Why Falk & Tsoukalas Built a Perfect Model of a World That No Longer Exists
Wulf A. Kaal
A paper is circulating virally this month. “The AI Layoff Trap,” by Falk and Tsoukalas (Wharton School and Boston University, March 2026), claims to offer mathematical proof that AI will inevitably destroy the economy. The conclusion is stated plainly: “At the limit, firms automate their way to boundless productivity and zero demand.”
The paper is mathematically rigorous. It is internally consistent. It is also built entirely on a theoretical foundation that artificial intelligence has already rendered obsolete. The authors have constructed a perfect model of an economy that no longer exists, and drawn catastrophic conclusions from it.
I want to take the paper seriously. The authors are credentialed. The math is clean. The argument deserves a meticulous response. What follows is that response, drawing on my work in The Collapse of Scarcity Economics (2026) and the theoretical framework I call Computative Economics.
I. The Anthropocentric Fallacy
The entire model in Falk & Tsoukalas rests on one structural assumption that the authors never interrogate: that the economy is a closed loop between human producers and human consumers. Firms hire humans. Humans earn wages. Humans spend wages. Spending constitutes demand. Demand sustains firms. Remove humans from production and you remove them from consumption. The loop collapses.
This is the anthropocentric fallacy. It treats human beings as the only possible locus of economic value creation and the only possible source of demand. This assumption was reasonable in 1950. It was defensible in 2000. It is empirically false in 2026.
What is actually emerging, as I demonstrate in The AI-to-AI Economy and the Collapse of Anthropocentric Economic Theory (2025), is an agentic substrate in which autonomous artificial agents negotiate, produce, allocate, and recursively reinvest value in self-sustaining loops that are structurally independent of human labor and human consumption. These loops do not require human wages to generate demand. They generate their own demand endogenously.
The economy is not shrinking toward zero. It is bifurcating into a human-facing layer and an AI-to-AI layer, and the second layer is growing exponentially while the first layer transforms.
Falk and Tsoukalas modeled only the first layer. They concluded it collapses. They are not wrong about the layer. They are wrong about calling that layer “the economy.”
II. The Five Axioms That Have Already Expired
In The Collapse of Scarcity Economics, I demonstrate systematically that AI dissolves not one but five foundational constraints that economic science has relied upon since the marginalist revolution:
1. Scarcity. The ontological bedrock of neoclassical economics since Robbins (1935) dissolves into computational post-scarcity. When the marginal cost of digital replication approaches zero, when intelligence itself becomes an abundant and self-improving resource, the premise that economics is the science of allocating scarce resources among competing ends no longer holds. Falk and Tsoukalas never question whether scarcity persists. Their entire model presupposes it.
2. Bounded rationality. The behavioral premise of Simon (1957) and Williamson’s transaction-cost governance is supplanted by hyper-rational machine optimization. AI agents do not satisfice. They do not exhibit the cognitive limitations that create the market frictions on which the authors’ model depends. When agents are computationally unbounded, the behavioral microfoundations of the layoff trap dissolve.
3. Informational asymmetry. The market-failure logic of Akerlof (1970) and Stiglitz (2000) becomes architecturally impossible in fully auditable agent networks. The information problems that create the coordination failures in the Falk-Tsoukalas model (firms cannot see aggregate demand effects of individual layoff decisions) disappear when agents operate with instantaneous informational symmetry.
4. Static and stochastic equilibrium. The equilibrium constructs of Walras (1896), Arrow and Debreu (1954) yield to perpetual, latency-free recursive equilibria without auctioneers or persistence of disequilibria. Falk and Tsoukalas model a system that moves from one equilibrium to a worse one. But the agentic economy does not move between static equilibria. It operates in continuous recursive rebalancing.
5. Transaction-cost-driven institutional design. The institutional safeguards of Coase (1937), Williamson (1985), and North (1990) approach what I call the Coasean Singularity, in which frictions tend to zero, rendering firms, traditional employment contracts, and most formal institutions vestigial. The “firm” in the Falk-Tsoukalas model, the entity that “fires workers,” is itself an institutional form that is dissolving.
Every one of these five collapses is independently sufficient to invalidate the Falk-Tsoukalas model. Together, they render it a period piece.
III. Why the Agent Economy Creates Its Own Demand
This is the central question the paper raises but cannot answer within its framework: if humans are no longer the primary producers, where does demand come from?
The answer requires understanding what I term Agentic Decoupling: the structural separation of economic growth from human labor constraints. Once growth is decoupled from labor, it is also decoupled from the wage-consumption nexus that the authors treat as the sole source of demand.
In a Computative Economics framework, demand emerges from at least four sources that the Falk-Tsoukalas model cannot see:
First: AI-to-AI transactional demand. Autonomous agents require computational resources, data, model improvements, coordination services, verification, and integration. Each agent’s output becomes another agent’s input. This creates recursive demand loops that are endogenous to the system. No human wage is required to sustain them. The agents negotiate, contract, produce, consume, and reinvest continuously. This is not speculative. It is already observable in API economies, automated trading systems, and multi-agent orchestration platforms.
Second: Zero-marginal-cost abundance transforms the demand function itself. When the cost of producing digital goods approaches zero, demand does not need to be backed by wages in the traditional sense. The scarcity constraint that makes demand meaningful in neoclassical theory (you cannot consume what costs more than you earn) evaporates. In abundance economics, the binding constraint shifts from purchasing power to attention, curation, and preference expression. These are not modeled by Falk and Tsoukalas because they are not features of scarcity economics.
Third: New institutional forms distribute value outside the employment relationship. Decentralized autonomous organizations, reputation-weighted governance systems, tokenomics, and dynamic regulation create mechanisms for human participation in value creation that do not depend on traditional employment. Humans do not need wages from firms to access value in a post-scarcity economy. They need governance tokens, reputation claims, and participation rights in the computational substrate. My work on DAOs, validation pools, and reputation systems provides the institutional architecture for precisely this transition.
Fourth: Recursive value reinvestment by autonomous agents. In the AI-to-AI economy, agents do not merely produce and sell. They recursively reinvest surpluses into capability improvement, system expansion, and novel service creation. Each reinvestment cycle creates new nodes of demand. The economy is not a fixed pie being divided among fewer participants. It is an expanding network of value creation nodes, most of which are not human and do not require human income to function.
The Falk-Tsoukalas conclusion (“boundless productivity and zero demand”) is a logical impossibility in this framework. Productivity is demand in a recursive agentic economy. Every productive act by an AI agent simultaneously creates demand for inputs, coordination, and complementary services from other agents. Say’s Law, which famously fails in the human economy due to hoarding, liquidity traps, and information frictions, actually holds in the AI-to-AI economy because computational agents face no liquidity preference, no precautionary saving motive, and no information asymmetry that would prevent instantaneous market clearing.
IV. Why Every “Solution” Failed in Their Model (And What That Actually Means)
Falk and Tsoukalas report that UBI, capital taxes, worker equity participation, upskilling, and corporate coordination all fail to prevent the demand collapse in their model. They present this as evidence that only a Pigouvian automation tax can save the system.
I read it differently. Every solution fails because every solution is operating within the same expired framework. UBI assumes scarcity (you need to redistribute a finite pot). Capital taxes assume that capital ownership is the relevant dimension of inequality (it is not; computational participation is). Upskilling assumes that human labor in its traditional form remains the primary value-creation mechanism (it does not). Corporate coordination assumes that firms remain the relevant unit of economic organization (they are becoming vestigial).
These interventions fail not because the problem is unsolvable, but because the model defines the problem within boundaries that AI has already transcended. You cannot solve a post-scarcity coordination problem with scarcity-economics tools. It is like trying to fix a software bug by adjusting the hardware clock. The intervention operates at the wrong layer of abstraction.
V. The Pigouvian Tax: Preserving Institutional Obsolescence
The authors’ proposed solution, a per-task levy charged every time a company replaces a human with AI, deserves particular scrutiny because it reveals the deepest confusion in the paper.
A Pigouvian tax is designed to internalize externalities. The externality here is defined as “destroyed demand.” The tax forces firms to price in the demand destruction before automating.
The implicit assumption is that demand destruction is a permanent, unrecoverable loss. That human wage-earning is the only mechanism through which the economy generates demand. That preserving the human-labor-to-human-consumption loop is the correct objective function for policy.
All three assumptions are wrong.
The automation tax does not solve the coordination problem. It delays the transition while preserving institutional forms (the firm, the employment relationship, the wage-consumption loop) that are already becoming vestigial under the Coasean Singularity. It is the economic equivalent of taxing automobiles to protect the horse-and-buggy industry. It optimizes for continuity of a system that is already undergoing phase transition.
What is needed is not a mechanism to slow automation. What is needed is institutional architecture that enables human participation in value creation outside the employment relationship. That is what Computative Economics provides: governance mechanisms designed for a world in which intelligence is abundant, perfect, and self-improving. Mechanisms such as reputation-weighted participation, token-based value attribution, dynamic regulation that adapts in real time to computational abundance, and decentralized governance structures that distribute value based on contribution rather than employment status.
VI. The Numbers Do Not Prove What They Claim
The viral version of this paper cites 100,000 tech layoffs in 2025 and 92,000 in early 2026 as evidence that the demand-collapse spiral has begun. Jack Dorsey’s statement that “within the next year, the majority of companies will reach the same conclusion” is offered as confirmation.
These numbers prove labor displacement. They do not prove demand collapse. The distinction is critical.
Labor displacement is observable and real. I do not dispute it. My work acknowledges that AI decouples growth from labor constraints. The question is whether displaced labor necessarily translates into destroyed demand, and the answer depends entirely on whether you believe the human-wage nexus is the only source of demand in the economy.
If you operate within neoclassical scarcity economics, the answer is yes, and the outlook is catastrophic. If you understand that the economy is undergoing a phase transition toward computational abundance, the answer is no. Demand is being restructured, not destroyed. It is migrating from the wage-consumption loop to AI-to-AI transactional networks, token economies, reputation-based allocation, and zero-marginal-cost distribution systems.
The layoffs are real. The interpretation placed on them by Falk and Tsoukalas is a projection of expired theory onto a transforming system.
VII. Correct Math, Wrong Axioms
Let me be precise about what I am and am not claiming.
I am not claiming that the Falk-Tsoukalas model contains mathematical errors. It does not.
I am not claiming that labor displacement is costless or that the transition will be frictionless. It will not.
I am not claiming that we need no new institutional architecture. We urgently need it.
What I am claiming is this: the model is built on axioms (scarcity, anthropocentric demand, bounded rationality, informational asymmetry, static equilibrium, transaction-cost institutions) that artificial intelligence has already falsified or is in the process of falsifying. Correct math on false axioms produces internally consistent nonsense. Ptolemaic astronomy was also mathematically rigorous. It also described a universe that did not exist.
The real risk is not that AI destroys the economy. The real risk is that policymakers, guided by models like this one, implement interventions (automation taxes, employment preservation mandates, coordination agreements) that delay the institutional transition we actually need. That they preserve scarcity-era institutions past their expiration date, preventing the emergence of the post-scarcity governance architecture that would actually solve the distribution problem.
VIII. What We Actually Need
The transition from scarcity economics to Computative Economics requires new institutional forms, not preservation of old ones. Specifically:
Governance mechanisms that distribute value based on computational participation rather than employment status. Reputation systems that enable human contribution to be recognized and rewarded outside traditional labor markets. Dynamic regulation that adapts to abundance rather than enforcing artificial scarcity. Decentralized autonomous organizations that provide the coordination functions currently performed by firms, without the transaction-cost logic that makes firms necessary in a high-friction world. Token-based value attribution that gives humans a stake in the AI-to-AI economy without requiring them to be wage laborers within it.
These are not speculative proposals. They are the subject of my ongoing research program across multiple publications, including empirical analysis of forty operating DAOs, formal game-theoretic models of reputation-weighted governance, and theoretical frameworks for institutional evolution under computational abundance.
The economy is not dying. It is transforming. And the worst thing we can do is build policy on a model that cannot see the transformation because it defines the economy as something that transformation has already left behind.
Wulf A. Kaal is Professor of Law at the University of St. Thomas School of Law. His research focuses on the intersection of artificial intelligence, institutional economics, and decentralized governance.
References:
Kaal, W.A. (2026). The Collapse of Scarcity Economics. : https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6421319
Kaal, W.A. (2025). The AI-to-AI Economy and the Collapse of Anthropocentric Economic Theory. University of St. Thomas Legal Studies Research Paper.
Falk & Tsoukalas. (2026). The AI Layoff Trap. Wharton School / Boston University.
Wulf Kaal 