This is a long-form essay summarizing the core arguments of two research papers, for readers interested in how an AI economy may transform the structural logic of market systems. Because both papers are technical and academic in style, this note offers a more readable summary — though some technical language is unavoidable.
Introduction: An Old Question in a New Form
In 1885, Karl Marx published Capital, Volume II—arguably the least-read volume in his trilogy. It has none of the dramatic language of exploitation and none of the literary force of commodity fetishism. Instead, it asks a dry but profound technical question: can capitalism reproduce the conditions required for its own existence?
Nearly 140 years later, that question is becoming urgent again—not because of proletarian revolution, but because artificial intelligence is doing what Marx only anticipated in outline: pushing the productive forces to a point where existing relations of production may no longer hold.
The two research papers summarized here revisit this question using contemporary economic tools. The first, Stability without Reproduction (2025), poses the problem. The second, Income Regimes After Wage Anchoring (2026), develops an answer. This essay summarizes the core logic of both and explores the philosophical and political implications.
Part I: Stability without Reproduction — When the System Stops Reproducing Itself
The Core Problem
The first paper begins with a simple observation: capitalism not only produces goods; it also reproduces purchasing power. In the standard cycle—production → wages → consumption → production—firms hire workers, workers earn wages, wages become consumption, and consumption returns as firm revenue. The cycle continues.
Where does AI break this cycle? Not at production. AI can raise production capacity. The break appears in the income channel: as AI substitutes for labor, aggregate wage income contracts. Output rises, but mass purchasing power weakens.
The paper formalizes this with a minimal setup: let \(Y\) be aggregate output, \(W = wL\) wage income, \(\Pi = Y - W\) profits, \(\theta\) the share of profits redistributed to households, and \(T\) state transfers. Then aggregate demand is \(D = c(W + \theta\Pi + T)\), where \(c\) is the marginal propensity to consume.
As AI raises productivity \((A)\), output grows \((dY/dA > 0)\), but wage income falls \((dW/dA < 0)\). If \(\theta\) is low and \(T\) does not scale, aggregate demand can decline even while productive capacity expands.
Stabilization ≠ Reproduction
The central contribution of paper 1 is to separate two concepts that are often conflated.
Stabilization means preserving macro stability through external intervention. The state injects transfers to close the demand gap. This can work, but it must be ongoing and, in the model, must rise with productivity \((dT/dA > 0)\).
Reproduction means the system generates demand endogenously—through a wage channel that automatically links production and purchasing power, without permanent external management.
The conclusion is: in a post-AI setting, an economy may remain stable while no longer reproducing itself internally. Stability becomes externally maintained. Hence the title: Stability without Reproduction.
Marx and Keynes in the Framework
The paper does not choose sides between Marx and Keynes; it uses both analytically under new technological conditions.
For Marx, the focus is not exploitation or class morality, but reproduction in Capital II: can the system reproduce its own conditions? The model is not Marxian by assumption, but it is Marxian in implication. The issue is internal coherence, not distributive ethics.
For Keynes, the paper accepts the power of Keynesian economics in explaining demand insufficiency, but argues that its domain of applicability narrows under AI-intensive production. The Keynesian transmission mechanism (fiscal expansion → employment → income → consumption) presumes labor remains the main link between production and income. As AI weakens that link, stabilization may still work, but the mechanism's domain of applicability contracts.
Main Weakness
The key weakness of paper 1 is clear: \(\theta\) is set low exogenously. That effectively embeds the result in the assumptions. If \(\theta\) were high (through broad equity ownership or profit sharing), the diagnosis could change. So the right question becomes: is \(\theta\) truly exogenous, or should it be endogenized?
That is exactly what paper 2 attempts to do.
Part II: Income Regimes After Wage Anchoring — Why Low \(\theta\) Becomes Structural
From Parameter to Strategic Choice
Paper 2 endogenizes what paper 1 took as given. Instead of fixing low \(\theta\), each firm chooses its own \(\theta_i\). The question is: under decentralized competition, where does the system converge?
The answer unfolds in two steps.
Step One: Coordination Failure — Nash Equilibrium at \(\theta = 0\)
Redistribution is modeled as a game in which each firm selects \(\theta_i\). Each firm is small relative to the system, so its unilateral \(\theta_i\) has negligible effect on aggregate demand.
Result: increasing \(\theta_i\) alone does not materially improve demand, but it does reduce that firm's retained earnings. So the best response is \(\theta_i = 0\).
Even if \(\theta = \theta^*\) (the demand-sustaining level) is Pareto superior, it is not a Nash equilibrium because each firm has an incentive to deviate unilaterally. This is a system-wide prisoner's dilemma logic.
Step Two: Darwinian Drift — Even High \(\theta\) Erodes Over Time
The second step is even more forceful. Suppose high \(\theta\) is imposed at the start—through law, social norms, or institutional legacy. Can it persist?
The paper uses replicator dynamics from evolutionary game theory: firms that redistribute less retain more, invest more, grow faster, and gain market share. The aggregate \(\theta\) is pulled down over time. Formally, the change in aggregate redistribution is proportional to minus variance times growth advantage. With heterogeneity in \(\theta\), drift is downward.
No firm needs conspiratorial intent. Competition itself rewards strategies that erode redistribution.
The structural tension is concise: stability requires sufficiently high \(\theta\), while competition drives \(\theta\) downward. Those two conditions conflict by design.
Do Alternative Channels Work?
The paper evaluates four common alternatives:
Broad asset ownership: If households own enough equity, dividends could replace wage income. But households starting with low assets do not benefit proportionally. Ownership compounds along the same competitive gradient.
Credit expansion: Borrowing sustains consumption temporarily, but does not solve the income formation problem. Debt-to-income eventually becomes unstable, consistent with Minsky and Mian-Straub-Sufi dynamics.
New sector creation: AI may create new industries, but sectors only scale if demand exists. If median income stagnates, new sectors face demand limits too.
Human-capital adaptation: If technological thresholds keep rising, the share of workers who can adapt may shrink. A system dependent on exceptional adaptation is not structurally robust.
None of these options interrupts Darwinian drift without a coordination mechanism backed by an enforcement mechanism.
Conclusion of Paper 2
Only income regimes with enforceable coordination can sustain both a sufficient redistribution level and persistence over time. Purely decentralized systems are dynamically unstable in the long run.
This does not imply the end of markets. It implies that markets alone can no longer preserve the stability conditions they historically relied on.
Part III: Why Capital Volume II Returns
A Neglected Work
Capital, Volume II has had an unusual intellectual fate. Even many Marxists read it less frequently because it lacks moral drama; mainstream economists often avoid it because it bears Marx's name.
Yet its core ideas migrated into mainstream analysis indirectly. Reproduction schemas anticipate input-output logic later formalized by Leontief. Kalecki's effective-demand formulation—"capitalists earn what they spend, workers spend what they earn"—is essentially a modern-language restatement of Volume II concerns.
The Core Question
Volume I asks how surplus value is produced. Volume II asks whether the system can continue circulating and reproducing itself once that surplus exists. Marx's Department I/II structure highlights proportional consistency problems that capitalism does not automatically resolve.
As long as wage-based reproduction worked, this looked abstract. Under AI-led wage erosion, it becomes immediate.
Part IV: Academic Landscape
These papers sit at the intersection of several literatures.
Acemoglu & Restrepo show automation systematically lowers labor share and can reduce labor demand even when productivity rises. Their Econometrica work (2022) attributes roughly 50–70% of major U.S. wage-structure shifts to task displacement.
Summers re-centered secular stagnation. Mian, Straub, and Sufi formalized demand weakness under inequality and debt dynamics that standard policy tools struggle to resolve.
Empirical evidence is still mixed. The Yale Budget Lab has not yet found a clear macro-scale disruption, while Brynjolfsson reports about a 16% employment decline in AI-exposed entry-level jobs, and Goldman Sachs points to a continued decline in tech employment share. The signal is emerging, but not yet uniformly visible in aggregate data.
Part V: Open Questions
Can "Voluntary Markets" Solve It?
At first glance ESG appears to contradict the model. But it may instead confirm it: in 2019–2021, ESG expansion looked like a temporary rise in effective \(\theta\) through voluntary coordination; in 2023–2025, ESG backlash reversed much of that momentum. Under renewed competitive pressure, redistribution retreats—exactly what Darwinian drift predicts.
The Political Economy of Redistribution
If AI concentrates profits in a narrow corporate core, the taxable base overlaps with actors that have both the strongest lobbying power and the strongest incentive to resist redistribution. Historically, welfare-state settlements were often stabilized by the fear of social rupture and by labor's capacity to withhold. In an AI-intensive regime, displaced workers may lose that leverage, and enforcement politics changes fundamentally.
So the question is not only "Can the state pay?" but "Who can compel the state to pay?"
Where Is the Boundary Between Stabilization and Reproduction?
Capitalism has never operated without state institutions (property rights, currency, education). If transfers such as UBI become deeply institutionalized, are they still "external stabilization," or do they become a new internal reproduction mechanism?
Part VI: The Highest Irony of Capitalism
A Circuit That No Longer Closes
Putting both papers together: AI weakens wage-based reproduction; voluntary redistribution cannot stabilize endogenously; the state becomes a permanent coordinator. The core circuit shifts from production → wages → consumption to production → profits → taxation → transfers → consumption. At that point, the state increasingly determines who gets what.
At that point, purchasing-power allocation becomes increasingly rule/state-mediated rather than market-mediated. If transfers must scale with productivity, the state-allocated share of GDP rises over time. The system drifts from a market economy supplemented by the state toward a state-mediated economy in which markets function increasingly as production engines.
Capitalism Transcending Itself
AI, the apex of capitalist innovation, may force capitalism beyond its wage-anchored form—not through revolution, but through structural necessity. If production no longer requires labor at scale, and mass consumers no longer emerge endogenously, the system ceases to function as capitalism in its historical wage-anchored sense.
In Marxian terms, this is a contradiction between productive forces and existing relations of production. In this framework, the mechanism appears not through manifesto, but through replicator mathematics.
Final Note: The Question Still Open
Both papers are diagnostic frameworks. They aim to ask the right question and answer it systematically at the level of mechanism. But the largest question remains unresolved.
If the state becomes a permanent allocator of income, who governs that allocator? A state that determines economy-wide income shares holds immense power. Without strong accountability institutions, redistribution can become authoritarian.
Another open frontier is whether post-AI economies can invent entirely new income channels. Wage labor itself was once an institutional innovation. The next regime may involve channels we do not yet have stable names for.
So the core question is not whether AI will eliminate every job. It is whether labor can remain central enough to income reproduction to preserve the self-stabilizing properties of market economies. If not, the stabilizing node will move—and where it moves, who controls it, and how it is governed may be one of the defining political-economic questions of this century.