Which Future Are You Building Toward? The Four Futures of AI

Delia Reyes wakes at 11:47 a.m. In one version of the next decade, that's a tragedy. In another, it's the morning she's been waiting her whole working life for. The difference between those two mornings is not technology. It's a set of decisions being made this year, mostly by people who've never heard her name.

Some of them were in Paris. On February 10, 2025, delegates from 61 nations gathered at the Élysée Palace for the AI Action Summit; 58 signed a joint declaration on inclusive, sustainable AI development. The United States and United Kingdom declined. That refusal wasn't a technical disagreement. It was a choice about which institutional actors would shape the rules — and whose interests those rules would serve.

The tension: we keep arguing about the wrong variable

In 2016, a political theorist named Peter Frase published a slim, deliberately contrarian book called Four Futures. It earned admiring reviews, modest sales, and quick burial under the news cycle. Nobody took it seriously as a forecast — Frase wrote it as provocation, not prophecy. Nine years later, its central insight is the actual architecture of fights in legislative chambers and AI labs.

What Frase saw clearly is that the critical variable is not the technology. Technology generates capacity. The institutional architecture for distributing what that capacity produces is what matters. That's so obvious once stated that it's easy to underestimate how thoroughly it contradicts the dominant framing of AI discourse, which treats the technology as the primary subject and institutions as secondary constraints on it. The conventional debate asks how powerful AI will get. The more useful question is who captures what it produces.

The reframe: same row, different owner

Frase organized the choice along two axes. One: whether productivity gains are broadly shared or narrowly captured. Two: whether social hierarchy diminishes or compounds. Four quadrants, four futures.

Figure: The Four Futures matrix. Shared Abundance and Rentism sit in the same row — the same high-productivity economy. The only thing that changes between them is who owns the surplus.

• Shared Abundance: high productivity, broadly distributed. Human contribution shifts toward creativity, care, civic participation, and meaning. For Delia, the morning she gets up on her own terms.
• Managed Transition: high productivity, distributed through active intervention — retraining, income floors, democratic governance. Not utopia: scarcity persists, struggle continues, hierarchy bends without breaking. The most tractable near-term target, because it doesn't wager everything on abundance arriving on schedule.
• Rentism: high productivity, narrowly captured. AI generates real surplus, but it accrues to owners of infrastructure — compute monopolies, platform lock-in, IP regimes. The formal economy hums; most people depend on platforms whose terms they can't negotiate. For Delia, the income floor might be real. The ceiling is permanent.
• Exterminism: scarcity intensifies, hierarchy compounds, and the displaced population becomes economically superfluous — not oppressed in any legible sense, simply irrelevant. For Marcus Webb, not unemployment but irrelevance.

The geometry is the point. Rentism and Shared Abundance are in the same row. The same high-productivity, AI-driven economy generates both. One square to the left, Shared Abundance collapses into Rentism, and nothing else has to move. This is why the distributional question — not the productivity question — is the actual hinge of the next decade.

How it works: the evidence, and why standing aside is a vote

Read honestly, the evidence points toward Rentism as the default. NVIDIA held roughly 90% of the AI GPU market in 2024. Frontier models are trained by three or four firms. The freelance designer using Adobe Firefly, Shopify, and Stripe earns more than before — but the surplus her AI-augmented work generates flows as margin to three platforms whose terms she can't individually negotiate and whose rails she can't practically leave. That's rent in the classical sense: income from owning scarce infrastructure, not from value created. Block's February 2026 restructuring — 4,000 jobs cut, shares up 24% on the announcement that AI agents would coordinate the work — shows where the organizational logic leads when each firm's rational decision becomes the system's destabilizing one.

There's a countervailing force worth naming: monopolies break. Open-weight models have put capable AI in the hands of those who can't afford frontier pricing, and inference costs have fallen dramatically — by Sam Altman's own data, token costs dropped roughly 150× in 18 months. That's real. It's just not yet strong enough to change the structural calculus.

The thoughtful skeptic concludes from all this that Rentism is inevitable, because it requires nothing — no new institutions, no new coalitions, no governance architecture. Shared Abundance and Managed Transition both require deliberate action. I think that view is half right, and the half that's right is more concession than most writing in this space offers. Coordination at scale is genuinely hard; infrastructure is concentrating, not dispersing.

The half that's wrong rests on a fixed-sum assumption about human nature. Those same institutions looked equally unreformable in 1780, 1905, and 1932 — at every juncture when they were in fact about to be reformed. Abolitionists operated against what observers called immutable economic structure; suffragists were told, with genuine certainty, that society couldn't accommodate what they asked. Both dismissals mistook the conditions of a moment for permanent features of human nature. The finding that roughly 25% of workers believe their jobs are socially meaningless isn't just data about alienation — it's evidence of misallocated coordination capacity that could be redirected if the institutional channel existed. The School for Moral Ambition drew more than 10,000 applications for a few hundred places, mostly mid-career professionals who concluded another year in their current role was worth less than redirecting toward something harder.

The cynic who concludes from the difficulty of coordination that coordination shouldn't be attempted isn't clear-eyed. The cynic is paralyzed. And in this particular game, paralysis is a vote for Rentism. Standing aside is already a bet — and that bet pays out for the default.

What to do

The people building this technology can't agree on what the transition requires, but none of them believes the current default is adequate. Dario Amodei: "We need to stop sugar-coating what's coming." Altman concedes capital-labor balance "could easily get disrupted and may require early intervention." Musk proposes "Universal HIGH INCOME" precisely because he expects AI-driven unemployment at scale. Three diagnoses, none arguing that the trajectory is fine. That's the signal.

So the question isn't "What will happen?" It's "What are you building toward?" — not in principle, but in the actual design choices being made this quarter.

1. Embed values before defensibility. The moat amplifies whatever sits in the values layer beneath it. The Steward-Ownership: How Founders Lock Mission Into the Architecture models exist to prevent the predictable failure: the Series C closes, investors arrive with return expectations, and a values layer with no structural teeth bends back toward extraction. Redeemable shares, purpose-aligned funds, Delaware PBC status with transfer restrictions are available now and close progressively as valuation rises.
2. Begin transition infrastructure at the first automation decision. Not 18 months later when headcount has already contracted. The The Three-Pillar Bridge — reskilling, portable benefits, income floors — takes 18–24 months of institutional plumbing to build. Delia's income floor is what kept her financially stable long enough to still be at the food bank when a paid coordinator role appeared. The Bridge didn't give her purpose. It gave her time to still be in the room when purpose arrived.
3. Use procurement as a governance instrument. Conditioning public AI contracts on stakeholder-voice audits, alignment-debt transparency, and portable-benefits standards costs nothing in new legislation. It makes what governments already buy contingent on how suppliers treat affected people.
4. Govern the concentration your moats create. Participation dividends and steward-ownership redistribute the surplus without eliminating the moat. They're not opposites — they're two sides of the same architecture.

The technology enables all four mornings. It does not prefer any of them. But the structural forces aren't symmetric: Rentism requires only that current trajectories continue, while the better futures require deliberate construction. Every human who builds, governs, regulates, or simply uses this technology is already casting a vote between two organizing logics — extraction or stewardship — whether they know it or not. The Machine Core executes. The Human Cortex chooses what's worthy of execution. That relationship is the one thing machines cannot scale.