At Home He Feels Like A Tourist

Essays in Dignity and Political Economy

This essay was built in conversation with an AI — the same technology it is attempting to analyse. The reflexivity is intentional and not entirely comfortable. The tool that helped construct the argument is also the subject of the argument. What it means that a large language model can participate in the construction of a political economy critique of large language models is itself one of the questions this essay is circling. The irony is noted, the collaboration is disclosed, and the judgment — about what the argument means, whether it holds, what follows from it — remains human. Or at least: it remains the human’s problem. Which is, in miniature, exactly the point.

* * *

I. The Room Where It Happens

Sit in any room where the people with the money are talking about artificial intelligence and you will hear the same vocabulary deployed with the same confident imprecision. Transformation. Disruption. Productivity unlock. Agentic workflow. The language has the quality of a second language learned from a phrase book: grammatically plausible, semantically weightless. Nobody is lying, exactly. They just haven’t needed to think very hard about what they mean, because the money has been moving in one direction and the direction has, so far, felt good.

What they are not doing — what the structure of their professional formation makes almost impossible — is asking the question one level up. Not how do we deploy this but what does it mean that we are deploying this. Not what is our AI strategy but what kind of economy are we building, and how will it function when we have finished building it. These are not mystical questions. They are the questions that a functioning political economy requires someone to ask. At the moment, almost nobody in the rooms where the decisions are being made is asking them.

This essay is an attempt to ask them. It will not and cannot resolve them. The sincere position — which we will return to — is that some of the most important questions here sit at the intersection of economics, philosophy of mind, and political theory in ways that admit of genuine uncertainty rather than tidy conclusions. But uncertainty is not the same as ignorance, and the fact that the people currently allocating the capital are not asking the questions does not mean the questions are not urgent. It means the urgency is not yet being priced.

First an anecdote. A broker analyst, those slightly geeky, insistent types wheeled out on Bloomberg TV to explain a company’s financial results, once presented a discounted cashflow model for a mid-sized UK technology services company — an industry that employed, at the time, perhaps a hundred thousand people across its various operations — with a terminal growth rate of seven percent. For the non-specialists: the terminal growth rate is the rate at which you assume an industry will keep growing, compounding, indefinitely into the future. Someone in the room did the arithmetic on a calculator — this was before the smartphone made conspicuous arithmetic rude — and noted, with the careful politeness of someone delivering bad news to a true believer, that at seven percent compound growth from that base, by 2050 the implied workforce of the UK technology services sector alone would be somewhere in the region of thirty million people. At that point, comparative advantage would require the entire working population to be employed in selling software services. Presumably to each other. Ricardo’s law hyperscaled.

The model was technically coherent. Its relationship to reality was not. The story is useful not because it is unusual but because it is representative. The financial models currently being built around artificial intelligence contain assumptions about revenue trajectories, addressable markets, and perpetual growth that have the same quality: internally consistent, formally impressive, and dependent on a vision of the future that dissolves on contact with basic arithmetic. OpenAI, valued at $852 billion as of March 2026 and projecting $100 billion in annual revenues by 2029, is the terminal growth rate problem stated at civilisational scale. The number may be right. The question nobody in the room is asking is: right relative to what, and who is paying, and with what, and having earned it how.

These are political economy questions. They used to have a literature.

* * *

II. The Circuit, Stated Plainly

Economies are not collections of objects. They are systems of relationships, and the relationships are what matter. This is not a heterodox position — it is, at some level, what any serious economist will tell you if you press them past the introductory textbook. But the relational character of economic life is easy to forget when the relationships are functioning smoothly, and politically convenient to forget when naming them would implicate those with power in the consequences of disrupting them.

The basic circuit of a consumer capitalist economy runs like this. Production generates income — wages, salaries, profits, rents. Income generates expenditure — on goods, services, housing, leisure. Expenditure generates revenue for producers. Revenue funds further production. The circuit is closed. Break it at any point and you break it everywhere, because every act of production is simultaneously an act of income distribution, and every act of spending is simultaneously someone else’s revenue. This identity — that production, income, and expenditure are three descriptions of the same set of relationships — is the key to understanding why the demand deficit problem is not a footnote to the AI story. It is the story.

In advanced economies the expenditure numbers have a specific shape that matters enormously. Private consumption — households buying things, experiences, services — accounts for roughly sixty to sixty-five percent of GDP. In the United Kingdom the figure sits around sixty-two percent. In the United States, closer to sixty-eight. Government expenditure runs at twenty to twenty-five percent. Business investment is fifteen to twenty percent. The consumption share is the critical number. Nearly two thirds of everything an advanced economy produces depends, directly or indirectly, on people having money to spend.

In the postwar settlement — roughly 1945 to 1975, the period that Wolfgang Streeck describes as the class compromise embedded in Fordism — consumption was funded primarily by wage income, underpinned by collective bargaining, rising productivity shared between capital and labour, full employment as an explicit policy objective, and the social wage of public services that reduced the proportion of income households needed to spend on health, housing, and education. The circuit was self-reinforcing. Henry Ford‘s insight — that his workers needed to earn enough to buy his cars — was the microeconomic expression of a macroeconomic truth. Wages and consumption rose together, and the rising tide, for a specific historical period under specific political conditions, did raise most boats.

From roughly 1980 onwards that settlement broke. The labour share of national income — the proportion of GDP flowing to workers rather than capital owners — began a long decline across all advanced economies. Real wage growth decoupled from productivity growth. The income foundation of consumption eroded.

But consumption did not collapse. Something filled the gap. That something was credit.

The great consumer credit expansion — mortgages, credit cards, car finance, student loans, buy-now-pay-later in its successive iterations — became the mechanism by which consumption was sustained even as its wage foundation was hollowed out. Debt replaced income as the fuel of demand. This is Streeck’s central argument in Buying Time: that the neoliberal settlement survived as long as it did not because it solved the income problem but because it deferred it, substituting private debt for the public and collective mechanisms of income distribution that the postwar settlement had constructed. The 2008 financial crisis was the moment when that deferral mechanism reached a structural limit — when the debt loads became unsustainable and the circuit finally broke in ways that could not be hidden.

The circuit was patched with quantitative easing, near-zero interest rates, and a decade of financial repression that transferred wealth from savers to asset holders on an extraordinary scale. The patch held, more or less, until 2022. Now interest rates are higher, household debt-to-income ratios are at or near historic peaks in most advanced economies, and the capacity to substitute credit for wages has real and approaching limits. The tank is not empty. But the gauge is no longer comfortably in the full zone, and someone is about to ask the car to drive much further on what remains.

Into this specific historical moment — a consumption economy running on credit rather than wages, with the credit mechanism under pressure — arrives the largest single wave of capital investment in the history of technology, dedicated explicitly to automating the cognitive labour that constitutes the employment base of the service economy that replaced the manufacturing economy that replaced the agricultural economy. The timing is, to use a technical term, a bit shit.

The infrastructure and who really pays for it

The scale of the current buildout requires a moment’s pause. The five largest US hyperscalers — Microsoft, Alphabet, Amazon, Meta, and Oracle — have committed to spending between $660 billion and $725 billion on capital expenditure in 2026 alone, nearly doubling 2025 levels. Total hyperscaler capex nearly tripled from $162 billion in 2022 to $448 billion in 2025. Big Tech AI capex is projected to exceed $1 trillion annually by 2027. These are not ordinary investment numbers. They are the construction of a new physical substrate for the global economy.

The financial architecture of this buildout deserves more scrutiny than it receives in the rooms where the vocabulary is deployed. The scale of capital commitment has long since exceeded what equity capital can absorb at the required speed. The gap is being filled with debt. Corporate bond issuance by the major technology companies has been running at record levels. The private credit market — the unregulated, non-bank lending ecosystem that expanded dramatically after 2008, when regulation pushed risk from visible balance sheets into the shadows — is deeply embedded in the data centre financing chain, through infrastructure funds, private REITs, and credit vehicles whose aggregate exposure to AI infrastructure nobody has yet mapped with any confidence. This is not equity risk, where investors lose their stake and the loss stops there. This is debt risk, where the failure to generate the cash flows required to service the obligations spreads through the financial system in ways that are, by design, opaque. The parallel with the pre-2008 mortgage debt ecosystem is not exact. It is close enough to be worth noting.

The capital is moving at this speed for reasons that are not purely rational. The investment thesis for AI has acquired a quality that serious students of financial history will recognise — the collective arousal of a mania, the almost erotic intensity of a market convinced it is present at the creation of something epochal. This is not a dismissal. The technology is real. The transformation may be as significant as its proponents claim. But the South Sea Bubble was also responding to a genuine phenomenon — the expansion of global trade, the novelty of joint-stock finance — and the mania consumed the reality in ways that took a generation to resolve. The fear of missing the defining technological transition of the era is a more powerful allocator of capital than any discounted cashflow model, and the people who understand this are using it deliberately. The language creates the urgency. The urgency creates the flows. The flows create the valuations. The valuations validate the language. It is a loop with the structure of a collective hallucination that is also, partially, justified — which is precisely what makes it so difficult to interrupt from within.

The current phase of investment is also almost entirely business-to-business. The hyperscalers are selling compute to enterprises. Enterprises are using it to automate internal processes. The consumer is not yet in this loop except as the eventual customer whose demand is assumed rather than created. Capital is consuming capital. The circuit is running in a closed loop that does not replenish the consumption base it depends on.

The B2B loop has a further structural feature that the infrastructure conversation tends to obscure. The data centres require power — extraordinary amounts of it, growing at rates that have sent the hyperscalers into forward power purchase agreements, behind-the-meter generation projects, and increasingly urgent conversations about nuclear capacity. In the United Kingdom context, the request — would you please accept new nuclear infrastructure so that our data centres can function? — would unite the electorate in opposition in ways that almost nothing else currently manages. The planning consent alone would take a decade and a half. The answer the technology sector is moving toward is privatised generation — their own power, their own grid, their own solution to the constraint that their demand has created. Which is, structurally, the same move as the rest of the investment thesis: use the public infrastructure as the baseline, capture the returns privately, socialise the costs of the disruption your withdrawal creates.

The American version of this has a detail that belongs in the account of the constraints. The most capitalised infrastructure buildout in the history of technology is running into a shortage of electricians. Not a shortage of capital. Not a shortage of chips, or land, or planning permission, or political will. Electricians. The skilled manual workers who wire the data centres, run the cables, connect the transformers — workers whose training pipeline was not automated, whose skills cannot be generated by a language model, whose formation requires exactly the kind of on-the-job apprenticeship that the investment thesis is simultaneously making economically obsolete for everyone else. The irony has the quality of a system telling you something about itself.

* * *

III. Work Within Work

To understand what artificial intelligence actually does — not what the press releases say it does, not what the anxiety says it does, but what it materially does to the structure of work — it helps to think about work differently from how we usually discuss it.

Every piece of work, every professional function, every job that involves thought rather than just physical action, has a layered structure. The layers are not always visible from the outside, and people doing the work are not always conscious of them, because expertise is partly the process of making the layers invisible to yourself — performing the upper layers so fluently that the lower ones disappear into habit. But they are there, and the layers matter enormously for what can be automated, what cannot, and — most importantly — what happens to human capacity when the lower layers are removed.

The base layer is process. The sequence of steps that converts a defined input into a defined output. Document review against a checklist. Standard contract drafting from a template. Financial data extraction and formatting. Code generation from a specification. Customer query routing. These tasks have always been automatable in principle. What large language models have done is extend the reach of process automation far up the cognitive stack — into tasks that looked, from the outside, like they required judgment, because they involved language and reasoning rather than physical manipulation. It turns out that a substantial proportion of what passed for judgment in professional work was sophisticated pattern completion. And pattern completion, at extraordinary speed and across an enormous range of domains simultaneously, is precisely what a large language model does.

Above process is what we might call model construction: the capacity to build an internal representation of a system — a business, a legal situation, a codebase, a patient — that is generative rather than merely descriptive. Not what does this look like but how does this work, what are its failure modes, what is the key variable that everyone else is underweighting, what would have to change for the whole picture to look different. Large language models can do impressive things in this space. The outputs are often plausible, well-structured, and genuinely useful as a starting point. But there are specific leaps that remain unreliable in ways that matter. The capacity to identify when the model itself is wrong in principle rather than in detail — to notice that the terminal growth rate produces an absurdity, that the legal argument has a structural flaw rather than a factual error, that the diagnosis is formally consistent but does not feel right in ways that point to a missing variable — requires something that is not yet reliably present. Call it structural imagination: the ability to inhabit the counterfactual world the model implies and notice that it does not make sense.

Above model construction is what people in the technology industry are beginning to call, with some justice, the philosopher layer: the capacity to determine which model is worth building, what the relevant frame is, what you are actually trying to understand and why the question you have been asked may not be the question that needs answering. The senior consultant who tells a client that their problem is not the one they presented. The researcher who sees that the field has been asking the wrong question for a decade. The observation — made with increasing frequency by people who build AI systems — is that this is the layer where human value is migrating. The skill premium shifts upward. The philosopher, not the coder.

This is true, and important, and also insufficient as a response to what is happening. Because it omits something critical about how the philosopher layer gets produced in the first place.

Professional formation — the process by which a person becomes someone who can exercise high-level judgment in a complex domain — does not work the way a training programme works. It is not a matter of acquiring a body of knowledge and a set of techniques. It is a developmental process, and it requires the full ladder. The junior analyst who is told their model is wrong, and has to work out why, and is wrong again, and gradually develops an intuition for the difference between formal validity and substantive truth — that person is not acquiring a skill. They are being formed. The formation is inseparable from the experience of being wrong about things that matter, of having a stake in getting it right, of accumulating what you might call the scar tissue of judgment.

Remove the process layer — automate the junior work, thin the graduate intake, eliminate the entry point — and you do not get a workforce that skips straight to the philosopher layer. You get a skills cliff in fifteen to twenty years when the senior practitioners who were formed the old way retire and there is nobody behind them who went through the formation process. The pipeline does not just thin. It breaks.

This is the point that the just learn new skills response to AI displacement misses most completely. It is not wrong that new skills will be valuable. It is wrong about the nature of skill acquisition in complex domains. You cannot train your way from a university degree to philosopher-developer without the intervening decades of formation. The apprenticeship is not inefficiency in the system. It is the system. And the system is currently being dismantled at the base by people who are focused on the productivity gains at the top and have not done the arithmetic on what the base produces.

There is a lineage here that the technology industry would benefit from knowing about. Thomas Carlyle, in Past and Present in 1843, articulated what he called the gospel of work — the argument that labour is not merely instrumental to income but constitutive of human dignity, that the person formed through useful work is a different kind of person from one who has not been. William Morris extended the argument into a critique of industrial production that separated the worker from the product of their work. The labour movement institutionalised it as the demand not just for wages but for conditions, for craft, for the right to take pride. George Orwell cut through to the essential: that useful work is one of the things that makes a human life worth living.

Bernard Stiegler — the French philosopher whose Technics and Time trilogy is one of the most serious attempts to think about the relationship between technology and human capacity — gave this lineage its sharpest contemporary formulation. He called it proletarianisation, but he meant something more precise than Marx: not just the loss of ownership of the means of production but the loss of the know-how, the savoir-faire, that constitutes human cognitive agency. When a craft worker is replaced by a machine, something is lost beyond the income — a way of knowing, a form of attention, a relationship to the world. Stiegler’s argument is that digital technology represents the most radical proletarianisation yet, because it externalises not just physical skill but cognitive and even creative judgment into technical systems. The capacity does not just become unnecessary. Over time, and with sufficient dependence on the technical system, it atrophies.

The question this raises — and it is a genuine question, not a rhetorical one — is whether the pattern matching that large language models perform is categorically different from what happens in human cognition, or whether the difference is one of degree, implementation, and the presence or absence of a body with stakes in the outcome. The philosophy of mind does not have a settled answer. The predictive processing account of the brain — which has substantial empirical support — suggests that human cognition is, at sufficient resolution, a form of pattern completion operating on learned statistical regularities. If this is right, the difference between what a language model does and what a human analyst does is architectural and quantitative rather than categorical. If it is wrong — if there is something about situated, embodied, stake-holding cognition that generates genuine understanding rather than simulating it — then the categorical difference is real and the philosopher layer may be genuinely safe.

Civilisation-scale decisions about the structure of work and the distribution of its proceeds are currently being made on the basis of this unresolved uncertainty. This seems, at minimum, worth noting.

* * *

IV. The Demand Deficit, or: Whose Problem Is This

There is a moment, in the financial history of every major overinvestment cycle, when the people who built the thing and the people who funded the thing look at each other across a conference table and have a conversation that was always implicit in the numbers but somehow never had explicitly. The moment when the terminal growth rate meets the calculator.

The AI investment cycle has not reached that moment. It may not reach it in the form of a sudden crisis. The technology is real in a way that the pets.com business model was not, and the infrastructure being built will persist and compound whether or not the specific companies building it survive the financial cycle. But there is a prior question that the rooms full of confident vocabulary have not yet asked: even if everything works — even if the revenue projections are correct, the technology does what its proponents claim, the infrastructure gets built and powered and staffed — what happens to the economy that is supposed to be on the receiving end?

The circuit described in section two has a specific vulnerability that the investment thesis has not priced. Consumer capitalism requires consumers. Consumers require income. Income, in the present structure of advanced economies, comes overwhelmingly from labour — from wages and salaries, and the credit which is leveraged on this, that constitute the primary income source for roughly eighty percent of households. The capital share of income — profits, dividends, rents — flows to a much smaller number of people, and flows in a pattern that is heavily weighted toward assets rather than consumption. The rich save more. The mechanism by which productivity gains become demand is the wage.

What the AI investment thesis is building, at scale and with deliberate intent, is a set of tools explicitly designed to substitute capital for labour across the cognitive economy. Not to augment labour in ways that raise productivity and wages together — the Fordist settlement — but to replace it. The business case for enterprise AI is not your workers will be more productive and you will share the gains with them. The business case is you will need fewer workers. The efficiency gains flow to margins. The margins flow to debt providers and shareholders. The shareholders are not, in the main, the workers whose labour income constituted the consumption that justified the investment in the first place.

The human consequences, already arriving

The human consequences of the demand deficit are not evenly distributed, and they are not arriving in the future. They are arriving now, in the specific and measurable experience of the generation that was educated for the cognitive economy and arrived at its entry point as the entry point was being removed.

The data is unambiguous and worth stating plainly. Underemployment among recent graduates in the United States hit 42.5 percent in the fourth quarter of 2025 — its highest level since 2020. Employment among workers aged twenty-two to twenty-five in the most AI-exposed occupations fell thirteen percent between 2022 and 2025, driven not by redundancies but by the collapse of new hiring. Graduate unemployment is running above the rate for all workers — an unprecedented inversion. The entry-level training contract, the graduate scheme, the junior associate intake — the institutional forms through which the formation process operated — are thinning in exactly the sectors where AI capability is most advanced.

This is not the unemployment problem in the old sense. The old unemployment problem was cyclical: a recession contracts hiring, a recovery expands it, and the workers who lost jobs in the contraction find new ones in the recovery. What is happening to this generation has a different character. The jobs that are not being offered are not being held in reserve for an economic upturn. They are being redesigned out of existence. The formation pipeline is being broken at its base, and the people bearing the cost of that breakage are the ones who did everything that was asked of them and arrived to find the contract had been changed while they were in the library.

The mental health consequences are not a soft addendum to the economic argument. They are the mechanism by which the economic disruption becomes a social crisis. Youth unemployment and underemployment have well-documented relationships with depression, anxiety, the loss of purpose and structure that work provides, withdrawal from civic participation, delayed family formation. What is distinctive about this wave is that it is hitting the credentialled — the ones who were told that education was the hedge, that intelligence was the protection. The psychological violence of that particular betrayal is specific and severe. It is not you failed to prepare. It is you prepared for something we removed.

This is David Harvey‘s alienation argument applied to a situation Marx did not anticipate: alienation not from the product of labour, not even from the labour process, but from the formation process itself. The self that was supposed to emerge from the years of being wrong and learning and accumulating the scar tissue of judgment has not had the chance to form. The social architecture of work — the colleagues, the shared project, the community of practice, the daily structure that gives the week its shape — has been withdrawn before it could be inhabited. Alienation from work, from self, from others. Not as metaphor. As the lived experience of a specific generation at a specific historical moment.

Consider the geometry of this across the income distribution. At the top, the banker who sings in the bath and rings his mother. He is, for now, on the right side of the asset compounder divide. His job involves the judgment layer — the relationships, the narrative, the conviction that no language model has yet reliably replicated. His children are clever, he believes, and cleverness has always been the hedge. He can probably pay enough to save them, or to buy them time. He has not yet asked himself whether his children might be the wrong sort of smart — optimised for exactly the process and model layers that are being automated — or whether the rug might be pulled not gradually but suddenly.

A floor below him, in the middle office of the same institution, is a colleague whose name he does not know. She has spent fifteen years building expertise in exactly the kind of analytical and compliance function that enterprise AI is targeting first. She is not junior enough to have been excluded from the entry point. She is not senior enough to be protected by the relationship layer. She is in the precise middle of the capability distribution that the current deployment curve is approaching. She is doing the arithmetic about what comes next, and the arithmetic is not reassuring.

The distance between these two people — measured in floors, in salary bands, in the thickness of the financial cushion available to absorb disruption — is also the distance between someone else’s problem and my problem. The technology is crossing that distance faster than the asset compounder class has yet registered. Even where AI does not come directly for your job, it comes for the job of someone you love.

Kalecki and the discipline of replacement

Michal Kalecki — the Polish economist who derived Keynesian conclusions independently, from a Marxist starting point, in the 1940s — identified something that explains why none of this is generating the political response it should. In his 1943 essay on the political aspects of full employment he argued that capitalists have an ideological interest in unemployment that goes beyond the wage cost argument. Full employment threatens their power. Workers with no fear of unemployment are harder to manage, more likely to organise, more likely to demand a share of the productivity gains. A degree of unemployment — not too much, enough — is, from the perspective of capital, a useful disciplinary mechanism.

Kalecki did not anticipate AI. But the structure of his argument applies with uncomfortable precision. A technology that can credibly replace cognitive labour — even partially, even imperfectly, even with significant human oversight still required — shifts the threat calculation in every professional employment relationship. The implied message to every worker whose function overlaps with AI capability is not you will be replaced tomorrow. It is you could be replaced, and you should factor that into how you negotiate, how you complain, how much you ask for. The disciplinary function of technological unemployment does not require the unemployment to materialise. It requires only the credible threat of it.

This is where the someone else’s problem defence reaches its structural limit. The demand deficit is not a problem that the “market” will solve, because the market has no mechanism for internalising the cost of destroying its own consumption base. It is not a problem that individual firms will solve, because the firm that unilaterally maintains employment levels while its competitors automate will be outcompeted. It is not a problem that the technology sector will solve through philanthropic redistribution, because the scale of redistribution required to replace the wage income being automated away is beyond the capacity or the incentive structure of private charity. It is, in the most precise sense, a political problem.

And here is the contradiction that the investment thesis cannot resolve from within its own logic. The state that might bridge the demand gap — through public investment, through income support, through the commons equity mechanisms we will come to — is fiscally dependent on the tax revenues generated by the labour income that the investment thesis is eliminating. Furthermore, the state’s debt capacity relies on the growth generated from the 65% generated by consumption. Automate the workers. Reduce the wage bill. Improve the margins. Watch the income tax receipts fall, the national insurance contributions fall, the VAT receipts fall, bond yields rise, as consumption contracts. Watch the fiscal space for the public intervention that might bridge the gap contract simultaneously with the gap it needs to bridge. The doom loop is not a worst-case scenario. It is the arithmetic of the circuit, followed without stopping.

This is what a structural contradiction looks like from the inside. Not a crisis that arrives suddenly and can be resolved. A set of relationships, each internally rational, that together produce a result that nobody chose and nobody can individually exit.

* * *

V. The Intelligence Commons, or: You Did Not Build That Alone

The political economy of the demand deficit has a specific property that distinguishes it from most of the problems that democratic governments are equipped to address. It is not a market failure in the conventional sense — a negative externality, an information asymmetry, a monopoly distortion — that can be corrected at the margin by regulation or taxation without disturbing the underlying structure. It is a structural contradiction in the circuit itself. The mechanism that generates the productivity gains is the same mechanism that destroys the consumption base those gains require. You cannot fix this with a windfall tax and a retraining programme. The arithmetic does not work at that scale.

The responses currently on offer from the mainstream of policy debate are, without exception, inadequate to the problem they are nominally addressing. Universal Basic Income runs into the fiscal arithmetic immediately: the tax revenues required to fund it at meaningful scale are generated by the wage income and corporate profits that the automation is reducing. You are trying to fill the bucket with the water that is draining out of it. Retraining programmes address a skills mismatch that is not the primary problem. The problem is not that workers lack the skills for the available jobs. The problem is the structural reduction in the number of jobs that require human cognitive labour at all. Retraining people for the philosopher layer when the philosopher layer has a limited and highly competitive intake is not a solution. It is a description of musical chairs with extra steps.

None of this is an argument against these measures. It is an argument that they are necessary but not sufficient. The sufficient response has to address the ownership structure directly. It has to be, in Mariana Mazzucato‘s framing, about who owns the returns from public risk rather than just who bears the cost of private failure.

Mazzucato’s documentation of the public foundations of private technology wealth is the empirical grounding for everything that follows. The internet: DARPA. The touchscreen: CIA and NSF funded research. GPS: Department of Defense. The algorithm underlying Google’s search: NSF grant. Siri: DARPA again. The foundational investments that created the conditions for the current AI buildout — the decades of publicly funded research in mathematics, computer science, and cognitive science, the public universities that trained the researchers, the legal and institutional infrastructure that made the private investment possible — were made by the state, on behalf of the public, with public money. The returns flowed, through a series of political choices that were made so consistently and over such a long period that they came to look like natural law, entirely to private capital. This is not natural law. It is a political choice. And political choices can be unmade.

The value of every major AI system rests on foundations that are irreducibly collective. The training data — the accumulated written, creative, and intellectual output of human civilisation across centuries — was not produced by the companies that used it. It was produced by everyone who ever wrote a sentence, argued a case, told a story, worked through a problem in public. The network effects are produced by users whose contribution is uncompensated. The public research infrastructure, the legal framework, the educated workforce, the stable institutions that make large-scale private investment possible — these are collective goods, produced and maintained at collective expense.

The private ownership of the returns from these collective inputs is not a principle. It is a power relationship that has been institutionalised as a principle. The tech founder made very real contributions — of vision, of risk, of execution, of specific knowledge — that deserve recognition and reward. They did not make all the contributions. The citizen whose data trained the model, whose consuming behaviour constitutes the market, whose taxes funded the foundational research — she gets nothing. This is not the natural order of things. It is the current order of things. The distinction matters.

The proposal: correct initial distribution

Every technology and financial institution above a defined scale issues equity annually — we say one percent — to a publicly governed commons fund, in proportion to the value created by collective inputs. Not a tax on profits after the fact. Not a voluntary sharing scheme. A mandatory ongoing equity stake at the point of value creation, flowing to a fund that represents everyone whose contribution has so far gone unrecognised.

This is not redistribution. It is correct initial distribution. The employee who writes the code gets equity because her contribution is recognised in the ownership structure. The argument for the commons fund is identical in structure: the contribution is real, it has not been recognised, the ownership structure should reflect that truth from the beginning rather than pretending the value was created without it.

The arithmetic, run against current hyperscaler valuations of approximately $10 trillion combined, generates roughly $100 billion flowing to the commons fund in year one, compounding as the valuations compound. Within a decade, at reasonable growth assumptions, the fund accumulates a genuine public ownership stake in the most valuable assets of the era. The returns fund the public goods — education, healthcare, social infrastructure, the care economy — that the same technology is simultaneously threatening to defund by displacing the labour income that currently pays for them. The hole and the plug are the same size. This is not a coincidence. It is the structure of the proposal.

The Norwegian Government Pension Fund — the largest sovereign wealth fund in the world, currently managing assets of approximately $1.7 trillion — is the existence proof that this is not utopian. Norway took public ownership of a natural resource, North Sea oil. that could have been entirely privately captured, invested the returns patiently and at scale, and now holds a genuine public ownership stake in the most productive assets of the global economy on behalf of every Norwegian citizen. The intelligence commons is the oil field of the twenty-first century. The question of whether it gets a Norwegian settlement or a Nigerian one — whether the resource is governed for collective benefit or captured by the few with access to it at the moment of extraction — is a political question, not a technical one.

Governance: sortition and the unbiddable mechanism

The governance of the fund answers the capture problem before it is made. The fund is governed by sortition — random selection from the full eligible population, which is practically everyone. And everyone gets a dividend once the cash starts to flow. Unconditionally. Musk gets his dividend. The uncontacted tribesperson gets theirs, by proxy through whatever representative mechanism their situation requires. No means test, no deserving poor, no exceptions. The universality is not political strategy. It is the logical expression of the founding claim: the value was created collectively, the stake belongs to everyone, the return flows to everyone. Your money, whether or not you know how any of this works. People generally do know, at the level that matters — the felt sense of extraction is widespread and real, and the dividend reframes it from grievance into ownership. Grievance organises against. Ownership invests in. The political psychology is completely different and significantly more stable.

The selection mechanism for governance is automated and cryptographically unbiddable. The panel cannot be lobbied before selection because it does not exist. It cannot be captured after selection because it serves a defined term and returns to the population. It is very well rewarded for its contribution. It is John Rawls made structural rather than hypothetical — the veil of ignorance is not a thought experiment but an institutional fact. The selected panel members are drawn from the full range of human circumstances. They govern remaining distribution rules, not investment decisions. In any event the fund is passive. It holds a defined equity stake without discretion. The constitutional questions — distribution formula, eligible uses, amendment process — can be in large part be codified in advance. The advisor whose business model requires active discretion has nothing to extract. You could, in principle, ask an AI to draft the constitution. The model proposes. The sortition panel disposes. The population ratifies. Each component doing what it is designed to do — which is also, as it happens, a description of how this essay was written.

The incumbents’ interest in the proposal is more aligned than it first appears. The hyperscalers can absorb one percent annual dilution. The startup trying to reach their scale cannot. The mandate entrenches while it extracts, which means the monopoly capital that the left reflexively treats as the enemy of the commons may, on reflection, prefer a managed equity contribution to the genuine competitive threat that a commons-funded alternative infrastructure would represent. The hyperscalers will continue to move downstream towards the consumer. The closer they get the more the rational self interest in bridging the demand deficit will be privileged. It is in essence an insurance premium costing no more than the current level of equity issuance to their own people. Monopoly capitalism, on this reading, can be weaponised for the commons rather than against it.

The fund also shares the downside. When the pack of cards comes down — and the scenario in which the private debt spreading through the unregulated shadow banking system finds its 2008 moment is not a fantasy — the commons fund is in the room as an owner, not a creditor. Not standing outside the crisis making claims against the estate. Sitting inside it, with a stake and therefore a voice, at the table where the decisions about what happens next get made. The shared risk is not a cost of the proposal. It is its deepest justification.

The jurisdictional problem — capital routing around national regulation — is addressed by constituting the fund at the level of the institutions rather than the jurisdictions. Market access conditionality is the enforcement lever, as it was for GDPR. You want access to the markets, the infrastructure, the regulatory permissions that operating in major economies requires. You contribute. The EU moves first. The template is adoptable. The incumbents, for reasons just described, may not resist as fiercely as expected.

The dividend lands in individual hands. People spend it on individual things — most on private consumption, which is precisely what the demand deficit argument says the circuit needs. From informations commons to helicopter. Milton Friedman would surely approve. The state taxes the dividend as income and directs the proceeds to collective provision. The dividend is yours. The tax on the dividend is the collective’s. The distinction is clean. Neither pretends to be the other.

There are other contradictions that the commons equity settlement does not resolve, and honesty requires naming them. The ecological existential sits alongside, actually above, the demand deficit as the other great unpriced contradiction of the current system. The cost of treating the natural world as an infinite externality is arriving faster than the models predicted, and the same political economy that produced the demand deficit produced the carbon economy. The commons fund does not solve this. What it does — and this is not a small thing — is buy the time and build the institutional capacity to address it collectively, through an institution that is structured to think at the timescale the problem requires.

The fund addresses the temporal structure that the quarterly earnings cycle makes impossible. It accumulates over decades. The generation that bears the cost of the transition receives the early dividends from a fund that is still small. The generation that benefits from the full accumulated stake is the one that comes after. This is not a flaw. It is the deepest virtue of the proposal — a structure that requires the present to act in the interest of the future, which is exactly the temporal architecture the ecological crisis also demands, and which no other available institutional mechanism currently provides.

The jurisdictional innovation is where the proposal becomes genuinely new rather than a variation on existing mechanisms. Capital is global. The technology companies operate across borders and treat jurisdictions as variables to be optimised. The commons fund, constituted at the level of the institutions themselves rather than the jurisdictions they happen to be incorporated in, is potentially the first institutional form that matches the scale of the problem. Not a global government — the democratic legitimacy problem there is genuine. But a global commons stake, governed by the people whose data and demand and institutional infrastructure created the value, across the borders that the companies themselves treat as irrelevant when it suits them. A new way of thinking together about the contradictions that will otherwise consume us separately.

They call it stakeholder capitalism. They invented the language and mostly failed to act on it, because shareholder primacy reasserted itself every time the rhetoric was tested against actual capital allocation decisions. The commons equity proposal takes the language seriously in a way that corporate social responsibility never did. You are not a stakeholder because a communications team says you are. You are a stakeholder because you own a stake. The difference is the difference between a press release and a constitutional fact.

And beyond the rational self-interest. Beyond the insurance premium and the shared risk and the jurisdictional innovation and even Rawls behind the veil of ignorance. There is the question of what we actually want.

We want to think together about the problems that will otherwise consume us all. We want the structure in which the connections — between people, between generations, between the human project and the natural world that hosts it — are recognised as the foundation of value rather than the friction on its extraction. We want the institutional form that makes us all stakeholders in the same enterprise, which means stakeholders in each other. Not equal. The proposal does not pretend to eliminate inequality. But connected. In the same boat. When it thrives, thriving together. When it fails, failing together and therefore with some possibility of collective response.

This is what the arithmetic requires. It is also what we want.

* * *

VI. Capital is Forcing Labour to Go on Strike

We return, at last, to the room where it happens.

The people with the money and the vocabulary are not, in the main, ideologues. This matters because the ideological framing — capital versus labour, the class analysis, the tradition of political economy that the banker class would rather not name — is accurate at the structural level and unhelpful at the level of individual motivation. The banker who sings in the bath and rings his mother is not plotting the immiseration of anyone. He is optimising within a system whose structural contradictions are Someone Else’s Problem, and whose consequences for the middle-office colleague whose name he does not know have not yet arrived at the floor where the decisions are made.

But here is what the circuit tells us.

Capital is forcing Labour to go on strike.

Not as metaphor. As structural description.

A strike is the withdrawal of labour from the production process — the refusal to provide the human input that makes the circuit function. What is happening now is the inverse: capital is withdrawing the wage relationship, automating the cognitive labour that constitutes employment for a growing proportion of the workforce, removing the income that constitutes the consumption that closes the circuit. Labour is not choosing to withhold. It is being withheld from. The agency is inverted. The economic consequence — production without consumption, supply without demand, a circuit that runs but does not close — is structurally identical to a general strike. The system that spent forty years telling Labour it had no power has built a machine that demonstrates, with mathematical precision, that Labour was the demand all along.

This is the pyrrhic victory that no terminal growth rate can resolve. You can model the revenue. You can project the productivity gains. You can build the infrastructure and train the models and sell the subscriptions and watch the margins improve. You can, in the limiting case, automate everything. And then you have won completely and lost everything, because the economy that was supposed to receive the productivity gains has no consumers, the state has no tax base, and the asset values are denominated in demand that no longer exists.

The Cassandra is deafening. The room is still discussing productivity unlocks.

There is a prior scenario, noted in passing and not to be dismissed, in which the demand deficit never materialises because the investment thesis collapses first — in which the private debt spreading through the unregulated shadow banking system finds its 2008 moment before the technology finds its 2029 revenue. The calculator is available to anyone who wants to use it. But the more structurally interesting scenario is the one where everything works and the circuit breaks anyway, and the conversation that the numbers have been trying to start for years finally has to be had.

We are making that argument to the rational self-interest of capital as much as to the moral imagination of anyone else. Not because we have abandoned the political economy analysis — the circuit, the contradiction, the Harvey seventeen, the Kalecki discipline, the Polanyi counter-movement that is already forming in ways that the rooms full of vocabulary have not yet registered. But because the argument from rational self-interest is, for a specific audience, the one that lands. Give away one percent a year to the commons as you give to yourselves. Not from generosity. From intelligence about what your victories are worth if the circuit that produces them breaks.

And here, finally, is the provocation that the political economy cannot quite reach on its own.

You can take your machines to Mars. The fantasy is coherent on its own terms — escape velocity, the backup drive, the multiplanetary hedge against the civilisational risks that the rest of us have to live with. Capital without labour. Intelligence without embodiment. Connection without dependence. The clean solution to the messy problem of other people.

But consider what you are taking with you. And what you are leaving behind.

The thing that makes human intelligence different from pattern completion — the thing that the philosophy of mind has not resolved and perhaps cannot, that the consciousness work this project has been building explores and will not rehearse here — is not processing speed or synthetic capability. It is the fact of connection. The fact of needing and being needed. The fact that meaning is made between people rather than inside them. That the self is not a substance but a web of relationships that requires other selves to exist at all. We have established elsewhere in this project that there may be no concrete things. Only the relationships between them.

Exile from that web is not liberation. It is the abolition of the thing that was supposed to be liberated. The victory that requires the elimination of everyone else’s stake in the outcome is not a human victory. It is the achievement of something that has optimised away its own humanity in the process of winning.

One percent a year. The commons fund. The shared stake, the shared risk, the new institutional form adequate to the scale of what is coming. The Norwegian settlement for the intelligence commons.

Not because justice demands it, though it does. Not because Rawls would choose it, though he would. Not because the political counter-movement is coming and it is better to be on the right side of it before it arrives, though it is and it is.

But because you are still, for now, human. And the humans you are considering leaving behind are not a problem to be solved or a demand base to be replaced or an externality to be managed. They are the condition of possibility for anything worth having.

The connections are not the constraint.

They are the point.

* * *

Thinkers drawn on in this essay

David Harvey — on capital’s internal contradictions, the spatial fix, and alienation. Wolfgang Streeck — on buying time and the deferral of the income problem through private debt. Mariana Mazzucato — on the entrepreneurial state and the public foundations of private technology wealth. Bernard Stiegler — on proletarianisation and the externalisation of cognitive know-how into technical systems. Michal Kalecki — on the political aspects of full employment and the disciplinary function of unemployment. Karl Polanyi — on the double movement and the social consequences of market expansion. Thomas Carlyle and William Morris — on the dignity of labour as constitutive of human identity. John Rawls — on justice behind the veil of ignorance.

This essay is part of the Pig Iron series: Essays in Dignity and Political Economy, published at athomehefeelslikeatourist.blog. It was built in conversation with Claude (Anthropic). The judgment is human.