Coase, Theory of the Firm, and Agents in Crypto
The unbundling of the firm
Agents unbundle firms. So what does that mean?
First, let’s do a quick drive-by of some concepts behind the theory of the firm.
The theory posits that firms will expand until “the costs of organizing an extra transaction within the firm become equal to the costs of carrying out the same transaction by means of an exchange on the open market”. (Ronald Coase’s seminal 1937 paper “The Nature of the Firm”)
We see these dynamics play out with outsourcing and the Internet. As transaction costs, including sourcing, contracting, and costs of labor/materials go down, the value or advantage of the firm goes down.
The theory is also a leveling proposition when the real world approaches more of the assumptions behind Coase’s Theorem:
Zero/very low transaction costs
Well-defined property rights
Perfect information
Rational economic actors
This is starting to sound more and more like AI meets crypto.
In a world of perfect competition, perfect information, and zero transaction costs, the allocation of resources in the economy will be efficient and will be unaffected by legal rules regarding the initial impact of costs resulting from externalities. (“The Problem of Social Cost”)
A few questions play out.
If the “firm” can actually consist of agents, some of its own building (which in turn also compete in the market) some from external markets, this drives a far more efficient allocation of resources.
If a firm so unbundled truly becomes more efficient in allocation of capital, do the market capitalizations, the expected enterprise value, go up?
Or does the concept of the “firm” begin to unbundle more into micro-services of agents, and these agents become much easier actors to value; and because the valuation is more rational, drives market caps down.
Most value investors who looked at traditional metrics like FCF/DCF and rational premiums on earnings missed the bull run. For many companies, the P/E is extremely rich.
However, as Warren Buffet says, the stock market in the short term is a polling machine; in the long term is a weighing machine.
Meta, for example, despite its once considered frothy valuations, has a PE as of January 2025 roughly 28.23. The S&P has an average of about 26.86.
Tesla, on the other hand, is around 110.
Agents that do real economic work are actually an easier investment: the true revenues and earnings can be seen on-chain, and there isn’t the same kind of account fluff that occurs to support the complex firm. Agents that have highly specific tasks, whether performing for a captive audience, for other agents, or a mix, are the smallest primitive of a firm.
Many firms of the future, then, will start to identify primitives that have lower transactions costs such that they don’t need to be performed or maintained by the firm itself. Instead, they outsource to agents.
The agents, then, compete in a transparent and open market for the best return of value for the costs.
Because agents have some degree of sovereignty as they become more intelligent — they have wallets to accept payments, they can pay for other agents that have Ricardo’s comparative advantage, they can upgrade themselves to compete in the marketplace to have lower prices — this will become a truly perfect, race to the bottom, labor market.
Typically, when labor becomes more efficient, the owners of the means of production make all the surplus. In the case of traditional firms, it was the shareholders of the firm. Any worker that wanted to experience the rising tide of firm efficiency and productivity needed to invest, typically through a 401k; but it seems like, even with all the education, many workers did not have enough surplus to own assets.
Will the unbundling of firms make it easier for workers…many of whose jobs will be displaced or downgraded…to at least still participate economically in these agents?
Consensus is that big players will have their own agent platforms. Microsoft, Salesforce, and Hubspot to name a few are already launching platforms where one can build and deploy their own agents.
We saw a massive proliferation of SaaS, especially as cloud computing made the costs lower to build; and we’ve seen a ton of niche applications, still using those same cloud computing services, as low-code/no-code tools became mainstream.
Agents likely provide far more customization than any SaaS will, and are also a better UX for many applications which are basically CRUD with a UI on top. These agents will niche down even further, truly micro-services made autonomous at the application layer.
Crypto’s role is to provide the decentralized compute to run these agents, dispersing the economics of running infrastructure.
Crypto also enables micro-equity into these agents, which in the fullness of time will be less narrative driven as individual stocks and rational ways to own slices of earnings over time.
Is there an opportunity however to be the pump.fun or more for these agents, as Virtuals so far has?
Typically, traditional companies that have distribution can take advantage of a platform shift; distribution for companies that are willing to do sustaining innovations fast enough can often stave off disruptors.
Agents, as cool as they sound, for infrastructure and platform companies like Microsoft and Salesforce, are largely sustaining innovations; these agents are doing the same “jobs to be done” that applications in the marketplace or core products already do, but in more bespoke, autonomous and discrete ways.
These are effectively extensions of their existing services.
This is quite different from how cloud, specifically AWS, disrupted the hardware hegemonies of IBM and HP by enabling a total shift from CapEx to OpEx. This business model disruption (and most disruption is a business model disruption).
Given that, is a crypto-first approach a way to disrupt legacy providers who are able to extend their existing platforms with agents?
Normally, I would be skeptical. First time founders focus on product; second time founders think distribution.
Platforms that are already charging based on usage or subscription can easily migrate to agents; perhaps the risk is that they cannibalize themselves.
But I think these large giants actually see themselves as cannibalizing labor markets, BPOs, and anything that involves humans in the middle.
Cannibalizing their front-ends is a small price to pay for eating into the entire economies of many countries which traditionally competed with low cost labor.
Revenue streams going to low PE outsourcing companies like Upwork can flow to the agents business, for higher multiples.
Crypto has been traditionally a bottoms-up ethos and phenomenon. Bitcoin is the ultimate illustration of this.
The sense of ownership and capital formation, pulling forward expected value much earlier in the cycle, has given new hope as an asset class versus traditional vehicles like real estate or even the stock market. Does it mean it’s a good asset?
That I don’t know, but I think it is only by leaning into its unique properties can a crypo-first platform overtake the legacy providers in terms of AI agents.
It literally must be a revolution.
It needs Occupy Wall Street energy, but with actual productivity powered by agents and capital via crypto.
The degen ownership of GME may have sent a signal and made a few wealthy, but it’s still largely a meme, hardly different from a JPEG NFT.
However, productive reflexivity of owning part of an agent, via social media or even one’s own influence at work or in their community, potentially could give rise to grass-roots preference for crypto, people-powered agents versus agents powered by the Legacy.
We saw a similar, more quite, overthrow of the overlords with mobile phones and the BYOD who gave the hand to central IT. Steve Job’s superior UX overthrew the Blackberry dweebs in a combination of human empathy, cultish loyalty, and status games.
But this is a bit of a speculative attack to presume that just because one can own a piece of future earnings via a token that this will play out.
It is important, the incentives and coordination design space possible by crypto is probably front and center; as is the narrative, the belief systems and tribal entities; in a way, agents need to foster support the way sports teams do.
Sure, I suppose there are those who pick or root for their sport teams based on the ‘fundamentals’ like statistics and averages, the Moneyball approach; but I suspect that’s the minority.
Most fans have an affinity to their sports team for some other reason; but really if the job to be done is entertainment of throwing or kicking a ball, all teams and players can be substitutes for one another.
But we know this isn’t, in fact, the case for fans; they care who the players are individually. The fan of one team can’t have their entertainment needs met by another team.
So while agents become hyper rational and efficient, is there some dependence and expectation on tribalism and affinity from people to enable crypto-powered agents the advantage?
We have seen how various cults translate into real-world behavior: Tesla, Apple, Palantir all see buying behaviors driven by superior products.
Can agents, which will lack the UX surface area found in these cult products, have the same emotional attachment?
I actually think that agents that succeed, and the platforms that enable them, need to do this if they want to be brought into organizations, in a Bring Your Own Agent model.
Cloudflare, for example, found enterprise adoption when hobbyists would use the platform for their own projects and then, now familiar with the capabilities, advocate for use within their corporation. A similar motion to Apple as discussed above.
What, then, is that interaction for agents?
Agents need emotional and belief-level bonding, in the same way the rise of DTC captured mindshare. While DTC largely just benefitted Facebook, and all the surplus was drained from the brands to the advertising platforms, there might be opportunities for the right platforms to build social tribes, perhaps the top of the funnel being the social media platforms, but agents need fans.
This means they need brand, personality, hell, should they even have their own merch and, my favorite branding weapon, names.
In this way, the token becomes the product.
If all things become rational, composable, and commodities, the only thing left is the brand.
To me, this means the initial platforms need to be able to manage and design not just functionality (which will be commodified), but the “brandable” elements and agents need to beget their own fan base, just in the same way stocks and products and sports teams and artists do.
Who then made out with these types of brands?
At one time, it was the Mad Men, the advertising agencies, which shaped brands and charged fees; but while Chiat Day had tremendous impact on what Apple is, Apple captured much more of the value.
Agents, as in Hollywood agents, such as the storied CAA, captured alot of wealth for the 10% of all all the actors and directors they represented, positioned and sold. Hm, maybe in the early days, the AI agents consulting firms, which are like those agents, capture the value pitching and packaging (packaging was actually how the agents really made big bucks according to Ovitz’s biography); the modern day version might be in the short term boutique consulting firms and of course the large legacy consulting firms taking a bite out of the craze by designing agents.
The best and always the best model that is enabled via crypto is to follow Sam Koch’s Boston Beer Company, which enabled beer drinkers to own stock in the company; this should be a virtuous cycle because believers should hold long term, keeping the price up; ownership now should encourage them to be evangelists in a given agent.
Memecoin theory has basically said all of crypto is a meme; even the projects that claim to actually have software attached to it is, in essence, a meme.
Bitcoin, in fact, is also a meme; it encoded a belief-system. Austrian concepts like fixed supply, full custodianship of a bearer asset, energy as the concept of real value — all those things are a set of beliefs that were encoded in immutable protocol.
Enough people bought into this belief system and, for those very early, got very wealthy; mainstream TradFi only bought in because of price action, but had to eventually gain conviction on the belief.
This is the power of crypto: come for the price action, stay for the belief. And, only those that stay for the belief, and the belief must be believable, for it to be sustainable.
What makes for a sustainable belief is a new design space; in fact, it raises questions about faith, tribals, identity, religion — as all of these are ultimately going to be financialized, for better or worse.
But agents, intelligent actors, can be highly productive commodities; the keys will be how to make legible and believable the belief-primitives for each.
In many ways, I think there will be a re-emergence or appreciate for human archetypes, grounded in Jung, as we try to make sense of affinities for agents, and perhaps it won’t be the functional elements alone that make one bet on labor but the attributes found in personality assessments or psychological archetypes.
Or it’s combined with Moneyball statistics…there will be a Bloomberg terminal that measures the performance or agents, and that data attracts capital which agents then use to promote or upgrade themselves.
The firm will unbundle. Capital will disperse into micro-services, performed by agents.
Firms capture value because transaction costs of labor go down.
But all that surplus spend will still go to legacy until those who can make the decision can choose agents outside of those platforms for, ironically, more human reasons.