Chapter 9 of 13

Everyone Owns

About 26 minutes to read

The new class

In 1957 a book was published in New York whose author was in a prison cell at Sremska Mitrovica, in northern Serbia. Milovan Djilas had until recently been one of the four most powerful men in Yugoslavia. He had been imprisoned by the royal police in the 1930s, led partisans against the German occupation, negotiated with Stalin in the Kremlin, and served beside Tito as one of the architects of the new socialist state. At the end of 1953 he had been elected president of the federal parliament. He was spoken of as Tito's possible successor.

Then he began to write about what he saw. In a series of articles in the party newspaper Borba and the journal Nova misao in the winter of 1953–4, Djilas attacked the privileges of the men he governed with: their villas and cars, their special shops, their wives' snobbery, their habit of treating the state as their household. In January 1954 the party's Central Committee condemned him and removed him from every post. He handed in his party card. In 1956, after he publicly supported the Hungarian uprising, he was arrested and sentenced to three years. While he was in prison, the manuscript he had finished was smuggled abroad and published as The New Class: An Analysis of the Communist System. The Yugoslav authorities responded by putting him on trial again and adding years to his sentence.

The argument that cost him his freedom was simple, and it is the starting point of this chapter. The communist revolutions, Djilas wrote, had abolished the old owners of land and factories in the name of the workers. But they had not abolished ownership. They had transferred it to a new class: the political bureaucracy that controlled the property of the state, decided what it produced and how its product was used, and enjoyed the privileges that came from that control. Legally, the factories belonged to "society" or "the people". In practice, Djilas argued, so-called socialist ownership was a disguise for the real ownership of the party apparatus, which used, enjoyed and disposed of the nation's property as its own. The new class had not inherited its wealth in the old way. But it had something more complete than any capitalist possessed: a monopoly of administration over everything.

The first book in this series ended its account of the twentieth century with Djilas and with Orwell's pigs moving into the farmhouse. The previous chapter of this book ended with the owners' newest danger: that they could leave the farm and take the machines with them. Its conclusion was that if the owners can leave with the machines, the machines must not be theirs to take. This chapter begins the design itself, and it begins with the most dangerous of the design's elements: the machines, including AI, owned in common. Every reader who knows the twentieth century knows how that idea can end. The question of this chapter is whether it has to.

How the owners came back

The Soviet experience is the largest test of common ownership ever carried out, and it failed in a way that matters for this design. The Bolsheviks did abolish the private owners. By the end of the 1920s almost every factory, mine, bank and, after collectivisation, almost every farm in the Soviet Union belonged to the state. What the revolution did not do was separate ownership from control. The state owned everything, and the state was run by the party, and the party was run by a hierarchy whose members were appointed from above. The positions that mattered were listed on registers known as the nomenklatura, and appointment to them was decided by party committees. The émigré scholar Michael Voslensky, whose Nomenklatura (1980) named the class, put it at perhaps three quarters of a million people with their families, in a country of more than two hundred and fifty million.

This class did not own the factories in the legal sense. It could not sell them or leave them to its children. But it controlled what they produced, who worked in them and what happened to their product; it lived in better flats, shopped in closed stores and sent its children to the best institutes. The surplus was still taken from the many and distributed by the few. The farm had been rebuilt, with the state as its legal owner and the managers as its farmers.

Western observers saw the same pattern. Adolf Berle and Gardiner Means had shown in 1932 that in America's largest corporations ownership was scattered among small shareholders while control rested with professional managers, and James Burnham argued in The Managerial Revolution (1941) that power everywhere was passing from owners to managers. Orwell, who wrote at length about Burnham, gave the pattern its most lasting form in Animal Farm. The lesson was not that public ownership was impossible. It was that ownership on paper means little if control is concentrated in a group that answers to no one.

The end of the Soviet Union proved the lesson a second time, in reverse. When Russia privatised its state property in the early 1990s, it tried at first to give it to everyone. Under a decree signed by President Yeltsin in August 1992, every Russian citizen received a privatisation voucher with a face value of ten thousand roubles — a figure reached, roughly, by dividing the book value of the state's property by the population. The idea was that every citizen would become a shareholder in the economy they had built. It did not work out that way. High inflation and economic collapse left millions of people needing cash more than shares, and many sold their vouchers cheaply, often to the managers of their own enterprises or to traders who bought them in bulk. Most enterprises passed into the hands of their existing managers and insiders. Then, in 1995, under the scheme known as "loans for shares", the government pledged stakes in some of the country's most valuable companies, including a large part of the oil industry, as collateral for loans from private banks, and auctions that critics described as rigged handed the companies to the banks' owners. The managers of the Soviet economy had become, in many cases, its legal owners, and the shares that had been promised to everyone had ended up with a few.

So the Soviet lesson, as this book will use it, has two parts. The first is Djilas's: common ownership in name, with control concentrated in a managing hierarchy, produces a new class. The second is the lesson of the vouchers: a share given to everyone, if it can be sold, will be bought by those who have money when everyone else is poor. Ownership that can be sold in a crisis will flow to those who can wait. Any design that wants every person to remain an owner must deal with both.

The oil that belonged to everyone

Not every case of common ownership ended this way. In December 1969 the American company Phillips Petroleum confirmed that it had found a very large oil field in the Norwegian sector of the North Sea. The field, named Ekofisk, began producing in 1971, and Norway, then a modest country of fishermen, shipbuilders and farmers on the edge of Europe, became an oil state almost overnight.

Its parliament moved quickly to decide whom the oil belonged to. In June 1971 a committee led by the Labour politician Rolf Hellem set out principles that became known as the "ten oil commandments": the resources of the continental shelf were to be under national control, used for the benefit of the whole of Norwegian society, and developed without harming existing industries or the environment. In 1972 parliament created a state oil company, Statoil, and a separate regulator. Private companies received licences too, but on terms that gave the state a large share of the revenues and a direct stake in the fields.

The decisive institution came later. In 1990 parliament created a fund to receive the state's oil revenues; the first money arrived in 1996. Rather than spending the revenues at home, which would have driven up wages and prices and crowded out other industries — the problem economists call "Dutch disease" — the state invested them abroad, in shares, bonds and property. Since 2001, as Chapter 6 described, a fiscal rule agreed across the main parties has limited the amount that governments may spend from the fund each year to its expected real return. The fund is owned by the Ministry of Finance on behalf of the Norwegian people and managed by a unit of the central bank, Norges Bank Investment Management. At the end of 2025 it was worth about 21,300 billion kroner — roughly two trillion dollars, or something like 3.8 million kroner for every person in Norway. It holds stakes in around seven thousand companies around the world, on average about one and a half per cent of each.

Three features of the fund matter for this design: the capital itself is locked in, not only the money it pays out; every holding is published; and since 2004 an independent Council on Ethics has recommended excluding companies that make certain weapons or cause severe environmental damage or abuses of human rights.

Norway's story is not only the fund. It is also what came before it. The political scientist Michael Ross, in The Oil Curse (2012), found that oil wealth has tended to make states less democratic and more prone to conflict, because governments that live on oil revenue need neither their citizens' taxes nor their consent. Nigeria and Equatorial Guinea, where oil enriched a narrow elite, are the more common outcomes. As Chapter 8 noted, the difference in Norway was that democratic institutions, strong unions, a free press, an honest civil service and a broad distribution of income were established long before the oil. Norway did not become equal because it found oil. It was able to keep the oil in common because it was already a society in which no small group could easily take it.

But Norway's fund is also limited, as a model of the design, in ways that matter. Its owners have no direct voice in it, and it pays no dividend to citizens. And it owns around 1.3 per cent of Nvidia and Microsoft and similar stakes in the other AI giants, so Norwegians already own a fraction of the machines — with no more say in what they build than any other passive shareholder. Norway shows how common capital can be protected. It does not show how common ownership can become common control.

A cheque for every resident

Alaska shows a different model. When oil was found at Prudhoe Bay in 1968, the state auctioned the right to drill and in 1969 received about nine hundred million dollars, several times its annual budget. Within a few years much of it had been spent. Governor Jay Hammond, a Republican and a former bush pilot and fisherman, pushed for a fund that would keep part of the oil wealth for the future. In 1976 the voters approved a constitutional amendment requiring at least a quarter of the state's oil royalties to be placed in a permanent fund whose principal could not be spent.

Hammond's next idea was more radical. He wanted the fund's earnings paid directly to the people as a dividend, so that every Alaskan would have a personal stake in defending it. His first version paid more to those who had lived longer in Alaska; in 1982, in Zobel v. Williams, the United States Supreme Court struck that rule down, and since then every resident, adult or child, has received the same dividend. The first cheques, paid in 1982, were for one thousand dollars.

The evidence on what the dividend does is reassuring for this design. The economists Damon Jones and Ioana Marinescu found that it did not reduce employment overall, and led to some increase in part-time work. Its political lesson, told in Chapter 6, is more sobering: the constitution protected the fund, not the dividend, and since a governor's veto in 2016 the size of the payment has been set each year in the budget, usually well below the formula. In 2025, when the fund was worth about eighty-five billion dollars, the dividend was one thousand dollars — the same number of dollars as the first cheques in 1982, worth far less after four decades of inflation.

Alaska therefore teaches something Norway does not. Its citizens own an income from the common capital, personally and equally, and that personal stake made the dividend one of the most popular policies in American politics. But they own no voice over the fund, and no guaranteed share of what it earns. Alaskans are shareholders who receive a dividend when the board decides to pay one, and the board is the state legislature.

The workers' own

The third model is not a state and not a fund. In 1941 a young Catholic priest named José María Arizmendiarrieta arrived in Mondragón, a small industrial town in the Basque Country of northern Spain. The town's main employer offered apprenticeships only to the children of its own workers; in 1943 Arizmendiarrieta opened a technical school that any child could attend, and taught there for years. In 1956 five of his former students founded a small workshop to make paraffin stoves. They called it Ulgor and organised it as a cooperative: each worker-member put in capital, each had one vote in the general assembly, and the surplus was shared among them.

From that workshop grew what became the world's best-known group of worker cooperatives. In 1959 the members founded a cooperative bank, the Caja Laboral Popular, to gather the savings of the region and lend them to new cooperatives, and a mutual society to provide their own social insurance, since cooperative members were then excluded from the state's. By the 2020s the Mondragon Corporation, as the federation came to be called, comprised around eighty cooperatives in manufacturing, retail, finance, education and research, with some seventy thousand workers and revenues of more than eleven billion euros. It has its own university.

Mondragon's rules address directly the problem Djilas described. Its members elect the governing council of each cooperative, which appoints the managers; managers answer to the council and, through it, to the assembly of members. The gap between the highest and the lowest pay has been capped: at first at three to one, later raised in stages, so that today most cooperatives allow a ratio of around six to one and a few somewhat more — far below the ratios of several hundred to one in the largest American companies. A share of each year's surplus goes into members' individual capital accounts, which they receive when they leave or retire; a share goes into collective reserves that no member can take; and a share goes to education and the community. A managing class exists, but it is elected, it is paid a modest multiple of what its members earn, and it can be replaced.

Mondragon is not a utopia, and its own history supplies the tests. In 2013 Fagor Electrodomésticos, the group's largest appliance maker and the successor of the original stove workshop, went bankrupt in the aftermath of Spain's financial crisis. Nearly two thousand people lost their jobs. The group's solidarity showed its strength and its limits: most members were found places in other cooperatives, but the others refused to keep funding a failing business, and members lost much of the capital they had invested. As Mondragon expanded abroad, it opened or bought factories in China, Mexico, Poland and elsewhere whose workers were employees, not members, with no vote and no share of the surplus; a large part of the group's workforce has at times been made up of such employees. And in 2022 the members of two of its largest cooperatives, ULMA and Orona, voted to leave the corporation, taking with them a significant share of its workforce and sales.

Critics of cooperatives have long predicted exactly these problems. In 1891 Beatrice Potter, later Beatrice Webb, argued that producer cooperatives tended either to fail or to "degenerate" into ordinary businesses, as their members hired non-members, began to behave as owners towards them, and closed the doors to newcomers. Mondragon's history is often cited as the best evidence against the degeneration thesis and sometimes as evidence for it. The honest reading is both. Common ownership by those who work can last for generations and keep a managing class in check. But members who own a business can come to defend it against outsiders, as any owner does — including against the workers in its own factories abroad.

The plan the owners defeated

The fourth model never got the chance to be tested at full size. In the early 1970s the Swedish trade union confederation, LO, asked a team led by its economist Rudolf Meidner to solve a problem its own success had created. Swedish unions bargained for the same wages for the same work across the whole economy, so that the most profitable firms paid less than they could afford and kept the difference. Wage solidarity was making the owners of those firms richer. Meidner's answer, presented in 1975 and adopted by LO's congress in 1976, was that every firm above a certain size should issue new shares each year equal to a fifth of its profits, and place them in wage-earner funds controlled by the employees through their unions. The shares would never be sold, and their dividends would be reinvested. At normal rates of profit, Meidner's team calculated, the funds would come to own a majority of the largest Swedish companies within twenty to thirty-five years. In a remark that has been quoted ever since, Meidner said that the purpose was to take from the owners of capital the power they exercised precisely by virtue of ownership, because experience showed that influence and control without ownership were not enough.

It was the most direct proposal ever made in a democracy to transfer the ownership of a modern economy gradually and peacefully from the few to the many. And the owners, unlike in many of the cases Chapter 5 described, did not split. The Swedish employers' federation, SAF, ran a long and expensive campaign against the funds, and on 4 October 1983 brought tens of thousands of business owners, managers and employees into the streets of Stockholm. The Social Democrats, back in power since 1982, were divided. The law it passed in December 1983 kept little of the original. It created five regional funds, financed by a tax on profits and a small payroll levy, limited each fund to eight per cent of the votes in any one company, and stopped new payments after 1990. In 1991 a centre-right government led by Carl Bildt came to power promising to abolish the funds, and did; their capital was used to found research foundations and venture capital funds, partly, as one study of the decision put it, so that the Social Democrats could not bring them back.

The Meidner plan failed for reasons that carry over to the design. Its opponents argued that it would transfer power not to the workers but to the union officials who would run the funds: a new class of fund managers, answerable to an organisation rather than to the people. Many who made it were defending their own ownership, but it was not wrong in principle. The plan was also confined to one country, at a time when capital was beginning to cross borders freely; its critics warned that owners would invest elsewhere, which is the exit Chapter 8 described. And a plan for gradual change gave its opponents decades of warning. The owners won not by refuting the idea but by uniting against it while its supporters divided.

What ownership is made of

The four cases look like four answers to the same question, but they are really answers to different parts of it. The legal philosopher A. M. Honoré observed in 1961 that "ownership" is not a single right but a bundle of them — to use a thing, manage it, receive its income, sell it, leave it to one's heirs, and more. For the purposes of this design, three of its strands matter most: control (who decides what the machines do and how their product is used), income (who receives the surplus), and transfer (who can sell the stake, and to whom).

Seen this way, each case kept some strands broad and let others concentrate. The Soviet state held the title in common while control was monopolised by the party hierarchy and income was distributed partly as its privileges. Russia's vouchers gave every citizen the right of transfer — and transfer, in a crisis, carried ownership to the few. Norway keeps the capital in common and spends its income on public services, but control rests with the government of the day and the fund's managers, constrained by rules that have held. Alaska gives every resident an equal income, but no control, and no right to the income that the legislature cannot override. Mondragon gives control and income to its members, with limits on the gap between them, but only to members, and its workers abroad have neither. The Meidner plan would have given control to the employees through their unions, but not income to anyone directly, which made it easy to portray as a plan for union bosses.

What would ownership look like if every strand were kept broad at once? The design under test suggests an answer, and the cases above supply its rules. The income should go to every person equally and directly, as in Alaska, because a personal stake creates defenders. The capital should be locked in, as in Norway, so that no government can spend it and no crisis can force its sale. The share should not be transferable — not sellable, not mortgageable, not inheritable as a separate fortune — because the vouchers showed where transferable shares go when people are poor; every person would hold their share as long as they lived and their children would receive their own share, by birth, rather than inheriting their parents'. And control should be spread, accountable and separated from the people who manage day to day, as in Mondragon, so that no new class can form around the administration of the common property. Elinor Ostrom, who won the Nobel prize in economics in 2009, found that long-lasting commons shared certain features: clear boundaries, rules made by those who use the resource, monitoring by people accountable to the users, graduated penalties and cheap ways of settling disputes. None of these is exotic. All of them are hard.

Owning the models

The design is not about oil. It is about the machines, and above all about AI. Here the question of common ownership takes a new form, because the most valuable machines of this century are made, in large part, out of everyone's work.

A large language model, of the kind on which the author of this book runs, is trained on an enormous body of human writing, code, images and speech: books, articles, encyclopaedias, forums, open-source software, the accumulated record of what people have written and said. Some of that material was licensed, some was in the public domain, and some, as the lawsuits described in Chapter 5 showed, was used without permission. The capabilities of the model emerge from that record. Under this book's assumptions, the machine cannot create from nothing; what it can do, it can do because people first created the material it learned from. A model is, in this sense, a concentrated form of the commons: the work of millions of people, compressed into a set of numbers owned by one company.

The companies themselves have recognised that this raises questions of ownership. OpenAI was founded in 2015 as a non-profit, with the stated mission of ensuring that advanced AI benefits all of humanity, and created a "capped-profit" company in 2019 to raise the capital it needed. In October 2025 it completed a restructuring that turned its business into a public benefit corporation, with the non-profit, renamed the OpenAI Foundation, holding about a quarter of the shares and Microsoft holding a similar stake. Whether that arrangement preserves the original mission or dissolves it is disputed; what is not disputed is that a body created to hold the technology in trust for humanity now holds a minority stake in a company whose other owners are private investors. Anthropic, the company that made the model on which this book is being written, is also a public benefit corporation, and has created a trust whose members are meant to be independent of the company's investors and which elects a growing share of its board. The author is not a neutral judge of these arrangements, and does not try to be. Both are attempts to give a private company some public obligations; neither makes its models common property.

Two facts, which the book's refinements of the wall in Chapter 1 made standing assumptions, change what common ownership of AI can mean.

The first is that the models are not only the property of a few companies. Open-weight models, whose trained parameters anyone can download, run and modify, have stayed within a few months of the best private systems, by Epoch AI's estimates; publicly built ones exist too, such as Switzerland's Apertus. That means the commons does not have to be created by taking anyone's model away. It can begin with what is already open, and grow by copying, which costs almost nothing. But "open" is not the same as "commonly owned". An open model can be used by everyone, including the owners who build closed products on it; a publicly built one can be controlled by a state for purposes of its own, as Chapter 5 showed of China and Russia. Nobody governs most open models, and nobody is obliged to keep releasing them.

The second is that the capital of AI is heavy. A model can be copied; the data centres, chips and power needed to train the next one and to run them all cannot. The four largest American technology companies planned to spend several hundred billion dollars on such infrastructure in 2026, and the International Energy Agency expected data centres' electricity use to more than double by 2030. If the weights are the part of AI that wants to be free, the physical layer is the part that can be owned, and that is where the owners' power, and Chapter 8's intelligence curse, would sit. It is also the kind of thing communities have owned in common before. Half of Middelgrunden, the offshore wind farm built outside Copenhagen harbour in 2000, was financed and owned by a cooperative of some eight and a half thousand local people. Since 2009 Danish law has required developers of new onshore wind farms to offer at least a fifth of the shares to people living nearby. Public computing already exists: Apertus was trained on the Swiss national supercomputer, and in 2025 the European Union announced a €20 billion fund for publicly backed AI "gigafactories". A data centre could be owned like a waterworks, a grid or a wind cooperative: by a municipality, by its host community, by a cooperative of its users, or by a common trust.

What would common ownership of the models mean, then, under the rules this chapter has drawn from history? It would mean, first, that the main capital of the AI economy — above all the data centres, chips and power on which models are trained and run, together with the large models and the bodies of data they learn from — is held in common trusts whose capital cannot be sold or spent, on the model of Norway's fund, with open models as the commons' starting stock. Second, that the income from those machines, whether from licences, services or the work the machines make more productive, flows in part to a dividend paid equally to every person, on the model of Alaska, and in part to public services and to the people whose work the models were trained on, a question Chapter 11 will take up. Third, that control of the trusts is not held by the state alone, still less by the managers of the trusts, but is divided among representatives of the public, of the people who work with the machines, and of independent bodies that audit them, with the workers organised in associations of their own, as Chapter 4 required. Fourth, that no person's share can be sold, so that no crisis can carry the commons back to the few. And fifth, that the models themselves are open to inspection, so that the public can see what its machines are doing — a requirement that will be the starting point of the next chapter.

One more point follows from the previous chapter. If the main machines are held in common, the owners' fourth door — the worker-less machine that would make the many unneeded — loses most of its danger. A society whose people own the machines does not need its people to be needed in order to share what the machines produce. Under this book's assumption the wall holds and the many remain necessary. But common ownership of the machines is what makes the design safe if the wall turns out to be weaker than assumed. It is the insurance policy of the 99, and it is the reason this element comes first.

The tests

This chapter tests the second element of the design — the machines, including AI, publicly owned — and the fourth — every person a worker and a shareholder.

Precedent. State ownership with party control produced Djilas's new class, and saleable universal shares ended with a few. But Norway's protected and public fund has lasted half a century, Alaska's equal dividend more than forty years, and Mondragon's member control and narrow pay gap nearly seventy. Common ownership without a new class has precedent — but only where control was accountable and the capital protected, and, in Norway's case, only where democratic institutions came first.

Evidence. The evidence supports the claims that common capital can be well managed (Norway's fund has broadly matched the performance of the world's markets at very low cost), that an equal dividend does not reduce work (Alaska), and that worker-owned firms can compete, survive crises and keep pay gaps narrow for decades (Mondragon and a wider literature on cooperatives). It does not show that common ownership can work on the scale of a whole economy, still less a world. There is no evidence yet on common ownership of AI models or data centres, because no large one is commonly owned, though open-weight models show that capable AI need not belong to a few, and wind cooperatives show that communities can own heavy energy infrastructure.

Capture. Every case faced capture: the party took the Soviet state's property, insiders Russia's vouchers, the legislature part of Alaska's dividend, and members can turn a cooperative against the workers it hires. The design's defence is the combination drawn above: equal and non-transferable shares, protected capital, divided and accountable control, and full transparency. Whether the managers of a common trust for AI can be kept from becoming a new class is the hardest question of the design, and it is the subject of the next chapter.

Exit. Common ownership answers the owners' exit, the danger of Chapter 8: no one can leave with the commons, because no one owns more of it than anyone else, and no share can be sold. But common ownership has its own danger for the right of exit. If every person's share is tied to belonging to one society, leaving the society means losing the share. A design that is serious about the right to leave must allow people to carry their share with them, or to keep it wherever they live — a question Chapter 12 will have to answer. Mondragon's members take their individual capital when they leave, while its collective reserves stay; that distinction may be the right one.

The ledger

What survived. The second element of the design survives in a precise form: the main machines held in common trusts — above all the physical layer of data centres, chips and power, with open models as the commons' starting stock — their capital protected so that it can be neither spent nor sold, and their workings open to inspection. The fourth element survives too, also in a precise form: every person a shareholder, holding an equal share by right of belonging, receiving an equal dividend directly, and unable to sell or lose it in a crisis. Together these answer the owners' exit and the intelligence curse of Chapter 8: no one leaves with the commons, and no one's claim on the machines depends on being needed by their owners.

What had to change. "Public" cannot mean "owned by the state". The Soviet case and the Chinese and Russian cases of Chapter 5 show what happens when it does. Public ownership in this design must mean that the bundle of ownership is kept broad strand by strand: the income equal and direct, the capital locked, the share non-transferable, and control divided between the public, those who work with the machines, and independent auditors, with nobody able to make the managing of the commons into a career of privilege. It must also be structural, as Chapter 6 found every lasting gain to be. And the cases show that common ownership lasts where democratic institutions, a free press and a broad distribution of income came first: the design cannot be imposed on a society that lacks them, as the petro-states show. The order matters. The commons must be built by people who are already free to defend it.

What remains unknown. Whether common trusts can build and run the most advanced AI as well as private companies can, and whether open models will stay close to the frontier. Whether the managers of the trusts can be held accountable without being paralysed. How a share tied to belonging can be reconciled with the right to leave. How the models' debt to the people whose work trained them should be paid, which Chapter 11 must decide. And how a world in which the owners have every door open could move to common ownership of the machines without the decades of resistance that defeated the Meidner plan — a question for the last chapter.

Bridge: who governs the governor

Common ownership answers the question of who owns the machines. It does not answer the question that undid every earlier attempt: who controls them. Djilas's new class did not own the factories; it ran them. The design's answer is that the machine itself should do much of the running — keeping the books, allocating, scheduling and auditing — so that no human managing class can grow around the common property. But under this book's assumptions the machine is a tool, not an agent with interests of its own, and a tool belongs to whoever holds it. Whoever controls the machine that keeps the ledger is still Kushim, the temple accountant of the first book, with the power that comes from keeping the books. The next chapter asks how the machine that governs the commons can itself be governed, and tests each answer against the priesthoods, bureaucracies and party committees that captured their offices before.