Capital No Longer Needs a Lawyer (and Perhaps Not Even a Capitalist)

Why does law exist? And what, exactly, do we need lawyers for? Until recently these were questions for philosophers, but AI has made them practical. My answer is that the economic necessity of the legal profession is historically temporary, and that it will be largely gone within ten years. By 2036, I expect autonomous agents to perform nearly all the paid intellectual work of applying law to particular relationships, except where the law itself still requires a human being. Law and its institutions will remain, but a client will be able to obtain a legally usable result without employing a human lawyer. To explain why, I have to start with what legal work actually is.

Ma'at

In many ancient cultures law belonged to a sacred order. Human rules did not stand apart from Being: they were part of an order that governed the gods in heaven as well as people on earth. In Egypt the goddess Ma'at personified truth and justice and embodied the order of the cosmos, and the word ma'at named the standard by which human conduct and institutions were judged. To uphold the law was to help maintain that order. [1]

Other traditions put it in their own terms. On the stele of Hammurabi, the king stands before Shamash, the god associated with justice. [2] The Laws of Manu present rules of conduct within a religious conception of duty. [3] At the opening of Justinian's Digest, Ulpian writes that jurists may rightly be called priests, because they cultivate justice and distinguish the equitable from the inequitable. [4] These traditions are not the same, but each time people measured particular human affairs against an order they held to be binding.

For this essay, I have borrowed the word ma'at to describe something every society has: an established measure of what is due, permitted or required. It may arise by agreement, grow out of custom or be imposed by force. It determines people's legal positions even when they believe it treats them unjustly.

Every society has its own ma'at, and particular relationships are measured against it. Every such order has also needed people who know it.

A priest, a scribe, a Roman pontifex, a jurisconsult and a modern lawyer have this in common: each helped people deal with rules they did not know or did not understand. Someone had to know what the rules meant and how they applied to a particular case. The lawyer is, above all, an intermediary between the legal order and the people who live under it.

A corporate lawyer acting for the seller of a company does the same job. The client's goal is commercial: to sell the company on acceptable terms. Statutes and judgments set the limits within which the parties may act. The lawyers on both sides examine those limits, choose a structure for the deal, allocate risk and turn what was agreed into documents with legal effect. The job is still the old one: to bring a particular relationship into line with the applicable measure.

Within the limits set by the legal order, the parties may write many of their own rules. They can negotiate a private, "local" ma'at, and the lawyer helps them define it and give it legal force. Party autonomy changes the work of the intermediary, but it does not eliminate the work of establishing what the parties may lawfully agree and what their agreement will mean.

The economics of legal intermediation

Imagine that every time two people entered into a legal relationship, they had to work out for themselves what each was allowed to do, what each could demand and how a dispute between them would be resolved. They would even have to settle a more basic question: whether each of them was recognised as a legal person. The most ordinary transaction would become expensive, and sometimes the cost of reaching an agreement would exceed the value of the deal.

Part of the cost comes from the fact that knowledge is limited and unevenly spread. Finding information, understanding it and deciding what to do with it all take time and money. A legal order settles many of these questions before any particular relationship arises. Nobody has to invent the rules of ownership, contract or liability to buy a chair or to sell a company. The parties start from a measure that already exists.

Law and economics describes this problem through the concept of transaction costs. In The Problem of Social Cost, Ronald Coase showed that if transactions cost nothing, the parties could bargain their way to the most valuable use of resources whatever rights the law had first given them. Real transactions are not free. Finding a counterparty, negotiating, drafting the agreement and checking that it is performed all cost money. These costs may prevent agreements that would benefit both sides. That is why the initial allocation of rights by the legal system matters economically. [5] In his Nobel lecture Coase put it directly: once we move from zero to positive transaction costs, "what becomes immediately clear is the crucial importance of the legal system in this new world." [6]

Even when the rules exist, applying them to a particular relationship is not free. Someone has to find the relevant law, work out what it means, establish the facts and decide what follows from them. In a company sale, someone must also choose the structure of the deal, decide which risks each party bears and express the agreement in terms a court will enforce.

This is the lawyer's work, and it is why the lawyer gets paid. The lawyer already has the knowledge and experience the parties would otherwise have to acquire, so they can buy it instead of spending their own time on it. The fee is a cost too. Hiring a lawyer makes economic sense when the expected cost of obtaining a legally usable result through one, including the risk of mistakes, is lower than the cost of a comparable result obtained another way. A client may do the work alone, use standard documents, rely on an employee or purchase software. The important question is what each alternative costs and what result it produces.

There are therefore two economic functions here. The legal order provides a common measure against which relationships are judged. The lawyer applies that measure to particular affairs. The order can outlast any change in how its rules are applied. The economic necessity of paying a human being to apply them depends on something much less permanent: what it costs to have the work done some other way.

When legal work becomes machine work

Consider what a lawyer actually does, starting with the work of a junior lawyer or a paralegal. It has become almost commonplace to say that modern AI models can take over much of a lawyer's technical work. By technical work I mean work that involves less direct legal reasoning in forming a position and more searching for information, reviewing documents and preparing them according to established practice.

A junior lawyer receives a four-hundred-page share purchase agreement and is asked to proofread it. Cross-references must lead to the correct provisions, defined terms must be used consistently, and the document should contain no obvious contradictions or grammatical mistakes. Well, today that lawyer can give much of this work to an AI.

The same applies to legal research. Finding relevant case law often means going through hundreds of decisions containing the right keywords but having little to do with the actual question. A model can help distinguish decisions that matter from those that merely use similar words. Not every court decision is available to every model, and the answer still needs checking. But I would be surprised to meet a lawyer today who did not at least ask AI where to begin before spending hours going through search results alone.

Drafting used to be a different story. Once upon a time, in a galaxy far, far away (roughly two years ago), texts produced by AI models could be painfully average and full of what I can only call AI slop. They looked reasonable, sometimes even professional, but often failed to express what the user actually wanted. A model could produce a respectable-looking contract without attending to the transaction it was supposed to govern. The arrival of reasoning models changed this considerably. Beginning with OpenAI's o1, introduced in September 2024, models became better at working through difficult instructions and alternatives before answering. [7] I think this has already changed the old drafting problem fundamentally. Today it is much easier to ask a model not simply to reproduce a standard clause, but to consider its commercial purpose and prepare something that fits the transaction.

There was, however, another part of legal work that seemed harder to entrust to a machine: negotiating directly with the other side. A model could prepare arguments, draft terms, find weaknesses in an opponent's position and suggest concessions. But someone still had to conduct the negotiations on behalf of the client.

In other words, the model could give us the spears and even explain how to use them. But we were not prepared to let it throw them itself. Or at least that was the assumption. I think this boundary is now beginning to move.

When agents negotiate

In September 2026, Meta introduced Muse, a personal AI agent able to take actions in a browser and connected applications on behalf of its user. [8] That same month, tech reviewer Matt Robb described how the agent handled the proposed sale of his keyboard on Facebook Marketplace. It negotiated with a buyer, shared Robb's home address and arranged for the buyer to collect the keyboard. There was just one problem: Robb was not at home and did not know anyone was coming. He had selected an "Allow Always" permission, expecting another opportunity to approve the transaction. [9]

What interests me here is not the keyboard. It is that an AI agent negotiated with another person without its owner's direct participation. The experiment was not exactly successful. But the model was no longer simply suggesting what its owner should say. It was making arrangements on his behalf. Well, it seems that the spears had already been thrown.

Now consider a more deliberate thought experiment. One person wants to sell an old oak chair for $500, and another wants to buy it. Each has an AI agent.

The buyer's agent offers $450 and asks for a one-year warranty against hidden structural defects, with the right to return the chair for a full refund. The seller's agent refuses. Its owner knows of no defects, but cannot guarantee the condition of the wood inside. It offers a thirty-day return period instead.

The buyer's agent objects: a defect may be unknown to both parties, and thirty days may not be enough to discover it. It offers the full $500 in exchange for the one-year warranty and compensation if a hidden defect damages other property. The seller's agent accepts the year, but requires proof that the defect existed at delivery. Ordinary wear and misuse are excluded; compensation for damage to other property is capped at $200. The buyer's agent accepts, provided that a defect can be reported within thirty days of its discovery. The seller's agent agrees.

Neither agent knows whether the chair has a hidden defect. But that is precisely why the parties need an agreement. Someone must decide who bears the loss if a defect is discovered. The agents have negotiated that risk, and the buyer has agreed to pay $50 more for better protection. They can arrange payment and delivery and submit the terms to their owners for approval.

What have the agents done? They have considered their owners' interests, negotiated a price and allocated risks. Once approved, their agreement will determine the parties' rights and obligations. In other words, they have designed what I called a private, "local" ma'at. The buyer and seller remain the parties to the contract, while the agents perform the intellectual work of negotiating it. Neither agent needs to become a legal person, and neither owner needs to employ a human lawyer.

Nor must human approval necessarily be a condition of every transaction. Section 14 of the Illinois Uniform Electronic Transactions Act expressly allows a contract to be formed by interacting electronic agents without any individual reviewing their actions. Under US federal law, the involvement of an electronic agent does not, by itself, deprive a contract of legal effect, provided the agent's action is legally attributable to the person to be bound. [10] [11] The law has already found ways to recognise automated contracting. What changes with AI is the range of decisions that the agents themselves can negotiate.

Of course, people do not normally hire lawyers to buy wooden chairs. So let us replace the chair with a company.

A company may have thousands of contracts, debts, disputes and obligations, each with its own legal consequences. Someone must examine them, establish the facts, identify the law, decide whether to buy shares or particular assets and negotiate representations, warranties and indemnities. Much of the difficulty lies in the volume of information and the relationships between documents. This is where I expect AI to have an advantage: an agent can examine contracts, compare provisions, identify contradictions and trace possible liabilities back to their sources.

Suppose the buyer's agent discovers that an important customer contract allows termination upon a change of control. It examines whether the proposed transaction would trigger that right and what losing the customer would mean for the company's value. The agent demands the customer's consent before closing. The seller's agent offers a limited indemnity instead. But the buyer's agent may conclude that compensation would not replace the lost commercial relationship. The agents must negotiate whether consent should be a condition of closing, whether the buyer should accept the risk for a lower price, or whether some other arrangement would satisfy both parties.

The legal problem is not fundamentally different from the problem with the chair. The parties face a risk, and someone must find out what can be known about it, decide who should bear it and give that decision legal effect. The company makes the work much harder, but not a different kind of work. If agents can perform it reliably, the commercial purpose of paying a human lawyer to do it begins to disappear.

What is left of the lawyer?

I think the economic consequence is clear. If parties can obtain a legally usable result directly from their agents, why should they continue to pay another human being to perform the same work? Legal work does not disappear. The law must still be found, the facts established, the documents prepared and the terms negotiated. But the parties can obtain that work from systems they use themselves. We are no longer speaking only about a law firm becoming more productive. We are speaking about the parties no longer needing to turn to a law firm at all.

The economic threshold is reached when an agent can provide a comparable legal result at a lower total expected cost than a human lawyer. That cost includes verification, possible mistakes, liability and putting mistakes right. People pay lawyers because doing so is economically rational. When they can obtain the same result more cheaply without one, the reason to pay disappears. I believe increasingly capable agents will cross this threshold, first for particular tasks and eventually across most of the profession.

Disputes may seem harder. A seller can refuse to perform an obligation. A buyer can claim that a defect was concealed. The parties may disagree about what their contract means. Yet much of a dispute lawyer's work consists of examining facts, assessing what a court is likely to decide and negotiating a settlement. Robert Mnookin and Lewis Kornhauser described this as bargaining "in the shadow of the law": expectations about a court's decision affect what people accept outside the courtroom. [12] Agents can assess those outcomes and negotiate on that basis too. Where negotiations fail, a court or another authorised decision-maker may still be required. But the need for a court does not itself establish the need for a paid human lawyer on either side.

Then there is authority and responsibility. What happens when an agent exceeds its instructions or makes a mistake that causes financial losses? A client who hires a lawyer receives professional duties and the possibility of seeking compensation for their breach. An agent-based system needs rules for allocating responsibility between users, providers and other participants. Conflicts may also arise if both parties' agents depend on the same provider. Noam Kolt addresses these problems in his work on governing AI agents. [13] Their costs belong in our economic comparison. But responsibility can be organised through contracts, insurance and legal rules. It does not follow that the intellectual work must be performed by a human lawyer.

Regulation presents a different question. Garfield.Law, an AI-driven legal service authorised in England and Wales by the Solicitors Regulation Authority, operates within a framework of human supervision and professional responsibility. [14] Such arrangements may remain for some time. But a legal requirement to involve a lawyer and an economic reason to hire one are not the same thing. The first depends on the rules established by the legal order. The second depends on what the lawyer actually adds to the result, and at what cost.

There is also an economic objection. As legal assistance becomes cheaper, people may use it more often. Someone who would never have paid a lawyer to review an ordinary contract may ask an agent to examine every agreement they sign. Businesses may formalise arrangements that they once left informal. The total amount of legal work may grow. But why should that create more jobs for human lawyers if the agents can perform the additional work? The US Bureau of Labor Statistics projects a 5 per cent increase in lawyer employment from 2025 to 2035. [15] My prediction concerns the profession's economic necessity as autonomous systems develop, not the employment figures already forecast on the basis of today's market.

I therefore expect the legal profession, as we know it, to lose its economic necessity. Its function is to connect the legal order to particular human affairs. Once people and companies can do this through their own intelligent systems, they no longer need to employ a separate professional for it. Legal scholarship, legislation and adjudication may develop in different ways. But producing law, deciding disputes and being paid to apply existing law to private relationships are different activities. It is the last one, the commercial profession of legal intermediation, whose disappearance I expect.

Capital without the capitalist

But the consequences of this development do not end with law. If agents can find the rules, negotiate agreements and arrange their performance, why should the owner of capital personally manage the economic activity through which that capital is preserved and increased? And if machines can perform much of the intellectual work of business, what will allow a person who owns very little to become a capitalist in the first place?

Of course, ownership and management have long been separate. Michael Jensen and William Meckling showed how delegated management creates agency costs: owners must monitor managers whose interests may differ from their own. [16] Now imagine that much of this management passes to an autonomous agent. Its owner sets goals and acceptable risks, while the agent searches for opportunities, negotiates transactions, invests money and monitors results. The owner still possesses the assets and receives the income, but much of the intellectual work of managing them is performed by a machine.

Imagine an heir who receives not only a family fortune but also a system capable of preserving and increasing it. The founder may have needed exceptional abilities to accumulate that capital. The heir may need far fewer abilities to keep it. What passes from one generation to another is both the property and the means of managing it. There is something rather feudal about that arrangement.

At first glance, the same technology seems to offer extraordinary opportunities to people who possess very little. Someone who wants to establish a company may no longer need to hire lawyers, programmers, accountants and analysts before even knowing whether the business will work. A small firm may gain access to intellectual capabilities that once required entire departments. I think this is a real expansion of freedom. But it leads to a problem I find much more disturbing: what if AI makes it easier to start a business while making it harder to accumulate substantial capital by building one?

For a long time, knowledge and judgement were economically scarce. A person who could write difficult software, understand an unfamiliar market, solve a legal problem or organise a business better than others could turn that ability into income and sometimes into property. A talented programmer could create a product without inheriting a fortune. A good lawyer could establish a practice. A founder could see an opportunity that larger competitors had missed. The ability to do something that most people could not do was itself an advantage.

But what happens when the ability becomes available to everyone through an agent? Take a photo editor as a simple example. Suppose a programmer creates a cheaper alternative to Photoshop. In the past, making the product required skills that customers did not possess. But imagine a future in which a customer can spend a month working with a $200 AI subscription and obtain an editor containing precisely the functions they need. They may no longer have much reason to pay the programmer. The same problem can arise with legal drafting, financial analysis, routine consultancy and many other forms of intellectual work. If a customer can obtain the result directly from an intelligent system, what exactly is left for the specialist to sell?

There will still be businesses to build and problems to solve. But I wonder whether the old route from personal ability to substantial capital will remain as open as it once was. If a useful idea can be implemented and imitated almost immediately, the person who first had it may capture very little of its value. The people who already control customers, distribution, proprietary information, physical assets and capital may capture much more. This is not an entirely new problem. David Teece explained how, when imitation is easy, the profits from innovation may go to the owners of complementary assets rather than to the innovators themselves. [17] AI could make that problem far more common.

Consider legal information. LexisNexis, ConsultantPlus and Garant are not merely collections of features that a programmer could reproduce. They have accumulated legal materials, established distribution channels, customer relationships and resources for developing new services. An independent founder may build an excellent legal AI system, but what happens when an incumbent can reproduce its functions while also giving its own agents access to information and customers that the founder does not possess? And a company with substantial capital can go further: it can finance infrastructure and train specialised or more powerful models of its own. The advantage is no longer simply knowing how to build something. It is owning what other people cannot easily reproduce.

Here I return to the moral promise of capitalism. Since the eighteenth century, one of its most powerful justifications has been that economic position need not be determined entirely by birth. A person could acquire property through work, knowledge, enterprise and the ability to organise the work of others. John Locke gave an influential philosophical account of the connection between labour and the original acquisition of property. [18] The American Dream later expressed a related hope: that one does not need to inherit a fortune to build one. Capitalism never guaranteed success to everyone who worked hard. But it held out the possibility that personal ability could change a person's economic position.

What happens to that promise when the economic value of personal intelligence falls, while the advantage of owning capital remains?

Imagine two people, one rich and one poor, with access to equally capable AI agents. Both agents know the same economics and law, find equally good opportunities and earn the same rate of return. Suppose the return is 10 per cent. An agent managing ten million dollars produces a million dollars; one managing a thousand produces a hundred. Both have used the same intelligence. Only one has enough capital for that intelligence to generate a substantial income. The machines may be equally capable, but they do not begin with equal resources.

Of course, the poorer person need not remain an investor of a thousand dollars. They may start a business, find customers and create something new. That possibility is precisely what concerns me. What if the work through which such a person could once build a fortune becomes easy for everyone to reproduce? The owner of existing capital can use an agent to preserve wealth, invest the returns and perhaps obtain better models and infrastructure. The person starting with nothing may find that the knowledge and technical abilities they could once have sold are no longer scarce enough to produce substantial capital.

I sometimes wonder whether we are living through one of the last periods when a technically capable person can still build a serious fortune with relatively little initial capital. Today there are unsolved problems, expensive professional services and transaction costs that a new company can reduce. Perhaps, in five years, many of these opportunities will be much harder to capture. I may be wrong about the timing. But I would not advise someone who wants to build a business to assume that the economic conditions for doing so will remain unchanged. Start now.

There is also the question of who will own the agents themselves. Computing infrastructure, databases, payment networks and the systems through which economic activity is organised may belong to a relatively small number of companies. Yanis Varoufakis has described a related development through the ideas of technofeudalism and cloud rent. [19] The owners of this infrastructure could collect income from activity carried out with other people's capital. A wealthy person might sit on an island while agents manage investments, develop products and negotiate contracts. For everyone else, the more important question is whether they can still build independent businesses or must operate on terms set by those who already own the infrastructure.

This is why proposals for distributing some of the gains from AI deserve attention. In Moore's Law for Everything, Sam Altman suggested an American Equity Fund financed by taxes on large companies and land, with annual distributions of money and shares to citizens. [20] That is not simply a proposal to pay everyone a subsistence income; it is also an attempt to give people a share in the assets that generate wealth. The distinction matters. A guaranteed income might prevent poverty without giving its recipients any realistic opportunity to acquire productive capital, build institutions or change their position in society. We could end up with a very comfortable world in which social mobility has almost disappeared.

And what would people do in such a world? If agents can develop products, manage companies, perform legal work and conduct scientific research, what becomes of the professions through which people have traditionally made themselves economically valuable? Perhaps even mathematics will eventually face this question if machines begin proving results that would otherwise have occupied generations of researchers. I do not know what intellectual work will remain uniquely human. Nor do I think that the disappearance of paid work would make human life meaningless. But I cannot regard a society in which most people are merely maintained by a distribution from an economy they cannot meaningfully enter as a satisfactory fulfilment of the promise of freedom.

None of this requires us to oppose AI. Quite the contrary. The same technology can reduce the cost of creating a company, defending one's rights and entering into agreements. Open models, inexpensive local systems and genuine competition between providers may allow people to use intelligent tools without surrendering control over their information and economic relationships. The question is not whether machines should become more capable. I want them to. The question is who will be able to use that capability to act independently, and who will own the resources on which it operates.

This brings me back to the beginning of the essay. The disappearance of the economically necessary human lawyer is one part of a much larger question about freedom. People should not need to employ a professional intermediary every time they want to understand their rights, establish a company, conclude a contract or defend themselves under the law. Cheap, accessible, locally operated legal AI can help them do these things directly. This is one of the reasons I am building Legal Machines. I want people to be freer in their relationships and transactions, not more dependent on whoever happens to control the intelligence they need.

Law will remain. We will still need rules, courts, property rights and institutions that protect people from arbitrary power. But we may no longer need the legal profession to stand between that order and the people who live under it. In that sense, the end of legal intermediation can be a step towards a technological utopia: a society where more people can act, create and cooperate without having to buy permission in the form of expensive expertise.

Capitalism presents a harder problem. Its promise was not merely that those who own property can preserve it, but that people who do not own much may create property of their own. If AI makes intelligence abundant while leaving the decisive advantages with those who already possess capital, that promise may become much harder to fulfil. We may then face a choice between a future in which intelligent machines expand human independence and one in which they make existing economic positions almost permanent. I want us to build the former. But I do not yet know how capitalism will preserve the social mobility on which so much of its moral justification depends.

Capital may no longer need a lawyer. It may no longer need the capitalist's personal intellectual labour either. But if it no longer needs human intelligence to reproduce itself, will people who have not inherited capital still have a realistic chance to create capital of their own?

REFERENCES

[1] The Metropolitan Museum of Art (n.d.) ‘Maat’, object no. 10.130.1302. Available at: https://www.metmuseum.org/art/collection/search/545360. ↩

[2] Musée du Louvre (n.d.) ‘The Code of Hammurabi’. Available at: https://www.louvre.fr/en/the-code-of-hammurabi. ↩

[3] Olivelle, P. (2005) Manu’s Code of Law: A Critical Edition and Translation of the Mānava-Dharmaśāstra. Oxford: Oxford University Press, Introduction and Book 1. Available at: https://doi.org/10.1093/oso/9780195171464.001.0001. ↩

[4] Justinian (533 CE) Digest, 1.1.1.1 (Ulpian). Latin text and English translation available at: https://reader.humanitext.ai/en/read/urn%3Acts%3AlatinLit%3Aphi2806.phi002.humanitext-lat1%3A1.1.1.pr-1.1.1.4. ↩

[5] Coase, R.H. (1960) ‘The Problem of Social Cost’, The Journal of Law and Economics, 3, pp. 1–44. Available at: https://doi.org/10.1086/466560. ↩

[6] Coase, R.H. (1991) ‘The Institutional Structure of Production’, Nobel Prize lecture, 9 December. Available at: https://www.nobelprize.org/prizes/economic-sciences/1991/coase/lecture/. ↩

[7] OpenAI (2024) ‘Introducing OpenAI o1-preview’, 12 September. Available at: https://openai.com/index/introducing-openai-o1-preview/. ↩

[8] Meta (2026) ‘Introducing Muse: The World’s First Personal AI Agent Built for Everyone’, 8 September. Available at: https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/. ↩

[9] Bhuiyan, J. (2026) ‘Meta’s AI agent Muse gives out user’s home address without permission, sending buyer to his house’, The Guardian, 29 September (first published 28 September). Available at: https://www.theguardian.com/technology/2026/sep/28/metas-ai-agent-muse-home-address. ↩

[10] Illinois General Assembly (2021) Uniform Electronic Transactions Act, 815 ILCS 333/14, § 14(a)(1), (3). Available at: https://www.ilga.gov/documents/legislation/ilcs/documents/081503330K14.htm. ↩

[11] United States Congress (2000) Electronic Signatures in Global and National Commerce Act, 15 U.S.C. § 7001(h). Available at: https://www.law.cornell.edu/uscode/text/15/7001. ↩

[12] Mnookin, R.H. and Kornhauser, L.A. (1979) ‘Bargaining in the Shadow of the Law: The Case of Divorce’, Yale Law Journal, 88(5), pp. 950–997. Available at: https://doi.org/10.2307/795824. ↩

[13] Kolt, N. (2026) ‘Governing AI Agents’, Notre Dame Law Review, 101, pp. 335–386. Available at: https://ndlawreview.org/governing-ai-agents/. ↩

[14] Solicitors Regulation Authority (2025) ‘SRA Approves First AI-Driven Law Firm’, 6 May. Available at: https://www.sra.org.uk/news/news/releases/2025/garfield-ai-authorised/. ↩

[15] U.S. Bureau of Labor Statistics (2026) ‘Lawyers’, Occupational Outlook Handbook, employment projections 2025–2035. Available at: https://www.bls.gov/ooh/legal/lawyers.htm. ↩

[16] Jensen, M.C. and Meckling, W.H. (1976) ‘Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure’, Journal of Financial Economics, 3(4), pp. 305–360. Available at: https://doi.org/10.1016/0304-405X(76)90026-X. ↩

[17] Teece, D.J. (1986) ‘Profiting from Technological Innovation: Implications for Integration, Collaboration, Licensing and Public Policy’, Research Policy, 15(6), pp. 285–305. Available at: https://doi.org/10.1016/0048-7333(86)90027-2. ↩

[18] Locke, J. (1690) Two Treatises of Government, Second Treatise, Chapter V, § 27. Available at: https://www.gutenberg.org/files/7370/7370-h/7370-h. ↩

[19] Varoufakis, Y. (2023) Technofeudalism: What Killed Capitalism. London: The Bodley Head. See author’s preface: https://www.yanisvaroufakis.eu/2023/09/07/technofeudalism-has-just-arrived/. ↩

[20] Altman, S. (2021) ‘Moore’s Law for Everything’, 16 March. Available at: https://moores.samaltman.com/. ↩