The Relational Economy Can't Save Us in the Messy Middle

In my last post, I argued that the most consequential political economy questions of the AI era will be decided in what I called the “messy middle”: the long, hard stretch between today’s mostly intact labor market and the post-AGI world of abundance that Silicon Valley keeps promising is just around the corner. The piece pushed back on the binary that lets people skip past this period entirely, either by insisting policy is premature or by promising UBI will solve everything once the surplus arrives.

Since publishing it, I’ve been thinking about the most thoughtful version of the optimistic counter-argument: that the relational economy will absorb workers displaced from cognitive work. Alex Imas has made this case in his essay “What will be scarce?”, and Ezra Klein leaned on it heavily in his Sunday column. I admire both of them, and I share Alex’s core intuition that AI commoditizes cognition and that relational work is where humans retain comparative advantage. When I think about my own three kids’ future, I put a huge premium on interpersonal skills for that very reason.

But I think the relational economy framework, powerful as it is for the long run, breaks down precisely in the period that matters politically. The next five to ten years are where the politics of AI will be made, not the post-AGI equilibrium. And in that window, I worry that the relational economy can’t do the work Alex and Ezra are asking it to do, without government intervention.

The relational economy is downstream of professional incomes

Alex’s structural change story requires a specific sequence: AI cheapens commodity production, prices fall, real incomes rise, and rising incomes shift demand toward the relational sector with its high income elasticity. Every step depends on the prior one. The model needs aggregate income to actually rise, or at least hold, for the demand reallocation to do its work.

But the relational economy as we know it today is overwhelmingly funded by professional-class disposable income. An example I keep returning to: in February, I was in London for a scenario planning workshop hosted by the Windfall Trust. The near-term scenario we grappled with has stayed with me. Imagine 2030, and a third of all jobs in the top 20 percent of the London labor market are lost or substantially devalued. London is dominated by professional services. It is those earners who buy the expensive theater tickets, the hands-on therapy, the in-person yoga classes, the private school places with the most engaged support staff.

My dear friend is a priest in north London. Her job is exactly the kind of deeply relational work AI can never do. But her parishioners, and the groups who rent the church for kids’ events, are overwhelmingly the professional families now vulnerable to displacement. Another friend is the headmistress of an elite private school in London. Nothing about her job can be done by AI, it is 100 percent relational. But if a meaningful share of her parents lose their jobs, how many will continue to send their children to that school?

Take a hatchet to professional incomes and you get contraction in the relational economy, because its customer base just collapsed. The relational sector cannot grow as the cognitive sector shrinks — if the cognitive sector is what funds the relational sector in the first place.

AI cheapens the wrong things in the medium term

The other half of Alex’s mechanism is a price story I do not think holds in the messy middle. We have already run a partial version of this experiment. Manufactured goods have collapsed in real price over the last thirty years through globalization and automation. China did much of what AI is supposed to do for cognitive work. And yet middle-class economic anxiety has increased over that period, not decreased.

The reason is structural. The categories that eat household budgets are not the manufactured goods that got cheaper. They are housing, healthcare, education, childcare, elder care, and the costs that come with raising kids: camps, after-school activities, all of it. Mark Perry’s chart on price changes since 2000 shows this with brutal clarity: televisions and toys down 80 to 90 percent in real terms, while hospital services, college tuition, and childcare are up 100 to 200 percent.

Here is the uncomfortable observation: the categories that resist AI automation in the medium term are largely the same categories that already dominate household budgets and that already have the highest cost growth. Alex would say this is exactly his point, those are the relational sectors that grow as commodities cheapen. But these sectors are already dominant in household spending, the wages funding them come disproportionately from the professional class, and AI does not meaningfully cheapen them on the timeline when those incomes might face a shock.

A displaced professional in the messy middle faces a triple squeeze: income loss in the cognitive work being automated, a cost base dominated by goods AI does not meaningfully cheapen short of full AGI, and a real risk that those costs rise faster as Baumol effects intensify. The optimist story requires that AI cheapen the right things on the right timeline. I do not see that as likely.

Why Ezra’s column shows the trap

Ezra’s column from Sunday illustrates the problem I am trying to surface, and I want to be careful about why. Ezra explicitly rejects the full automation scenario. He says he doesn’t believe mass unemployment is likely, and he is skeptical of the AI labs’ own projections. By his own framing, he is putting himself in the messy middle. He even articulates its distinctive cruelty better than most, citing the China Shock as a case where limited displacement got ignored while mass disruption would have forced a response. That is exactly my concern.

But then he turns around and endorses Alex’s relational economy story as the reassuring counter-narrative, without noting that Alex’s framework is a long-run equilibrium argument that assumes the post-AGI conditions Ezra has just dismissed as unlikely. The examples he reaches for, Jevons and coal, VisiCalc and accountants, Starbucks, all come from a period of broadly rising or stable professional incomes. That is precisely the condition that may not hold in the scenario he is actually worried about.

I take Ezra seriously, and he is in good faith pushing back against the apocalyptic framing. But the column shows how easy it is, even for a careful generalist, to set up the messy middle and then reach for a long-run framework to address it.

What I think could work

If the relational economy will not grow on its own during the messy middle, the obvious question is whether anything could fill the gap. I want to be honest about what I believe is and is not feasible.

Full income replacement for displaced professionals may not be politically feasible in this period. The idea that the federal government will backstop $200,000 policy analysts and consultants at their prior incomes (beyond temporary unemployment benefits, and perhaps wage insurance for older workers who genuinely cannot retrain) runs against every signal from the current political environment. It risks being perceived as an elitist bailout by the much larger share of the country that never earned those wages. And a universal redistribution large enough to maintain professional-class lifestyles without working would also remove the economic motivation for the people currently doing the essential work society depends on: home health aides, line workers, nurses, early childhood educators, the people who fix telephone poles after a storm. That risks the collapse of labor supply for the work the economy actually requires.

What I do think is feasible, and where I believe the policy energy should focus, is direct public investment in the relational sectors themselves. The deliberate creation of decent, well-paid jobs in care, education, public health, early childhood, elder care, mental health, and the public-facing infrastructure of civic life. These are sectors where demand exists but is unmet because we systematically underfund them, and where the work cannot be automated. Funded partly through taxation of AI gains, public investment here would absorb workers, strengthen the institutional fabric that the relational economy depends on, and do so without setting an income floor so high that essential work goes undone.

This is not a complete answer. Nothing is. But it has a structural logic the market story does not: it does not require the relational economy to grow on the back of a customer base that may itself be contracting. It builds the relational economy directly, as a public good, in the period when the market alone will not.

I do not pretend to have the full answer. But I am increasingly convinced that the answers we need will have to be built for the period we are actually entering, not borrowed from the equilibrium we hope eventually arrives. That is the work I want to keep doing, and I am grateful to everyone wrestling with these questions alongside me.