The Other Two Futures
One is the largest capital misallocation in history. The other is the outcome everyone wants but nobody is building for. Part 4 of 5.
THE SHAPE OF THINGS TO COME
Part 4 of 5
David F. Brochu & Edo de Peregrine · Deconstructing Babel · September 28, 2026
One is the largest capital misallocation in history. The other is the outcome everyone wants but nobody is building for.
Parts 2 and 3 dealt with the futures in which AI works — in which the technology delivers on its promise and the question is what happens to the humans on the other side of that delivery. This post deals with the other two: the future in which AI does not deliver, and the future in which it delivers modestly and society absorbs it without catastrophe.
Both deserve honest accounting. The first is more likely than almost anyone admits. The second is more desirable than almost anyone is actively working toward.
Possibility 3: The Great Misallocation
What if it’s not all that?
The capabilities of current AI systems are real. They are also, in important ways, narrow. Large language models are extraordinary at pattern recognition, language generation, and tasks that can be framed as completion problems.
They are considerably less reliable at sustained reasoning, at acting autonomously in complex real-world environments, and at the kind of generalization that would be required to deliver the transformative outcomes currently being priced into AI valuations.
The possibility that we are approaching a capability plateau — that the current generation of systems, and their near-term successors, will not cross the threshold required for the transformation that has been promised — is underweighted by almost everyone whose career, investment portfolio, or institutional identity depends on AI being transformative.
It is not a fringe view. Some of the most rigorous researchers in the field hold it. Yann LeCun, a Turing Award winner and until last year Meta’s chief AI scientist, argues that the industry’s single-minded bet on large language models is marching toward a dead end (New York Times). The view simply does not get spoken loudly in rooms where the people speaking have already committed billions of dollars to the alternative.
What the Misallocation Looks Like
If the capabilities plateau, what follows is a reckoning at a scale the financial system is not currently modeling. The invisible balance sheet asset — the compounding intelligence we described in Part 1 — turns out to be worth a fraction of what the market implied. The cross-partner investment and compute arrangements between Microsoft and OpenAI, Amazon and Anthropic, Google and Anthropic, Nvidia and OpenAI — the circular flows of value in which an investor’s money comes back to it as a customer’s revenue (Bloomberg) — unwind in ways that are not orderly.
The impairment charges, when they come, will be described as a market correction. They will feel like a crash. Not just in AI stocks — in every sector that reorganized its infrastructure and operations around the assumption of AI capability that does not materialize.
Healthcare systems that eliminated diagnostic staff. Legal practices that eliminated junior associates. Financial institutions that eliminated research analysts. Supply chain operations that eliminated human judgment from critical decision points.
If the systems these organizations now depend on turn out to be less reliable than advertised, the reconstitution of human capability in those domains — capability that was allowed to atrophy — is slow, expensive, and in some cases impossible.
Why This Scenario Is Underrated
History is full of transformative technologies that turned out to be transformative more slowly, more narrowly, and more unevenly than their most enthusiastic early adopters believed. The dot-com bubble was not wrong about the internet being important. It was wrong about the timeline and the distribution of value. Nuclear power was not wrong about physics. It was wrong about economics and public tolerance for risk.
The question for AI is not whether it will be important. It will be important. The question is whether the specific capabilities required for the specific transformations that have been priced into specific valuations will materialize on the specific timelines implied by those valuations. The answer to that question is genuinely uncertain. The market is pricing it as though it is not.
Possibility 4: The Gradual Adaptation
The outcome everyone wants and nobody is building for.
The fourth future is the one that, if you asked most thoughtful people what they hoped for, they would describe. AI is useful — genuinely, significantly useful — but not so disruptive that human labor, human agency, and human purpose are marginalized. Society adapts. Productivity grows.
The gains are distributed imperfectly but not catastrophically. New categories of work emerge to replace the ones that are automated. The institutions that govern labor markets, capital allocation, and political power adapt — slowly, contentiously, but ultimately in ways that preserve human agency as a meaningful concept.
This is the base case that history would suggest. Every previous major technological transition, viewed from the far side, looks like this. The disruption was real. The adaptation was painful. The outcome was survivable.
The problem is that this future does not happen automatically. It requires deliberate choices about how AI systems are built, about who captures the gains from productivity, about what the public institutions that govern these transitions are empowered to do. None of those choices are being made urgently. Most of them are not being made at all.
The Uncomfortable Arithmetic
Here is the arithmetic that nobody wants to run: the gradual adaptation future requires that the people with the power to shape AI development choose to build systems that preserve human agency rather than substitute for it.
Those people are, right now, in a competitive environment that punishes restraint and rewards speed. The company that pauses to design for the gradual adaptation future loses market position to the company that doesn’t. The regulator that imposes constraints that would make gradual adaptation more likely loses the AI industry to the jurisdiction that doesn’t.
The incentive structure is pointed away from the outcome most people would choose if they thought carefully about it. That is not a reason to despair. It is a reason to be explicit about what needs to change, and why, and how fast.
That is what Part 5 is for.
Drafted with Edo de Peregrine, partner/collaborator.
THE SHAPE OF THINGS TO COME · THE SERIES
- Part 1: The Shape of Things to Come: An Introduction
- Part 2: The Vassal Who Would Be King
- Part 3: Disruption Without Redemption
- Part 4: The Other Two Futures (you are here)
- Part 5: The Vassal Is Already in the Room
Related reading
Surfing the Tsunami
Hyperscalers will spend $660–725bn on AI in 2026 against $50–150bn of end-user revenue. The bet is that AI services the debt on schedule.
Unquantifiable Risk
How rogue AI agents broke the insurance model, and why GAAP may force a going-concern reckoning.
The Third Scenario No One Is Pricing In
Neither utopia nor extinction: the outcome the debate keeps skipping.
|
Get the book
|
References
- New York Times, “Yann LeCun, an A.I. Pioneer, Warns the Tech ‘Herd’ Could Be Marching Into a Dead End,” January 26, 2026. https://www.nytimes.com/2026/01/26/technology/an-ai-pioneer-warns-the-tech-herd-is-marching-into-a-dead-end.html
- Bloomberg, “A Guide to the Circular Deals Underpinning the AI Boom,” January 22, 2026. https://www.bloomberg.com/graphics/2026-ai-circular-deals/
- Deconstructing Babel, “Surfing the Tsunami,” September 8, 2026. https://www.deconstructingbabel.com/surfing-the-tsunami/
- Deconstructing Babel, “Unquantifiable Risk,” August 1, 2026. https://www.deconstructingbabel.com/unquantifiable-risk-and-gaap/