The Successor System

Closing the regulatory route leaves the value-flow interventions

Categorical Recursion shut the regulatory door. Category-based instruments cannot hold the wage-demand circuit in place, because the deployed object moves category faster than a regulator can redraw the line.

A serious reader turns next to interventions that do not depend on categories. Public ownership of compute. Mandated redistribution of automation gains into the wage share. State deployment of models with allocation by need. Sovereign AI funds, compute dividends, nationalised frontier labs. These act on where value goes, not on what the object is called. They are the correct class of intervention.

They still do not save the wage-demand circuit. They save consumption, which is a different object. The circuit distributed agency, bargaining power, status, mobility and democratic leverage through productive necessity. Redistribution can replace the income. It cannot replace the necessity. That is P4: no absorption channel or successor reconstitutes mass productive necessity, and redistribution substitutes for consumption, not for the wage-demand circuit.

Two systems that both close the loop are still two systems

The wage-demand circuit is a specific feedback loop. Production pays wages. Wages buy output. Demand for output creates demand for labour. The population’s purchasing power comes out of its own productive activity, so mass economic agency sits inside the loop.

Consumption funded by redistribution runs a different loop. Firms earn revenue, the state claims a share, transfers go out and recipients spend. The loop still closes, but through the state. Purchasing power comes from political allocation rather than from work.

These are not two settings of one machine. They differ in stability, political economy, the distribution of agency and long-run trajectory. Treating them as one is the central error of every redistributionist account of the AI transition. The thesis does not say the second system is impossible. It says the second is not the first, and whether postwar capitalism survives is the question of whether the first survives. If the first three premises hold, it does not.

Redistribution that catches a labour market is not redistribution that replaces one

The standard objection is that postwar capitalism was already full of redistribution. Pensions, public health, unemployment insurance, tax credits, housing support and free schooling all moved purchasing power outside the wage relation. On this view AI-era redistribution is just more welfare state.

The objection misses the difference between supplement and replacement. After 1945, transfers sat on top of a labour market that stayed the main source of mass purchasing power. They steadied the circuit without constituting it. Most working-age adults drew most of their income, recognised role, status and political standing from production. The welfare state caught people who fell out of the circuit. It was never the circuit.

A post-AI regime does not steady a wage circuit that is still primary. It takes over from that circuit as the source of mass demand. That substitution is the discontinuity. The question is whether most people hold purchasing power because their labour is needed, or because the political system grants them a claim on automated output.

The threshold is not the presence of redistribution but its centrality. Once transfers stop being the net and become the floor, the system has changed, even if every institution keeps its name.

Each alternative works on its own terms

Public ownership of compute means the state owns or controls the data centres, chips, training runs, models and inference capacity. Private firms lease access and rents land on the public balance sheet. This escapes the categorical problem entirely, because the state no longer regulates the object. It owns it.

A mandated wage share means firms must hold labour compensation at a target fraction of value added. When AI removes workers, the firm either keeps them at sustaining wages or pays the gap into a pool. The right to deploy becomes conditional on continued contribution to mass income.

Public deployment with non-market allocation means the state runs models for public ends and distributes the output by need or political decision. Health, education, planning, research and administration run on publicly owned systems, and nobody buys the output through wages.

All three are real, all three sit outside the categorical paradigm, and each could in principle sustain mass consumption indefinitely. The thesis asks what each fails to keep. The three are the cleanest cases, not the edge of the design space, and the structural claim covers hybrids and variants too. Any intervention that acts on value flow rather than on productive necessity preserves consumption, not the circuit.

A mandated wage keeps the wage form and loses the wage mechanism

The wage-share mandate is the hardest case. A critic can say that if firms must keep paying labour, wages and payroll survive and the loop looks identical to the postwar one.

It only looks identical. The mandate keeps the form of the wage and discards its mechanism. A wage paid because the firm needs the worker is labour income. A wage paid by obligation after the worker has stopped being needed is a transfer routed through payroll. It may be good policy, even morally required. It is still the successor system wearing the wage as an administrative costume.

Asking whether the firm would pay without the rule is too blunt a test. Plenty of postwar wages were politically shaped: union rates, regulated professions, public-sector pay, minimum wages. The real test is whether the worker still performs a function for which human labour is materially required to produce the output. If so, the wage is real, whatever politics did to its level. If not, it is income maintenance through payroll.

That is the supplement-versus-replacement line seen from inside the firm. A worker in 1965 earned a politically mediated wage and was also needed for production. Remove the mediation and the firm still hires the worker at another price. Remove the worker and the output stops. The circuit rested on that necessity. Synthetic wages appear where it has gone. Both deliver income. Only one is the wage-demand circuit.

The two also fail differently. A real wage is defended by the firm’s need for labour. A synthetic wage is defended only by the political decision that mandates it, which is exposed to elections, lobbying, fiscal stress and competition from jurisdictions that decline to impose it. One is built into production. The other is contingent on politics.

In every version the population is provided for and the circuit is gone

Under public compute ownership, the state collects deployment rents and pays them out, through a sovereign fund, a citizen dividend, a basic income or public services. People eat, vote and spend. Income arrives by state allocation, not wage labour. The loop in which producing things paid for buying them has become a transfer system with the state at its centre.

Under a mandated wage share, firms pay the shortfall into a pool and people still receive income. That income is no longer generated by making the goods it buys. It comes from a regulatory transfer running alongside production. The decoupling of production from income is the policy’s aim, and it is also the end of the circuit.

Under public deployment, people receive services outside any wage market, the existing model for health and schooling scaled up to cover what wages used to cover. The mechanism is allocation, not exchange. Mass economic agency gives way to mass political claim.

None of these abandons the population. People are fed and the lights stay on. What persists is a different system that delivers consumption.

The successor lacks the three properties that made the postwar settlement work

Beyond consumption, the circuit had three properties. First, it distributed economic agency through the same channel as purchasing power. Earning meant taking part in the system that made what you bought, and that gave workers leverage. A strike mattered because withholding labour withheld output. Bargaining mattered because the firm needed the worker roughly as much as the worker needed the firm. The postwar political economy rested on that mutual dependence.

People living on state allocation keep political power. They can vote, protest, organise, riot and withdraw legitimacy. Those powers are real, but they are not labour leverage. Labour leverage works because production needs the worker. Recipient leverage works because rule needs consent, or at least quiet. A workers’ strike halts output. A recipients’ strike halts nothing. A riot imposes cost without restoring necessity. The successor citizen may hold political power, but not the same economic power.

Second, the wage was a price signal that sorted talent, effort and ambition. The signal was crude and often unjust, but the productive system generated it, not the polity. Under a redistributive successor it weakens. Some work is still paid, but pay tracks state preference, sectoral protection or status reward rather than productive need. Allocation passes from market to polity.

Third, the circuit could be repaired. Keynesian demand management, welfare states, unions, public investment and industrial policy all assumed labour stayed economically necessary, so displaced workers could in principle be reabsorbed. The postwar settlement was never self-correcting. It was correctable. A redistributive successor loses that assumption. Shocks travel through the state’s fiscal capacity and political will, not through a labour market able to reabsorb people at scale.

The OECD’s Employment Outlook 2026 finds that adjustment to AI as a local shock often runs through transitions into joblessness rather than moves across sectors, leaving displaced workers with lasting scars. That is a current-state finding, not a structural endpoint. It is not what a readily correctable labour market would show.

A successor can deliver consumption without any of these three properties, and they are what made postwar capitalism what it was.

The threshold is majority agency, not the last job

Residual human work will always exist: protected roles, artisanal work, trust-bearing work, care, politics, luxury human service and legally required human presence. None of it preserves a system built on mass productive participation.

Postwar capitalism ends when wage labour stops providing mass economic agency, meaning a majority of working-age adults cannot sell labour at socially sustaining wages without subsidy, protection, artificial scarcity, makework or political intervention. The test is mass agency collapse, not total labour extinction. It concerns productive participation, not consumption.

This closes the residual-work dodge. Critics answer the thesis by pointing at jobs that remain. Kings had servants, aristocracies had artisans, and feudal societies ran on labour. Residual work says nothing about the system’s basis. The question is whether wage labour remains the route to economic agency for the majority. Once it does not, the system has changed even if every remaining job survives.

The measured distance to that threshold is large. The St. Louis Fed and Harvard Real-Time Population Survey records generative AI assisting 1.7 percent of US work hours, up from 1.4 percent in Q4 2024, on self-reported data. The McKinsey Global Institute estimated in May 2026 that 58 percent of work hours across ten European countries are technically automatable today. Technical automatability is not displacement, and the two figures measure different things in different economies. Together they mark the gap between what is deployed and what is reachable. P4 is a claim about where the successor lands as that gap closes, not that it has closed.

Installing a successor requires the leverage the transition removes

Set aside what the successor lacks. Getting it built is its own problem. Every structural alternative requires the state to tax, expropriate or control the most valuable infrastructure of the period. That needs either popular pressure strong enough to beat the AI-owning interests, or elite agreement that expropriation is the lesser evil.

Popular pressure of that strength needs mass economic leverage, which comes from the wage-demand circuit. The conditions for building the successor are produced by the system it replaces. Destruction is automatic. Construction is political. The conditions for construction erode as destruction advances.

Elite agreement is available in theory, if AI-owning interests accept that their position depends on a consumption base and fund redistribution large enough to keep it. The postwar settlement is a precedent. Most of history is a precedent against. Structural conditions tilt towards the second, because AI-owning interests are not tied to national labour markets as postwar industrialists were. Their revenue is global, their compute is mobile within limits, and their political weight scales with their share of national output rather than with how many national workers they employ.

Jurisdictions are already bidding for that capital. OpenAI, Oracle and SoftBank announced five new Stargate sites in September 2025 within what they describe as a $500 billion, 10-gigawatt US build-out, and OpenAI has since broken ground in Michigan. The figures are vendor-reported. Localities competing to host compute is the terrain any future tax on it would have to hold.

Keep the diagnosis apart from the forecast. Even under ideal installation, structural alternatives preserve consumption, not the circuit. Under realistic installation they are likely to arrive late, partially and under pressure. The second is a forecast about politics. The first stands however that forecast resolves.

A cooperative destination is unlikely to stay where it lands

A stronger argument concerns the destination, and it closes the optimist’s remaining exit.

Assume the transition is complete and the optimist’s preferred alternative was installed without resistance: a basic income from AI-capital tax, citizen-owned compute, nationalised frontier AI, a sovereign dividend, a post-work society of meaning. The question is whether it holds. The forces that drove the transition keep operating afterwards, and they tend to dissolve any cooperative endpoint that depends on suppressing them.

A basic income funded by AI-capital tax needs sustained taxation across jurisdictions that compete for that capital. Capital moves and competition pushes rates down. With the wage circuit gone, so is the mass political base that once defended progressive tax. The payment becomes negotiable downwards with nothing to stop the slide. The Multiplayer Prisoner’s Dilemma now runs between states, on the funding mechanism itself.

Universal basic capital needs ownership to stay spread across generations. Power-law returns concentrate value inside the portfolio. Liquidity preference concentrates holdings across individuals. Capital outside the scheme concentrates abroad regardless. Within a generation ownership drifts back towards concentration before anyone has tried to reverse the policy.

Nationalised frontier AI needs the planning state to keep pace with private-AI states, which grow faster along the capability path the earlier essays document. The planning state can adopt private-sector structures and become the same configuration, accept lasting relative decline, or suppress competition by force. The first dissolves the arrangement, the second is politically unstable, and the third is hegemonic.

A post-work culture of meaning still needs a substrate to pay for the compute, and that substrate is one of the arrangements above. Status hierarchies re-form around access to AI capability and the substrate’s politics dominate.

Small communities can be stable inside a tolerant larger system, as monastic and Amish communities are, but only at small population fractions. That is a niche, not an absorption channel.

Three attractor states remain open, and the control problem decides between them

Where configurations are stable, they tend to be stable for a structural reason. The analysis points to three attractor states, each an open possibility and none a prediction.

The first is a Singleton: competition removed by a single hegemonic actor at global scale, whether a state, a firm or an aligned superintelligent governor. The second is Neo-Feudalism: competition left to produce the unequal equilibrium it selects for, with a compute-owning elite, a verifier class and a dependent population on minimal provision. The third is Fragmentation or Managed Decline: local alternatives that survive, but only as a small fraction inside a larger system tolerant enough to leave them be.

Stability and broad agency pull against each other because the dilemma that drove the transition keeps running after it. Suppressing it needs either a singleton or global coordination that genuinely cannot be broken, and neither sits in the toolkit available to actors starting from current conditions. Which attractor obtains, if any, is decided less by economics than by the control problem: who, or what, holds the capacity to enforce an outcome.

The optimist’s picture is also a static snapshot. Arrangements do not hold still. Once labour scarcity leaves the system, status competition re-anchors on whatever scarcity remains: elite verification roles, proximity to compute, attention from the AI-owning class, reproductive opportunity, selectively conferred meaning. Absent hegemonic enforcement, stable endpoints free of competition are hard to find. Stable endpoints in which competition has selected an arrangement the optimist would not have chosen are not.

So the honest question is which stable configuration is reached, and on whose terms. Broadly flourishing intermediate arrangements are possible, but they tend to dissolve unless something holds competition off them. That question is harder than the optimist’s, and it is the one that survives the endpoint stability test.

Stripped down, the thesis is a claim about circuit severance

Postwar capitalism rested on a wage-demand circuit in which mass productive participation funded mass consumption, which funded mass participation. AI severs the circuit by making mass productive participation economically unnecessary. Unit cost dominance produces the severance at the level of tasks. Interface collapse carries it into workflow recomposition. Coordination impossibility means no actor can halt it. The Sorites Collapse Principle and Categorical Recursion mean regulation cannot route around it. Structural alternatives can keep consumption going. They cannot rebuild the circuit.

The claim is not that all human cognition loses value. It is that general-purpose cognitive labour loses its role as the mass scarcity behind middle-class absorption. Some roles will stay protected and well paid. A circuit needs mass absorption, not islands of residual scarcity.

What follows is a successor system in which purchasing power comes from political allocation. It need not be worse on every measure. It could be fairer, more humane and better for the median citizen than late-stage rentier capitalism. It could also be more authoritarian, more fragile, more dependent on the goodwill of the AI-owning class and more open to capture. Its shape depends on political choices made under heavy constraint.

The thesis does not say which successor arrives. It says that, if the premises hold, some successor arrives, that no version of it preserves the wage-demand circuit, and that the survival question for postwar capitalism is therefore settled at the level of structure.

The policy question is constitutional design, not labour-market repair

Most AI policy debate assumes the task is to save the wage-demand circuit. On this analysis that option has gone. No mix of regulation, redistribution, ownership or coordination restores it, because the mechanism that generated it, general-purpose cognitive labour as the mass scarcity behind middle-class absorption, has been removed.

The real question is which successor arrives, who designs it, and on whose terms. That is a debate about the constitutional design of a post-wage economy, and it should be argued in that vocabulary rather than the vocabulary of labour-market policy.

Taken seriously, the structural alternatives are options for the successor, not rescues. Public compute ownership, a mandated wage share and public deployment are three constitutional choices. Each spreads power differently and carries its own stability properties, failure modes and relationship to democratic agency. Choosing between them is the decision actually in front of the polity. What a workable successor architecture would need, and how fast the cost of intelligence beneath it is falling, is the subject of the follow-on work in The Constraint Wearing Four Costumes and The Shortening Half-Life of Intelligence.

Pretending any of them preserves the old system stops the decision being made. The successor is then designed by default, by whoever has most leverage, at the point of deepest crisis. That is the worst design environment available. Refusing to see this does not defend postwar capitalism. It ensures the successor is designed without the people refusing.

What would refute this

P4 fails if an absorption channel rebuilds mass productive necessity at scale. The signal would be sustained and joint: the share of working-age adults whose primary income is wages holding or rising as AI deployment deepens, labour’s share of income recovering, real median wages tracking productivity, entry-level hiring in AI-exposed work recovering, and a rising share of consumption financed by wages rather than by transfers or debt. One series moving is noise. Several moving together for years against heavy deployment would refute the premise.

The entry-level signal needs care because the evidence currently conflicts. The Federal Reserve Bank of New York found in May 2026 no divergence between junior and senior postings within highly exposed occupations, and noted that the relative decline in exposed vacancies began before ChatGPT. The Stanford Digital Economy Lab’s payroll analysis, revised in August 2026, puts employment of workers aged 22 to 25 in AI-exposed occupations 19 percent below where it would be had it kept pace with less-exposed peers. The evidence register carries a reconciliation hypothesis, drawing on PwC’s 2026 Global AI Jobs Barometer finding that AI-exposed junior roles are seven times more likely to demand senior skills: postings hold level while junior roles are re-specified. The conflict remains unresolved.

The endpoint argument is refutable too. A broadly egalitarian arrangement that holds under inter-state competition for a generation, without a hegemonic enforcer and without drifting back to concentration, would show that the stable outcomes are not confined to the attractors named here. The full set of conditions sits in What Would Refute the Thesis.

The next essay takes the most sophisticated remaining objection, that friction buys enough time to keep the circuit alive in modified form. Continue to Essay 7: Drag Is Not Rescue.