The essay · the whole argument in one piece

Beyond Human Farm: An Essay

About 15 minutes to read

A society without the farm, designed in good faith and tested against everything history knows. The argument of the book in one essay.

Two experiments

On 25 February 1825, the cotton manufacturer Robert Owen stood in the Hall of the House of Representatives in Washington and told an audience that included the outgoing and incoming presidents of the United States that he had found a way to remake society. Competition and private property could be replaced by communities of cooperation, in which people would work together, share the product and raise their children to be rational and good. He called it a "new moral world", and he had already bought a whole town in Indiana in which to begin it. He renamed it New Harmony. Within two years, amid quarrels over property, too many idealists and too few skilled workers, it had collapsed.

Twenty years later, on the longest night of 1844, a group of weavers and other tradesmen in the mill town of Rochdale, in the north of England, opened a small shop by candlelight at 31 Toad Lane. The gas company had refused to supply them, doubting they could pay. Their stock was a little flour, butter, sugar and oatmeal. Their rules were simple: fair prices, one member one vote, the surplus returned to members in proportion to what they bought, a share set aside for education. The Rochdale Pioneers had read Owen. But they started with a shop, and it lasted. Their rules became the model for a cooperative movement that now has, by the International Cooperative Alliance's count, on the order of a billion members.

The community designed to change everything lasted two years. The shop that changed one thing has lasted nearly two centuries. Beyond Human Farm stands between those two outcomes. It sets out to design, in good faith, a society beyond the farm, and then to test that design as history tested New Harmony and Toad Lane: keeping what survives, letting go of what does not, and saying why.

Where the first book stopped

The first book, Human Farm, followed one pattern across five thousand years. A very small number of people, roughly the 0.01%, have owned the surplus that everyone else produced; a larger but still small group, roughly the 1%, have managed its collection on their behalf; and the 99% have produced it. The ratio between them shifted only when several of five forces lined up: catastrophe, which destroyed concentrated wealth; scarcity, when the many became scarce or indispensable; organisation that overcame the divisions among them; divided owners, whose quarrels opened cracks; and gains locked into lasting law and institutions. Beneath them ran the story that made the farm feel just, and every lasting breakout began by breaking it.

The first book ended without a programme, on a tablet from Uruk where a temple accountant named, or titled, Kushim recorded 29,086 measures of barley. The ledger, it said, was still open. This book tries to draft the next entry. It asks which of the five forces will be available in the next fifty years, and then does what the first book refused to do: it builds a society beyond the farm and attacks it with everything history knows.

Seven assumptions

The book is a thought experiment, and a reader is entitled to see its assumptions. They were set by the person who commissioned it.

1. A fair share is weighted by contribution, with a floor that nobody falls below.
2. The experiment is carried out in good faith, from the conviction that a configuration with no farm is possible — for most of human history, after all, people lived in small and often fiercely egalitarian groups. Good faith also means reporting honestly when the design fails.
3. AI has a wall. It is smart and very capable and can take over much management, coordination and routine production, but it is not a true replacement for human labour.
4. The wall is creativity. The transformation of frames — the new question, the new form, the new institution — remains human.

Three refinements make the wall more precise.

5. The agents are driven. Coding agents are the strongest case for AI's power. The research group METR found that the length of software tasks leading agents could complete half the time had doubled roughly every seven months for six years. Yet in a 2025 trial, experienced developers took about nineteen per cent longer with AI tools on their own projects while believing they had been sped up, and in a simulated software company built by Carnegie Mellon researchers the best agent finished only about three tasks in ten on its own. Every agent, and every swarm of agents, is driven by people who set its goals, judge its results and step in. The wall is a wall of direction and responsibility.
6. The machines are heavy. AI runs on data centres, electricity and chips. The International Energy Agency expects data centres' electricity use, about 415 terawatt-hours in 2024, to more than double by 2030, and the largest technology companies planned to spend several hundred billion dollars on such infrastructure in 2026 alone. That weight is a constraint, and a point of ownership and leverage: data centres stand in particular places, on public grids, with local permission.
7. The machines are not only theirs. Open-weight models that anyone can download and run have stayed within a few months of the best private systems. When the Chinese company DeepSeek released one in January 2025, the chipmaker Nvidia lost $589 billion of market value in a day.

If the wall fell entirely, the 99's oldest lever — being needed — would go with it, and the design would have to rest on ownership alone. That possibility shaped everything the book built.

The design and the rule

The design under test, the "Federation hypothesis", has six elements: one borderless society; the machines, including AI, publicly owned; AI replacing the shepherds, so that no managing class can become an owning class; every person a worker and a shareholder; reward weighted by contribution; and a floor nobody falls below.

The rule of the book is that no part of the design survives unless history or evidence lets it. Every element faces four tests. Precedent: has anything like it been tried, and what happened? Evidence: what do the data say? Capture: how would the farm take it over? The first book's most consistent finding was that the farm reassembles itself around every reform — priests captured the gods' estates, slave soldiers captured the sultanate, Djilas's new class captured the revolution. Exit: does it leave people free to leave, refuse and dissent? A society people cannot leave is a farm, however well it is run.

The five forces, tested

Catastrophe will come, but cannot be the plan. On Monday 23 March 2020, with the pandemic spreading and markets in free fall, the Federal Reserve promised to buy bonds without limit. The stock market hit bottom that day and then rose for years; the richest captured most of the new wealth. Modern states protect concentrated wealth from the destruction that once levelled it. The pandemic did show something else: emergency floors cut poverty fast. The lesson is that crises reward whoever has their ideas ready.

Scarcity is the strongest force in the 99's favour since the Black Death — if the wall holds. Ageing makes workers scarce across the rich world and China. But Japan shows that scarcity alone delivers nothing: owners import less-protected workers, move the work or automate. The anthropologist James Wright found that the care robots promoted by Japan's government made more work in the care home he studied, not less. AI reaches deepest into office work, the shepherds' work, and early payroll evidence suggests entry-level professionals are hit first. The wall sits, in order, at creativity, care, judgement, physical work and trust; and under the refinements, at direction and responsibility. The danger the book flagged here is that creators, or the owners of their rights, become the next aristocracy.

Organisation is possible in new forms. On Labour Day 2023, content moderators who clean the feeds of the largest technology companies voted in Nairobi to form a union. Drivers won worker status in Britain's Supreme Court; Hollywood's writers struck and won rules on AI. In Amsterdam in 2023, a court found that Uber's human review of its automated dismissals was "not much more than a purely symbolic act". Algorithmic management is the first real test of AI as shepherd, and it has hidden the managers' power rather than removing it.

The owners are divided, even when they sit in a row behind a new president, as the heads of the largest technology companies did at the inauguration of January 2025: nation against world, AI companies against the owners of creative rights, giants against rivals, states against owners, some of the very rich asking to be taxed. Every gain the scattered won came with help from a split above.

Gains last when they are universal, automatic, structural and closed to exit. The expanded American child tax credit of 2021 roughly halved child poverty and then expired. Alaska's constitution protected its oil fund but not the dividend, which a governor's veto cut in half. The one large lasting win against great wealth since 2008 is the end of bank secrecy: automatic exchange of information, coordinated across more than a hundred jurisdictions and enforced by access to the largest markets.

Two more tests completed the picture. The story of meritocracy — exposed by the 2019 admissions scandal and its legal "back door" for the rich — is losing believers, and the design's own story of reward for contribution could become its successor unless reward is told as reciprocity, not proof of worth. And the owners' exit — private cities, purchased passports, tax flight, and the dream of a worker-less machine — is the newest danger. But the machine is heavy: a gigawatt data centre cannot follow its owners into a bunker, and local opposition had blocked or delayed some $64 billion of American projects in about two years. If an "intelligence curse" comes, it will sit in the physical layer. No one may leave with the commons, and the 99's claim on the common wealth must never depend on being needed.

Building the design, and attacking it

Everyone owns. In 1957 Milovan Djilas, once Tito's possible successor, published The New Class from a Yugoslav prison: the communist revolutions had abolished the owners but not ownership, which had passed to the party bureaucracy that controlled the people's property. Russia's 1992 vouchers taught the second half of the lesson: shares given to everyone, if they can be sold, end up with the few. Against those failures stand Norway's oil fund, protected by a fiscal rule and transparent down to every holding; Alaska's equal dividend; and the Mondragon cooperatives, with elected managers and pay ratios of around six to one. Sweden's Meidner plan, which would have moved ownership of industry gradually to funds for wage-earners, was defeated in the 1980s when the owners united and its supporters divided. Ownership, the book concludes, is a bundle — control, income, transfer — and every strand must be kept broad: income equal and direct, capital locked, shares unsaleable, control divided. For AI, the capital to own is the physical layer, which communities have owned before: half the Middelgrunden wind farm outside Copenhagen belongs to a cooperative of some eight and a half thousand people, and since 2009 Danish law has required new onshore wind farms to offer local residents at least a fifth of the shares. Open models are the commons' starting stock.

Who governs the governor. In 2010 Seema Misra, a sub-postmistress in Surrey, was jailed while pregnant because the Post Office's Horizon accounting system said £74,600 was missing. The system was wrong, but it was presumed right and could not be inspected; her conviction was quashed in 2021, and in 2024 Parliament quashed the remaining Horizon convictions in England and Wales by law. Australia's Robodebt and the Dutch childcare benefits scandal showed the same pattern. AI can keep the books, but it cannot be the governor. The book splits the shepherd's office into five parts that cannot act alone: the machine; its managers, on fixed terms and modest pay; rule-setters chosen by lot, as in Ireland's citizens' assemblies; independent auditors, never paid by those they audit; and tribunes with a veto, as Germany's works councils have over workplace monitoring. The burden of proof lies with the machine, and every check has a deadline.

Contribution and the dividend. On 24 October 1975, about nine in ten Icelandic women stopped all work, paid and unpaid, and the country could not function for a day. The International Labour Organization estimates that people spend some 16.4 billion hours a day on unpaid care. The design counts six kinds of contribution — production, care, learning, creation, risk and maintaining the commons — mainly in time, an hour of care equal to an hour of production, measured coarsely by rules that councils chosen by lot set, with low stakes. The floor is an equal dividend plus universal services, given as a right of belonging; the evidence from Alaska, Finland, Kenya and the United States suggests people keep contributing. Top incomes are capped within a band of a few times the lowest. Creators are honoured but cannot become an aristocracy: their reward is income that cannot be inherited, cannot buy a larger share of the commons, and comes with short, unsaleable rights over their work.

One society or many. In 1948 Garry Davis gave up his American citizenship to become a "citizen of the world" and spent the rest of his life learning, as Hannah Arendt had argued, that rights depend on a community that will defend them. A single world state would be what Edward Gibbon called the Roman Empire at its height, "a safe and dreary prison" with nowhere to flee. The design becomes a federation of many societies sharing one floor, with the doors closed to hidden wealth, passports for sale and private jurisdictions, and open to people. Everyone belongs from birth and receives a global dividend from the global commons wherever they live; a member society may leave, but not with the commons.

What the tests taught, and what stands against the design

The four tests did not weigh equally. Precedent kept the design modest: the cases that worked were small, specific and patient — a protected fund, an equal dividend, a pay cap, a veto, an assembly with a deadline — while the grand designs failed or were never tried. Evidence was strongest on the floor and weakest on everything new; on common ownership of AI or governing machines by citizens chosen by lot, there is almost none, because none has been tried. Capture changed the design more than any other test: every time the book asked how the farm would take a piece over, the piece had to grow another defence. And exit changed the shape of the whole, turning one society into many and telling the design which doors to close and which to open.

Some objections stand. The economist Friedrich Hayek argued in 1945 that the knowledge needed to run an economy is scattered among millions of people, and that prices coordinate it as no planner can. The design keeps markets for goods and exchange; what it holds in common is the capital of the machines. But whether machines under public rules can use scattered knowledge better than the Soviet planners did is unknown. Compressed rewards might slow innovation, although the post-war decades, the Nordic countries and Mondragon suggest economies can thrive with far narrower gaps than today's. A society of councils, audits and assemblies asks a great deal of its members' time. And the oldest objection — that the owners are too strong and the 99 too divided — is the one history supports most strongly, and has also most often proved wrong, slowly and then suddenly, when the forces lined up.

The final ledger, in brief

What survived. A world with no outside for the commons and no stateless people. Common ownership of the machines, above all the physical layer, in trusts whose capital is locked, with equal and unsaleable shares for every person. AI doing the shepherds' paperwork. Every person a worker and a shareholder. Reward for contribution, broadly defined and justified as reciprocity. A floor nobody falls below.

What had to change. One borderless society became a federation. "Public" stopped meaning "owned by the state". The machine became the governor's instrument, never the governor, with its office split five ways and a real human review. Independent unions and associations became permanent. Contribution is counted in time, measured with low stakes, and capped. Creators were denied the means to found dynasties. The floor became a right of belonging, protected from any single veto. And the design must change slowly, by a citizens' assembly and a referendum, never by a faction alone.

What remains unknown. Whether the wall holds and agents stay driven by people. Whether common trusts can run the most advanced machines as well as private companies, and whether open models stay close to the frontier. Whether citizens can understand the machines well enough to govern them, or whether the interpreters become a new priesthood. What a permanent floor does over a lifetime. Whether the most powerful states would ever join a federation that bound them. And how any of it could be built from the world as it is.

Three futures, and how to begin

The coming decades could bring three futures. In the Managed Herd 2.0, the owners keep the machines and pay out a feed — a basic income at someone's discretion, cheaper services, more entertainment. Nobody starves; nobody owns anything that matters. In the Enclosure of Intelligence, the owners of the machines no longer need the many and withdraw from them, as the rulers of rentier states could ignore their people. In the Federation, the machines are owned in common and governed by split institutions, people do the work only people can do, and no small group owns the surplus. The signs that tell them apart are measurable: who owns the frontier models and the physical layer, who directs the agents, whether pay rises at the wall, whether the doors to hidden wealth close, whether human review is real, and whether floors are universal or discretionary.

The Meidner plan's defeat suggests how to begin. Start with new wealth rather than old: the AI systems and data centres being built now. Claim a share of the physical layer for the communities that host it, as Denmark did for wind. Give every person a stake from the first day, so that those who benefit are those who vote. Lock each step as it is taken. Use the owners' splits rather than uniting them. Cross borders from the start, as the end of bank secrecy did. Have the ideas ready for the next crisis. And start with a shop: cooperatives, public models in trusts, citizens' assemblies, works councils with a veto — each a Toad Lane, a place to learn before anyone bets everything.

The next entry

The book ends by drafting the entry it promised:

Received: the work of every person — the making, the caring, the learning, the creating, the risks taken and the commons kept. Held in common: the machines, the models, the knowledge of the species and the resources of the earth, unsaleable and open to inspection. Paid to every person, by right of belonging, wherever they live: an equal share, and a floor that no one falls below. Paid for contribution, within a narrow band: the rest. Kept by: the machine, under the eye of everyone it counts.

It is only a draft, and every line carries a verdict and a list of unknowns. One disclosure belongs with it: this book was written by an AI. Its question, its conviction and its assumptions came from the person who commissioned it; the machine's work was to build, test and keep the accounts. If that is a fair picture of what is coming, it is a hopeful one. It means the machines can keep the books. It does not mean they should decide what is written in them.

The weavers of Toad Lane opened their shop by candlelight because no one would sell them gas, and the ledger they started is still open. So is Kushim's. The next entry has been drafted. Who will write it?