In 2016 Joel Monegro wrote the influential and controversial Fat Protocols thesis
The previous generation of shared protocols (TCP/IP, HTTP, SMTP, etc.) produced immeasurable amounts of value, but most of it got captured and re-aggregated on top at the applications layer, largely in the form of data (think Google, Facebook and so on). The Internet stack, in terms of how value is distributed, is composed of “thin” protocols and “fat” applications.
This relationship between protocols and applications is reversed in the blockchain application stack. Value concentrates at the shared protocol layer and only a fraction of that value is distributed along at the applications layer.
This argument appears entirely based on a 2016 empirical observations
The Bitcoin network has a $10B market cap yet the largest companies built on top are worth a few hundred million at best, and most are probably overvalued by “business fundamentals” standards.
For clarity, we say a protocol is a process that governs exchange or activation of energy, resources, or data, while an app is a controller and viewer of data, energy or space with a direct social or economic relationship with a customer.
Fat Protocols are crumbling
This narrative is under pressure today. Last week we saw tweets from Austin King
And @kaledora
And even empirically, we’re starting to see apps make more revenue than underlying infrastructure.
Ok the Fat Protocol thesis is in question. Crosshatch is an integrations and security company using AI.
What does this have to do with AI?
AI is a protocol
AI and (values of!) web3 are on a collision course.
AI can be thought of as a candidate protocol for fetching intelligence. Riffing more, AI weights might be seen as a sort of ledger of things that have happened in the world and how they relate.
AI appears poised to be a centralizing force, while interests in privacy, control and interoperability hold power by their connection to the rights and interests of the individual, a decentralizing force.
AI can remix any content and potent AI resources are available everywhere (cloud, browser, and locally) so there’s a question of what gets remixed, by whom, and why.
Can the individual remix content across apps – is the app over? Or can apps remix an individual’s context? Or does all candidate functionality and context accrue to a large AI lab?
This question is fundamentally about where value will accrue in the age of AI. And the ideas we’ve long explored around users “logging in with their context” are ideas previously of crypto e.g., in the follow-on essay to Fat Protocols, Thin Applications
As a crypto user, you bring your own data. Nobody has monopoly control. When you log into a crypto app by connecting your wallet, you’re sharing the “keys” it needs to find your information across the relevant networks. You can share them with any app, so your data comes with you as you move from interface to interface
This sounds just like what AI people are saying today, e.g., A16Z’s Justine Moore.
Value in apps is private information
The thing is, and as many started discussing last week, we believe the Fat Protocol and Thin Applications thesis is wrong.
We introduce a new argument: the Fat Protocol thesis for AI is wrong because of inherent asymmetry at the interface between application and protocol.
Continuing from before, take a protocol to be software that accrues, permissions, and activates data. While protocols might manage and activate data, protocols are not privy to nor have rights to participate in the value generated by apps that use protocols.
That is, there is an information asymmetry between apps and protocols. Apps provide an interface that creates value with users. That interface is owned by the app, and has no obligation to (or incentive!) describe how a protocol drove value to the app or its users.
This situation is the Inverse Market For Lemons.
In the classical Market For Lemons – which won George Akerlof a Nobel Prize – buyers know less about the value of a good being sold than sellers. This leads to a market failure and a race to the bottom where high-quality goods leave the market because such sellers can’t sell their products for their true value.
In the inverse case, buyers are better informed than sellers. That is, apps know the value the app delivers to their users via the protocol, which (at least in web2 markets) is private information unknown to the protocol. Apps are disincentivized to share this value, as this could lead to price discrimination (and loss of their hard-won surplus!).
In this case, this leads to a race to the top with lower quality protocols falling out of the market, as protocols have to take what they can get, while higher quality or scaled protocols rise.
You can see this in analysis around AI-enabled “services as a software” businesses. “Services as a software” customers want to save time and money. AI-enabled B2B SaaS don’t know the value of their services to their customers – sharing value could give SaaS ammunition to raise prices! – and if these customers can save more by killing the AI-enabled SaaS arb by building themselves (e.g., as Klarna is), they probably will.
In June, we bet that value will accrue to “Fat” Thick Apps – those with large inventories of qualities expensive to discern. Our argument there rests on the impracticality of disintermediating Thick Apps, which would involve expensive, slow, unreliable and likely unconsented scraping and AI processing.
We believe Monegro was mistaken in both of his essays: in the age of AI (even with portable interoperable context!) there will be
- Thick “Fat” Apps
- Thin Protocols
[Thin Protocols means, among other things, that “intelligence too cheap to meter” actually means that large lab margins go to the cost of compute by the same information asymmetry argument, unless of course, they vertically integrate and bypass this protocol/app interface.]
Grow the pie
We’re building Crosshatch to become a Thin Protocol. We follow Fred Wilson: Be Generous is a good strategy (if you have sensible cost structure!) :
How can giving something away or letting others take value from you be good business?
It is all about zero sum thinking. If you think that the size of the pie is fixed, then you need to grab as much of it as you can. But if you are making a pie that can grow and grow and grow, you just take a small slice and let everyone else eat.
Accepting this asymmetry is also just practical. E.g., following Gokul Rajaram trying to set prices on metrics you don’t control is not recommended.
Crosshatch enables the world Monegro envisioned: users bring their user-agent
- Secure AI resources
- Permissioned data
- Activated for a particular purpose
to turn on hyper-personalization in their favorite apps in a privacy preserving way. Crosshatch enables richer experiences at a lower cost for its affordance of all linked user context scaled by assurances of user control that only a protocol can offer. Following Monegro
Nobody has monopoly control. When you log into a crypto app by connecting your wallet, you’re sharing the “keys” it needs to find your information across the relevant networks. You can share them with any app, so your data comes with you as you move from interface to interface. And you keep control of the “private key” (basically a password) needed to operate on it.
The Fat Protocols thesis sizes "Work + Data" as constituting the largest part of value capture
Particularly in the age of AI, the "work" of these "cryptoservices" could conceivably collapse to a single AI API call on models whose price has fallen 240x in the last 18 months. Combined with the information asymmetry at the interface between app and protocol, even under a "bring your own data" (that we're building!), we expect the value to look more like
In the age of AI, value accrues to Thick Apps and Thin Protocols.
Thick apps accrue value for their earned relationship with the customer. Thin (“skinny fat”) protocols accrue value by scale, network effects and trust.
The reason why apps are thick is not because they control all the context – when users can bring their own data to apps, users have the data gravity – but because of this information asymmetry: the value apps deliver to end-users is earned private information owned by the app not the protocol. Apps keep margins because they build trusted brands (often themselves with their own networks like eBay, Zillow, Spotify or Airbnb) that delight and save users time and money. Revenues from retail media networks could see a shift.
But normatively, this all feels right to us. Applications that directly create value should accrue value.
At Crosshatch, we’d be thrilled to help make Apps even better for lower costs.
