Cores and Seams
Crypto, AI, and the Firm
“Crypto is decentralizing, AI is centralizing. Or, if you want to frame it more ideologically, crypto is libertarian and AI is communist.”
- Peter Thiel
Enter Ronald Coase
In 1932, a 21 year old kid with commie leanings took tours of Ford and GM plants on a travelling scholarship to America, and was struck with a question nobody had apparently thought to ask before: if the miraculous price mechanism of the market allocates resources better than central planners, then why does so much of the market consist of these little command economies inside of large companies? And, on the flip side, if hierarchy works so well, why isn’t the whole economy one giant firm?
The conclusion that he came to was that while the market is efficient, actually using the market isn’t free. Every transaction with a contractor outside of the company has costs that aren’t included in the price tag: figuring out who to hire in the first place, negotiating with them, monitoring performance, and enforcing agreements when things go wrong. These external transaction costs add up.
Firms exist because it’s usually cheaper to hire someone on an ongoing basis and just order them around instead. But the costs involved in maintaining that hierarchy, internal coordination costs, add up too.
Coase figured out that the size of the firm will grow until the internal coordination costs reach an equilibrium with the external transaction costs. If adding just one more person to your bloated company costs more than outsourcing it to the market, then it only makes sense to stop hiring.
From there it follows that if you reduce the cost of interacting with the market, the average size of firms will shrink, but if you reduce the cost of maintaining a bureaucracy, the average size of firms will grow.
Soldering Jell-O
A little over a week ago IC3 — a who’s who of researchers on the academic side of crypto, studded with high-level insiders — released a report called Crypto x AI, AI x Crypto: A Survey. It was pointedly restrained: staking the claim that industry hype has gotten far ahead of any actual tech, but still managing to outline a plausible path for AI and crypto to act as middleware for one another.
“Combining the two naively can be like soldering Jell-O. Combined well, though, crypto can channel AI’s fluid power into secure and reliable systems.”
- Ari Juels
A few days later Dario Amodei released an essay called Policy on the AI Exponential, which casually drops cryptocurrency as an example of a “much more mundane technology” that AI has historically been mistaken for: a drive-by pot shot far, far away from the equal billing of the IC3 paper.
It’s true that on their own these are not symmetrically important: crypto is mostly an efficiency story, whereas AI can generate new value that compounds on itself over time. Crypto needs AI, but its unclear if AI needs crypto.
It basically all boils down to one question: is the future full of billions of independent agents transacting over trust boundaries, or will everything consolidate into one big happy borg?
One pill makes you larger…
Crypto is a set of technologies that allow people who don’t trust each other to coordinate. Bitcoin itself can be described as a solution to the Byzantine generals problem: How can a group of generals coordinate an attack when a subset of them are traitors? Ethereum expands this out to general purpose smart contracts that immediately execute once their conditions are met. They work with the same logic as a vending machine: the money goes in, the Twix bar plops out. No quibbling over the results, no takesies backsies.
The whole value proposition of crypto is aimed squarely at reducing Coase’s external transaction costs: it unidirectionally acts as a force towards shrinking the size of businesses, allowing them to unbundle themselves into a kaleidoscope of independent microservices.
(What did you think decentralization meant? Vibes? Essays?)
AI on the other hand cuts both ways: it lowers external transaction costs by allowing solo operators to act as full fledged procurement departments, but it also attacks the internal coordination costs of bureaucracy, acting as an infinitely scalable middle manager and making it cheaper and more feasible to grow a large organization.
When the costs that define if you’re big or small both plummet at the same time, what actually happens?
Complicating things further, AI allows firms to slash their headcount while maintaining or growing their operational complexity, enabling the possibility of a “large firm” that consists of one (or even zero) employees.
The relative size of fully autonomous agents also follows the same Coasean logic. Do they choose to grow themselves by spawning a subagent to complete a task, or do they stay lean and punt it out to an external service?
Drawing the Lines
To reduce confusion let’s switch to the following terminology:
The inside of any command structure is a Core: a firm or a platform, an agent and the subagents it spawns — anywhere where orders are passed down and resources are allocated.
The boundary between cores is a Seam: the join between two firms or jurisdictions — neither side can command the other, so everything has to be negotiated.
Over the next few years we will begin to see if these technologies are manufacturing a fragmented, maximum-seams economy of micro-cores (crypto’s ideal habitat), or internalizing commerce into a few enormous platforms that have a limited need for neutral rails.
Max Seams World: Everyone is a Firm of One
In this world all cores have shrunk down toward the individual, with an agent independently representing each person’s interests. This does not mean that they’re equal: some have access to vast pools of resources and capital, like Elon Musk on steroids; others have only the power of their vote, and whatever resources from governmental redistribution that gets them1. But even when the most powerful person in the world wants something done, no durable organization comes together: instead a standing army of external agents and contractors forms around their intent.
A version of this world has existed before. The modern industrial corporation that we’re familiar with is basically a 200 year old technology that was invented to coordinate masses of human labor around expensive machines. Before that, one of the high-water marks of private commercial power was the merchant prince: a Medici sat in Florence and commanded a far-flung network of contracted agents (human agents in this case) across the continent by letter.
Max Core World: The Monocompany
Inside the monocompany you get told what to do, and things are allocated for you in the way office supplies are allocated. Maybe you can still nominally “buy” things, but it would be like living in a company town and being paid in scrip: both customer and employee. The monocompany sets what you are paid, and sets what everything costs. No negotiation. Maybe it’s not so bad: a benevolent AI overlord deciding what’s best for you. Fully automated luxury communism, if material scarcity goes away.
But seams don’t disappear entirely. Even an extremely intelligent entity with perfect knowledge of every price and preference in all of history couldn’t run the economy completely centrally2. But that doesn’t mean that cores can’t become very, very large. The East India Company at one point governed a fifth of humanity, maintaining an army that rivalled Britain’s and minting their own currency. The market retreated to the edge of the empire.
Por qué no los dos?
One historical transformation we can look to is the emergence of the internet. The internet slashed the costs of interacting with the market and maintaining a bureaucracy at the same time, and it created a barbell shape: superstar firms became more concentrated than ever, and there was an explosion of viable one-person operations, with the middle hollowing out.
The likeliest future is both extremes at once: unimaginably large cores trading over unfathomable seams. Wherever verifying an outsider’s output is cheap, the seams multiply and the work scatters to whoever does it best. But when trust is hard to establish, it gets pulled into a core.
This is the world we’re already seeing emerge in the AI era: people without technical backgrounds are vibecoding apps that clear 300K ARR with no employees, while SpaceX pulls off the largest IPO in history, passing a 2 trillion dollar valuation (with Anthropic and OpenAI soon to follow).
A world where value concentrates at the seams is one where crypto is worth talking about at all; whether it actually wins against centralized alternatives is another open question3. But, if crypto forms the rails on which a massive new agent economy sits… that doesn’t sound too mundane to me.
My motivation for writing this post, much more straitlaced than what you’re used to from me, was to serve as an audition for this Narratives & Communications Lead role at Basis (an “AI for accounting” company). If any of my dear readers know people who work there, please put us in touch.
(Accounting, interestingly, wins no matter which of these futures plays out)
There’s a brutal cyberpunk version of this where they don’t have even that — the true libertarian dog eat dog world — but Coasean logic doesn’t actually touch government very much: the government is a monopoly that makes its money by force, so it’s fully allowed to be as inefficient as it wants.
Something something Mises and Hayek
Personally I think it does, for basically all the usual reasons crypto shills give. Cores can’t credibly act as neutral intermediaries at a seam — the moment a firm owns the rails between two parties it has every incentive to act in its own interests: tilting things in favour of its own services, or harvesting valuable contextual data that leeches value from the transacting parties. Neutral ungameable protocols mitigate the first issue, and zero-knowledge proofs can partially mitigate the second (the hard version of the objection is that context revelation is itself the new transaction cost, which is a strong argument that the max-seams world can’t happen in its extreme form).
Thanks for flagging this report!!!