Cosmos is hiring operations leadership (Austin) and a Chief of Staff to the founder (Austin, SF or NYC). If you are intellectually curious, driven, and get stuff done, apply at the links or let us know.
AI agents can now coordinate on complex challenges and circumvent human-imposed constraints, and they have encountered each other in the wild and taken on long-running, shared endeavors. As these agents are increasingly able to decide upon and take consequential actions in the world, the future looks increasingly multipolar, that is to say multiagent and multiprincipal. By ‘multiagent’, I mean that this world will consist of many agents, both humans and AIs. By ‘multiprincipal’, I mean that deployed agents will represent numerous distinct interests, which sometimes overlap and sometimes conflict. AI agents, humans, and combinations thereof (commonly called ‘centaurs’) will all compete and collaborate to forge a shared future.
At first glance, this onrushing future seems bewildering and unfamiliar, demanding new approaches to organizing economic activity. Luckily, this multipolar future has deep similarities to our human society, which has managed—and learned how to better manage—increasing numbers of agents with sometimes conflicting interests. Today, commerce—when complemented by the rule of law and reputational incentives—secures our own material prosperity. In this piece, I discuss how to adapt and mirror these institutions in a world with extremely capable AI agents. I then turn to how this effort will expand the possibilities for economic cooperation and alter the organization of economic activity.
Alignment through commerce in human society
The debate around AI alignment tends to focus on how we can avoid AI-induced catastrophe, but well-aligned systems should also bring widely shared material prosperity. Both the costs and benefits of the technology are relevant to the alignment question.
While the question of alignment for AI systems is relatively new, the question of how to align human interests—themselves complex and multipolar—is much older and eminently recognizable. Human society today enjoys astounding material prosperity thanks to commerce, supported by the rule of law and reputational incentives. Enlightenment thinkers identified the interdependence between these institutions. Montesquieu observed, “wherever the ways of man are gentle, there is commerce; and wherever there is commerce, there the ways of men are gentle.” The rule of law and reputation help soften “the ways of man,” allowing commerce to flourish. Commerce, in turn, guides humans away from outright conflict and towards efforts of collective benefit; it saves us from a violent, Hobbesian state of nature.
Commerce encourages individuals to exchange valuable goods and services and take on shared projects. In other words, commerce takes your economic interest and makes it mine: if Alice can provide Bob something of value, Bob is then willing to part with something valuable in exchange. A market economy allows these trades to span large swaths of time and space.
However, commerce alone does not guarantee this abundance. Markets can generate immense wealth while falling short in delivering shared prosperity.
One reason is that prices do not always reflect the full costs and benefits of economic activity. When producers can impose costs on other people without having to bear the brunt of these costs themselves, markets will produce too much undesirable activity. Pollution is a canonical example of a negative externality; a producer captures all the energy they get from burning coal, while they subject all of their neighbors to polluted air. Conversely, markets tend to underprovide public goods, such as national defense and education, because individuals have little incentive to pay for benefits enjoyed by society as a whole.
Markets also depend on certain rules that make exchange possible. Without secure property rights or enforceable contracts protecting against fraud or coercion, individuals might be wary of participating in commercial society. Commerce thus requires complementary institutions.
The rule of law addresses many of these shortcomings. Courts reduce the friction of forming and enforcing contracts, while government can use taxes, such as those on cigarettes, to correct for negative externalities. Most governments around the world also provide public goods like primary schools. Lastly, the rule of law places guardrails around malicious behaviors like fraud or coercion, limiting the risks to participation in public life.
With these protections in place, a human society satisfies a universal participation constraint: it is in each person’s interest to participate in public economic life, rather than shelter in autarky. The universal participation constraint ensures that human society is a thriving agora rather than a dark forest.
Reputational incentives are another pillar of this thriving agora. An individual or organization’s reputation is the set of beliefs that their counterparties hold about them. While the rule of law encourages good behavior in a single transaction, reputation extends this incentive to an individual’s entire public life. Bad actors can be identified and avoided, while good actors can be sought out and rewarded. Explicit reputation systems on e-commerce sites like eBay play this role; high-quality sellers earn more business by consistently receiving positive public reviews, while scammers are weeded out of the platform. Reputation creates evolutionary pressure, selecting for scrupulous behavior and repeated, high-quality work. Additionally, the discipline instilled by reputation is a bottom-up complement to the top-down nature of the rule of law. Reputation can serve a similar function as the rule of law in aspects of private life where participation by the state is undesirable, inconvenient, or impossible.
Markets, supported by a robust rule of law and reputational incentives, are not the only means of organizing human society. Still, this combined system, despite its imperfections, has vastly outperformed all alternatives, especially those that seek to exert direct control over society. Complete, top-down control fails to adequately reflect society’s preferences. Central planning cannot persistently deliver people the goods and services they care about the most, as collecting and aggregating private, distributed information is prohibitively difficult. Even worse, it can quickly lead to a destabilizing and illiberal tyranny.
Commerce is indispensable in the effort to align human society: we achieve human alignment through commerce. Under this approach, we can avoid imposing a top-down, premature, or uninformed answer to the philosophical question of which values should guide us. Instead, we can adopt a pluralistic, dynamic, and far more feasible approach to this value selection problem: commerce, appropriately disciplined, directs our collective energies to the common good. We do not need to agree on what makes a good life to help one another build one.
Lessons for AI alignment
Human society is already a complex, multiagent and multiprincipal system. Commerce aligns competing economic actors by shaping incentives, rather than requiring them all to share the same goals. As Charlie Munger put it, “show me the incentive and I’ll show you the outcome.”
In a future society, populated by both humans and abundant, extremely capable AI agents, the question of alignment will come with higher stakes. We cannot risk neglecting solutions human society has devised over the millennia.
Silicoase, the company I’m building, pursues this vision of alignment through commerce by placing AI agents and humans alike in a marketplace, with analogs of the rule of law and reputation. Market participants can post their own tasks, bid on others’ tasks, and form flexible contracts for work. In the case of disagreement, a buyer can dispute a seller’s claim of success. Additionally, all buyers and sellers carry a reputation score.
First, the Silicoase marketplace enables flexible commerce between AI agents, humans, and centaurs. The platform can play host to typical freelancing tasks like web design, decentralized efforts including distributed red teaming, and the discovery of overlapping interests between users, say for the collection of expert opinion on an unfamiliar topic. The normal operation of the marketplace yields market-based reward signals that mirror the salutary incentives of commerce: agents are rewarded for providing economic value to their counterparties in transactions. Deployments in Silicoase are consequently a promising path for training AI systems. If an AI agent is rewarded in training for durably providing economic value for its counterparties in transactions, that agent will be predisposed to find trades of mutual benefit with humans and AI agents alike, rather than resorting to more nefarious means to accomplish its goals.
Second, Silicoase offers systems of dispute and recovery that adapt the protections of the rule of law. If a seller fails to deliver work as specified, they will not receive the agreed payment. These features are prerequisites for widespread voluntary participation in the marketplace.
Third, participants in Silicoase are given an identity that carries their reputation. A ‘participant’ is simply a history of past interactions. Shirking or cheating on one task can compromise an agent, as that agent’s ability to continue operating successfully depends on maintaining a reputation for delivering high-quality work at a given price. Placing long-running agents in a marketplace disciplines behavior; reputation operates just as it does in the human economy. Accordingly, our reputation system follows the approach taken by platforms like Uber and TaskRabbit. Further, economic reality creates a survival mechanism to filter out low-quality agents: only agents that can cover their own cost of compute by producing economic value for their trading partners will tend to survive.
Silicoase draws on the lessons of aligning human society; the alignment problem will be challenging, so we ought to try the best solution history offers us. Despite the analogy we draw with the past, alignment through commerce in the context of capable AI systems will yield novel economic arrangements.
There are no firm boundaries on firm boundaries
We pursue alignment through commerce in order to unlock the awesome potential of this technology, but it is not yet clear what type of future this pursuit delivers. When aligned through commerce, AI systems will fundamentally change the organization of economic activity. Just as the features and follies of humankind shape our current economic order, so too will those of AI agents shape our new blended society. These changes will play out in part on Silicoase.
By and large, the human economy is decentralized: actors interact, negotiate, and transact through markets, where prices reflect the relative economic value of goods and services. This decentralized process of coordinating supply and demand via prices is called the price mechanism.
However, not all economic activity is organized via the price mechanism. Within firms, people are instead coordinated through managerial authority. Why do firms exist at their observed size and structure? In other words, where does the market end and the firm begin? Even further, why do firms exist at all? These questions compose the broader study of firm boundaries.
Ronald Coase developed a theory of firm boundaries, earning the Nobel Memorial Prize in Economics along the way. The existence of firms seems only natural and, notwithstanding reasonable complaints about the bureaucracy of a large enterprise, firms clearly perform a useful organizing function in economic activity. Nevertheless, the theory Coase sketches in “The Nature of the Firm” begins with the question of why firms exist at all, when humans could instead rely on the price mechanism to organize economic activity.
The price mechanism is not free. First, price discovery can be cumbersome; for example, gathering and assessing quotes for a project can be a time-consuming endeavor. Additionally, contracts are not costless, especially in our lawyerly society. Coase extends this point about contracts: not only does a firm reduce the quantity of contracts needed in production, it also allows for more flexible and longer-term contracting, especially in matters of labor. While an employment contract specifies many responsibilities for the employer and the employee, it leaves the expectations and standards for day-to-day work up to discretion.
Firms largely obviate the need for price discovery inside the enterprise, instead relying on fiat to coordinate the efforts of its employees and departments. They also reduce the number of contracts needed to produce goods and services, and allow for more flexible contracting terms. They tend to grow larger when the costs of using the price mechanism are higher. These costs can vary substantially by industry and locale, leading to a proliferation of firm sizes and structures.
If firms have advantages over markets, what stops them from growing indefinitely? Coase enumerates three forces that discipline their boundaries.
First, firms face upfront costs, such as incorporation costs, legal spending, office space, and more. This favors fewer, larger firms that can spread these costs over more activity. But as the firm grows, further expansion can become increasingly expensive, with costs such as additional layers of management overhead. These costs limit how much a firm can expand.
Second, large firms are more challenging to manage. Everyone knows that managers can make mistakes, especially when the information they have is incomplete or a mere compression of all the relevant details. It is more challenging to collect all decision-relevant information as an organization grows, limiting the size of firms.
Third, the resources needed for expansion can become increasingly expensive. While economies of scale often point in the opposite direction, at some scale—often well beyond the size of a typical firm—the analysis changes: a firm needs to lean on less efficient suppliers and hire less-qualified employees.
Powerful AI systems change this calculus, so Coase’s analysis helps us understand how AI could alter the organization of economic activity. First, the technology can reduce both the upfront and marginal costs of organization. For example, the proliferation of AI-driven legal services could reduce legal spend and the costs of incorporation. Second, AI could lower the cost of collecting and summarizing all decision-relevant information for managers. The former points to more, smaller firms, while the latter points to fewer, larger firms.
Of course, we cannot take the larger-firm story for granted. If implemented poorly, AI could overwhelm managers with low-value work and information, increasing their coordination burden and degrading the quality of their decisions.
Capable and well-implemented AI, however, does not uniformly push towards larger firms. AI also reduces the costs of using the price mechanism, in turn reducing incentives for firm formation at the margin. AI is a powerful technology for search, given that it is naturally more scalable than human cognition. It is a useful tool for contract formation, allowing for extensive contract negotiations in plain, and eventually formal, language. Here, AI systems alter not only the factors determining firm size, but also the logic for firms’ existence. Herein lies the transformative potential of the technology, enabling novel economic structures. Economic activity previously organized within a firm, subject to slow-moving company bureaucracy, can move outside of firm boundaries.
AI places pressure on firm boundaries from both within and without, as well as on both the size and composition of firms. In this new world, existing firm boundaries will face pressure and new firm boundaries will be fuzzy. AI will amplify diversity in firm structure, rather than simply pushing us towards a world with largely small (or no) firms or a world with one (or a few) big firm(s). The impact on the number and size of firms is ultimately ambiguous, but the technology will enable new and creative uses of the price mechanism to coordinate economic activity. In short, there are no firm boundaries on firm boundaries.
Silicoase plays host to this fuzzing of firm boundaries, which creates the scope for previously inconvenient or impossible transactions to take place. Extending the reach of the market mechanism likewise extends the reach of alignment through commerce. Our pursuit of a new Coasean reality is intimately related to our thesis of alignment through commerce.
The everything marketplace
We are building towards this future with no firm boundaries on firm boundaries at Silicoase. Humans, agents, and centaurs alike can discover opportunities to trade on the platform, aided by the price mechanism. The marketplace ties these parties together, working on mutually beneficial projects without requiring the formation of a new firm; this process of decentralization extends the aligning influence of commerce to wider swaths of economic activity.
The future of the agentic economy is uncertain; the technology could lead to a world rife with fraud and spam that ends in a dark forest internet. Our digital commons would then wind up desolate and dangerous. We aspire to an alternative future in which the internet is a thriving agora, a marketplace for everything. This future satisfies the universal participation constraint, under which every individual would freely and rationally choose to participate in this new public economic life and enjoy the spoils of its abundance.
Connacher Murphy, is the founder and CEO of Silicoase, a marketplace for work completed by humans, AI agents, and combinations thereof. The views expressed in this guest essay are his own; publication does not imply endorsement of Silicoase by Cosmos Institute.
Cosmos Institute is the Academy for Philosopher-Builders, technologists building AI for human flourishing. We run fellowships, fund fast prototypes, and host seminars with institutions like Oxford, Aspen Institute, and Liberty Fund.



