Starbucks spent years removing workers from its stores, convinced that automation and mobile ordering efficiency would protect its margins. The bet failed and in late 2024, incoming CEO Brian Niccol reversed course , calling for more baristas, handwritten cup notes, and ceramic mugs because, as Niccol put it, "This is still a craft business." That reversal, at one of the world’s most standardized consumer companies may be a preview of the economic structure that AI is about to build.

A new essay by Alex Imas, behavioral economist at the University of Chicago Booth School of Business, argues that the standard fear about AI and employment gets the structural economics backward . Automation reallocates jobs into sectors where the human producer is part of the product's value, a dynamic Imas calls the relational sector. The research carries direct implications for where venture capital should be looking as AI commoditizes knowledge work.

Investors have poured capital into AI infrastructure on the premise that automating cognitive labor creates durable value. Global VC funding for AI startups reached $131.5 billion in 2024 , representing one-third of all venture investment that year. But the thesis embedded in most of those deals treats the economy as static: automate a task, capture the margin. Imas's framework suggests the more important question is what spending patterns look like once commodity production gets cheap, and whether the sectors absorbing that spending are ones venture capital has yet to price correctly.

The historical analogy is agricultural productivity; farm employment in the U.S. collapsed from roughly 40% of the workforce in 1900 to under 2% today, not because people stopped eating, but because rising farm productivity lowered food’s share of total spending. Workers and dollars moved elsewhere and Imas argues AI will trigger an identical reallocation away from automatable commodity production and toward what he identifies as high-income-elasticity sectors, places where spending grows faster than income as people get wealthier.

The empirical backbone comes from Comin, Lashkari, and Mestieri’s 2021 Econometrica study , which estimated that income effects account for more than 75% of observed structural change across economies. As incomes rise, consumers do not buy proportionally more of everything; their spending tilts toward sectors with higher income elasticity. The ironic consequence Imas draws out is that the more productive an automated sector becomes, the smaller its share of GDP, not larger.

Mimetic Demand and the Relational Sector

The behavioral mechanism Imas identifies is mimetic desire, drawing on Rene Girard’s theory that human wants are fundamentally social and comparative. People do not simply desire goods for their intrinsic properties. They want what others want, and they want it more when others cannot have it. Imas and co-author Kristof Madarász tested this experimentally: willingness to pay roughly doubled when a random subset of consumers was excluded from a good . The good itself was unchanged but the perceived value did.

In a separate study with Graelin Mandel , Imas found that AI involvement eliminates the exclusivity premium. Human-made art gained 44% in value from exclusivity; AI-made art gained only 21%. Human provenance is scarce by definition, because the scarcity comes from the humans themselves.

The sectors with these properties share a common feature: the producer is not separable from the product. Personalized medicine, bespoke professional services, artisanal food and hospitality, live performance, high-touch education, and complex care all fit the profile. These are categories Engel curve dynamics push spending toward as base needs get cheaper to meet.

What This Means for Founders and Investors

Imas is precise about the limits of his claim. He is not arguing that aggregate labor share must rise, or even hold steady. It may fall. The companion technical note works through the formal model and its conditions. The claim is about the sectoral: where spending and employment migrate once AI makes commodity production cheap, and whether those destination sectors structurally require human involvement.

For venture capital, the framework identifies a mispricing risk. Funds that built their AI theses on displacing human labor across automatable sectors may be correct about the displacement but wrong about where value accrues afterward. The sectors with the highest forward income elasticity are ones that combine human provenance with scalable infrastructure: platforms that verify, surface, or distribute human-originated goods and services without commoditizing them. The analogy may not be Airbnb displacing hotels but rather a set of businesses that make it economically viable to be the human that guests specifically want.

Starbucks is an imperfect but instructive data point. The company has a market capitalization exceeding $90 billion. Its leadership publicly concluded that removing human labor from the customer experience reduced rather than created value, even when the underlying product, espresso, is trivially automatable.

The economic question AI poses is not, as it is usually framed, whether machines can do what humans do. In most production categories, they already can, or will soon. The more consequential question is what consumers choose to spend on once that production becomes abundant. Imas's answer, grounded in both economic history and behavioral evidence, points toward scarcity that automation structurally cannot eliminate: the human in the transaction. For investors with capital to deploy, that is the market to map.