could afford one. The paradox is that making a resource cheaper expands the contexts in which the resource is used, which expands the demand for the human practices that surround the resource.

What is striking about Klein’s argument is that he runs the Jevons recoil through his own week. The AI flagged a symptom and he booked an appointment with his doctor. The AI offered a personal insight and he opened a new thread with his therapist. The AI validated a research angle and he brought new questions to his editor. The AI made video subtitling cheap and he hired more video editors. In each case, the cheaper cognition did not displace the human; it intensified the demand for the specific human who had been there all along. The piece, in effect, is not about how AI substitutes for human relationships but about how AI amplifies the want for them, in the way that a cheaper microscope creates more biologists, not fewer.

The atomization paradox

Halfway through the essay, Klein performs the move that gives the piece its weight. Having argued that the relational sector is about to expand, he notes that the relational capacity of the people who would have to staff it is, on every measure, collapsing. Friend time has fallen from twelve hours a week in 2003 to five hours a week in 2024. The proportion of high school seniors who date has fallen from eighty percent in 2000 to forty-six percent in 2024. About a quarter of Gen Z reports no sexual activity in the past year. Klein cites these figures not as a moral panic but as a structural problem: the most economically valuable skill class of the next twenty years is being eroded in the very cohort that is supposed to embody it.

His worry, and it is a careful one, is that AI does not have to displace humans en masse to do real damage; it only has to displace them at the margin where relational capacity is already fragile. The young man who is alone in his bedroom does not need an AI girlfriend that is good; he needs one that is good enough to make the costly project of a real relationship feel optional. What Klein calls the danger of “digital simulacra of friendship and love” is the supply-side hollowing of the relational sector, which is happening in parallel with the demand-side expansion. The two trajectories chime with each other in a way that is, in his account, the actual political-economic problem of the next decade.

The best human you can call

Mollick’s test, the one Klein keeps returning to, is the load-bearing image. Is the AI better than the best human you can call. The question is decisive because it is honest about how expertise is actually accessed in a life. Most of the time, the best human in the world for your problem is not in your callable set; what you have is whoever you can text on a Tuesday at four in the afternoon and reasonably expect a reply from before Thursday. Mollick’s test, in this sense, is not a global comparison between AI and human cognition; it is a local one, between AI and the cognition that is, for you, accessible.

What Klein’s essay does not name, however, is that the callable set is not given. It is the result of a long, mostly invisible accumulation of relational infrastructure: who you went to school with, who happened to overlap with you in your twenties, whose number you happened to keep, whose email you can write without it being awkward. The callable set is largely a function of biography, and biographies, as the atomization data Klein himself cites makes clear, are getting thinner. What might be described as the demand-side argument of the relational sector, namely the claim that the human element is becoming more valuable, is operating on a callable set that, for many people, is structurally incomplete in ways neither they nor anyone else can quite see.

This is where the analogy from Mollick’s test breaks down, and the break is illuminating. The interesting question for the next decade is not only whether the relational sector grows; it is whether the discovery layer for it exists. A person can have intact relational capacity and still meet a wall, because the human who would, in fact, be the most useful person to call is someone they have never heard of, and the existing infrastructure for finding such a person (LinkedIn, Twitter, the conference circuit) is built around credentials, audience, and proximity rather than around how somebody actually thinks. Klein names the hunger; the question of where the locator goes is, in effect, the unsaid of his piece.

The wager underneath Rodin is that the bottleneck of the relational sector, in the long arc Klein describes, is not relational capacity in the abstract but discovery. It is the question of how the best human you can call comes to be in your callable set in the first place. The fingerprint Rodin extracts is not a profile in the LinkedIn sense; it is a description, in the user’s own words and intellectual movements, of how a particular mind approaches its open questions, including its themes, its models, its blind spots, and the archetype the writing has been quietly building. Two such fingerprints can recognise each other in a way that no credential layer can replicate, because what they recognise is not what either person has already done but how each is in fact thinking. Whether that is enough to enlarge the callable set in any serious way, against the slow erosion of relational capacity Klein worries about, is, of course, the question I do not yet know how to answer. Klein names the hunger. What is left to build is the part of the infrastructure that knows where to look.