I. 

They are what might be termed self-defeating questions, and they haunt those of us foolish enough to dare build consumer apps:

1. Do we actually think value accrues to apps when all evidence points to protocols instead? Can apps actually, like, make money in a world where protocols like Ethereum easily generate fees, justify a token at the base layer of their security, and capture value from all apps built on top?
   

2. Can new apps even build sustainable users when apps, as a whole, look increasingly like a historical phenomenon of 2008-2012? After all, why haven’t there been any widely, durably adopted new apps in nearly a decade or, if we exclude TikTok, over a decade?

So dire is the obvious answer to both these questions—“Dumbass, no!”—that it prevents any other from being true. Once our wise investor overlords have concluded, understandably, that apps can’t accrue value, then investment in apps inexorably dwindles—and there remains very little hope for disproving the anti-app thesis at all.

Let’s be fair to the other side: a certain kind of contrarian might ask if such negativity is a pretty good sign that the app cycle has bottomed. Perhaps one app thesis has run its course, and another will rise? That presumes, of course, that the app cycle is exactly that, a cycle, and not simply a blip of the 2010s’ Decade of Web2.

But while this contrarian might dutifully marshal evidence for a new app thesis, they likely can only do so by making a much better case for protocols. Yes, our contrarian can cite an example like Friend.Tech, which saw about $1,000,000 in sales at its peaks; but today, that number is closer to $16,000, a decline of nearly 99% in two weeks. Likewise, our contrarian might argue that Friend.Tech takes far greater fees than the protocol it’s built on, Base; but if the app truly is a harbinger for more apps to come, then inevitably Base will make more in aggregate from its constellation of apps than any app makes individually.

The argument seems clear enough: even the strongest case for apps is, at heart, a case for protocols.

It is not a new argument. Joel Monegro’s “Fat Protocol Thesis,” written in the heyday of the app era in 2016, noted that “the market cap of the protocol always grows faster than the combined value of the applications built on top, since the success of the application layer drives further speculation at the protocol layer.” Many have tried to counter the Fat Protocol Thesis—myself included—but by 2023, it feels like all but doctrine.

In fact, the Fat Protocol Thesis gives us a tempting answer to both our questions: the reason there have been no great new apps in the past decade is precisely because value has started glomming onto protocols instead. 

So rather than attempt to quixotically prove that apps will somehow become better investments and revenue-generators than protocols, I’d like to argue that the real answer to our two key questions—Can today’s apps acquire value? Can today’s apps even acquire sustainable users?—is actually to question their own terms. Even to ask if value accrues to apps or infra is to assume that these are mutually exclusive categories. And to ask why we haven’t had any major new apps is to ignore the fact that we have—but that they might not look like apps in any traditional sense.

And why haven’t they?

Because the next wave of great social apps will all look like protocols—or, if you prefer, proto-apps.

II.

The simplest answer for why there’ve been no widespread, lasting apps since TikTok—just a smattering of novelty BeReals here and there, ebbing on seasonal zeitgeists—is what we might call The Social App Thesis, which has defined the fate of the major apps, the social apps, for the past decade. In short, there are a limited number of social categories, and the winners in each have attained distributional network effects so strong as to be unassailable to competitors.

Our app store’s dance card is filled, in other words, as we’ve exhausted the five major categories of online expression:

• long-form video (YouTube)

• short-form video (TikTok)

• long-form text (Reddit)

• short-form text (Twitter)

• image (Instagram).

In the early days of social apps, you could create a successful incumbent by hybridizing these categories as well through a Tumblr or Facebook that combined all the categories of image, short-form text, long-form text, and video. But this was a historical opportunity open only for a few years before social was unbundled.

Indeed, within the past five years, the only hope for kickstarting a successful new app has been to trailblaze a new category completely: live audio (Clubhouse) or live photo (BeReal), for example. But live experiences, for all their novelty, preclude continuous interactivity that’s been the basis of successful social. They’ve defied while defining what turns out to be the golden rule of social apps:

You should be able to interact with as much relevant content as possible whenever you tune in.

There just aren’t any categories left, in other words, and there’s no way to compete within the ones that now dominate our daily lives. You can create a Twitter competitor that’s 10x better than Twitter, but it will still struggle against Twitter’s distributional network effects: ultimately, Twitter is still the place where posters get the greatest reach to readers, and readers get the greatest access to posters. They simply cannot get that from another app without spending years to re-cultivate their social graphs. Every second spent doing so is also a second away from Twitter, where readers and writers keep the other encamped for so many hours a day that they consume each other’s attention to look anywhere else.

Distributional network effects, or lack thereof, also help explain why there are two major app categories where entrants can claim share alongside incumbents without ever quite disrupting them: messaging and dating.

Arguably, messaging and dating apps follow the same model. Because they’re not fluid public markets between creators and their audience, their network effects are weak. To put that simply, creators are not using messaging and dating apps to reach the widest audience of strangers, so they don’t lose anything by switching from one app to another to contact different groups. But note that here, too, the major players are a decade old: Signal was founded in 2014, Telegram in 2013, Discord 2012, Snapchat 2011, WhatsApp 2009; Bumble 2014, Tinder 2012, Hinge 2011, etc.

Even when the network effects are weak, the category remains too full for newbies. As Chris Paik put it in 2021, “The reason why we have not recently seen additional mobile-first social companies emerge is because we are well into the half-life decay of the smartphone.”

And here we arrive at the moral of our little tale, that according to The Social App Thesis, we’ve simply reached saturation.

Now.

It would be hard to say The Social App Thesis is wrong, exactly—within the terms of this tale, it seems fairly irrefutable. But it’s telling that The Social App Thesis neglects the fact that a major new app has ascended for the first time in nearly a decade. Yet so game-changing is this app that you might not even notice it’s an app at all. I’m referring, of course, to Chat-GPT.

Indeed, it is something of a stretch to say that Chat-GPT is any kind of social app. You can make the case, of course. You can argue that Chat-GPT, like every other major social app, is just another way to maintain an endless dialogue in a specific medium over all the things you’re interested in. The difference is that instead of posting content for the world to see, you post only for yourself, for a good reason: the being that you’re chatting with is, in fact, nothing less than a higher version of you, an aspirational you, a you who has all the answers.

But even if we refrain from categorizing Chat-GPT as a social app, well, then, that tells us something too—that we just took for granted the fact that major apps would be social apps, that the rules of user-generated-content and messaging apps would define the rules of all major apps. 

So bear with me here. Because I’d like to argue that what makes Chat-GPT distinctive as a new type of social app has nothing to do with whether or not we classify it as social at all. You can say it’s a new breed of social or not, and it doesn’t really matter to the point I want to make here, which is that we need a new model to define the successful apps of the future. So let’s go bigger here:

Chat-GPT is the first major app in nearly a decade because it’s the first example of a whole new category: proto-apps.

And to understand proto-apps, we need to understand what makes Chat-GPT an absolute exception to the rules of apps that have defined the previous decade.

III.

We might say that at its algorithmic heart, Chat-GPT breaks the traditional paradigm of successful apps in three ways:

1. It was launched as a Web App. While Chat-GPT recently released its inevitable mobile app, it launched—unlike nearly every other major app of the past 15 years—as a web app. And for good reason: there wasn’t really a need for a mobile app and there was very good reason not to have one at first. Let’s start with the obvious: design-wise, Chat-GPT requires little more than web1 UX, nothing more than a bar to ask a question and a window to answer it. With Chat-GPT, navbars, search, and feeds, those building blocks of web2 social, have gone magically extinct. And that’s why we need a Chat-GPT app about as much as we need a Google app; it makes far more sense just to embed these within browsers.
   
   But more practically, Chat-GPT is a web app because it would be a bad idea to launch a mobile app first. As much as common wisdom dictates that major apps should be mobile apps, downloading an app is a major point of friction increasingly afforded only to the most popular services. More to the point, mobile apps have the advantage of push-notifications, data collection, and interoperability with device assistants, but they have the cost of a 30% app store commission. I argued in “Apple is a State” that Apple’s vertical command of app store, operating system, and phone gives it hegemonic power to tax its citizens how it likes, but Progressive Web Apps mark the first fissure in the fortress as apps can undercut their fees while still enabling notifications and data collection.
   
   Of course, you’d only need to look at the history of crypto apps over the past five years to know these have all succeeded just fine as web apps too.
   
   

2. It’s Horizontal. It’s been a truism since the supposed Unbundling of Craigslist that apps with verticalized use cases for specific industries and user profiles are the ones that win. When we pitched a platform for contests that could be used for hackathons, grants, bounties, prediction games, giveaways, governance, etc. last fall, my cofounder and I were repeatedly met by a shared concern among VCs: we should pick one use case and tailor to its needs. While we might have argued that horizontal services (like, say, Thumbtack or Linkedin) typically win through distribution moats, economies of scale, and broader user bases, the VCs had a point. Even a Thumbtack or Linkedin are fairly verticalized services when you place them next to a Craigslist: the user profiles might differ within these apps (many professions), but their use case is basically the same (get a job).
   
   People have fundamentally used nearly every major app of the past decade to network professionally, to gossip personally, or in the towering case of Twitter, to network-by-gossiping all at once. And that’s it.
   
   Chat-GPT, by contrast, feels like a return to the Craigslist-Amazon “Everything Store” vision of web1. There is no set use case, and you can do literally anything you feel like within its interface—learn facts, plan routines, create content, get feedback, or just have a thoughtful counterparty listen to your woes. Part of the point of the proto-app thesis isn’t simply that the app and protocol layers are increasingly merged; more simply, the point is also that the apps really are proto-apps, reversions to the early web.
   
   But unlike Amazon, which bootstrapped its way to Everything through extremely specific use cases (books), Chat-GPT starts with Everything because, well, it supplies digitally abundant bits, not physically limited goods. The more things it can do, the more reason you have to return—because Chat-GPT’s greatest unlock might be that it doesn’t force you to define your category of use case at all. As AI, it responds on a case-by-case basis to whatever you’re thinking, so you no longer need to identify why you’re even there as you would in web1 (picking the proper Craigslist or Yahoo category link) or web2 (opening the proper app).
   
   The horizontalization of web services no longer looks like a limitation of forcing disparate services into the same UI as it might have in the Craigslist-era. It looks, instead, like a ploy to capture every possible use case of talking to an app according to the needs—and language—of the user.
   
   

3. It’s just a frontend for a protocol—that lets other services be integrated into its own frontend as well. And here’s the key point for why Chat-GPT can be horizontal in the first place after a decade of verticalization.
   
   Unlike traditional apps, which integrated the frontend and backend, Chat-GPT is, essentially, one possible version of a frontend among many tacked on top of a protocol, the OpenAI Model. You could throw other frontends on top of this engine, like papier-mâché bandages being thrown over the Invisible Man, and in fact OpenAI has: Dall·E is just a frontend to spit out visuals while Chat-GPT is a frontend to spit out text. As killer apps, both are in some ways afterthoughts to an underlying protocol that OpenAI assembled hastily so that people could interact with it.
   
   The decoupling of backend-protocol from frontend-app means, in effect, that:
   
   1) the app is just a circumscribed way to converse with the protocol within the limits of user-friendliness, a window onto a wider world, and
   
   2) in theory, anyone could build more frontend-apps on top of the protocol.

   
   For sure, OpenAI can and does power a number of consumer apps in the wild. But OpenAI’s great strategic decision is that it enables services to build on top of its protocol—through its own frontend in Chat-GPT. “Plugins,” as they’re called, let users connect to OpenTable, Expedia, Instacart, Zapier, Wolfram, etc., so you can order groceries and trips and (one day) market trades all through Chat-GPT. Ultimately, it can become the frontend not only for the OpenAI Model but for any service on the web. A few lines of natural language text, and you can execute your commands across the internet. Unlike every traditional data-siloed web2 model, it is a model build on composability—on the ability of one service to interoperate with any other across the internet.
   
   Chat-GPT is just the frontend for a protocol, I said. But that also means that anyone can build on top of that protocol through the app as well—and likewise, that it can become the frontend to any service that does so. Or more broadly: Chat-GPT can become a frontend to any transaction across the internet.
   
   Its power, as a protocol-app, is not just that anything can build on top of it, but that it can build on top of them in turn. It is a gateway to the web.

IV.

We can just look to a couple recent apps—Eco’s Beam Wallet, say, or Friend.Tech once again—to see the proto-app thesis starting to play out far beyond AI. They’re web apps, or specifically, Progressive Web Apps (PWAs), skirting the tax collectors of the Apple app store. They are frontends for smart contracts on a protocol, a blockchain. And the real question is whether they’ll let others build on top of them as well with services (raffles, giveaways, community-access, direct messaging, etc) that they could integrate into their frontend. Because doing so, becoming a portal to other services, is arguably what would let them go horizontal as well.

I’ve tried to avoid talking about crypto, but by this point, the jig is up: crypto is really all I’ve been talking about this entire time. With crypto, the foundational protocol is the blockchain itself, but any app you build on top of it can become a protocol in its own right too—because anyone can build on top of your app as well, permissionlessly, composably, reading the transaction data from your app’s open-source smart contract to write an action on their own app in turn. This is even, I’d argue, a massive advantage over a centralized AI protocol that can read and write any data in natural language, but siloes that data for its own end without enabling others to freely read it for their own apps as well. As my cofounder Sean has put it, blockchains, by contrast, are just open APIs for any service to trigger an action on any other service.

And that means, effectively, that every app is a protocol that can command fees off the blockchain through PWAs, outside the app store commission. Anyone can build a service on top of your app, and you, in turn, can integrate a service into your frontend. Every app in crypto can follow the same playbook that’s driven Chat-GPT to massive success.

This is also why, despite popular opinion, apps actually can make money when they merge with the protocol layer to let others build on top. By enabling others to monetize on top of them—through hooks, through funding, through group purchases, through collective investments, hell even through ads—proto-apps can safely take a cut. They make money when their users do.

And yet, because the consumer space in web2 and web3 remains ingrained in outdated theses that stopped working a decade ago, only a couple consumer services à la Lens Protocol actually want to build with composability as a core feature—and these remain the only proto-apps to date.

Uniswap v4, for example, is arguably the flagbearer for proto-apps (partly, to be sure, because there aren’t really any other options). By introducing “hooks” as a kind of onchain API for users to write their own code to automatically manage liquidity pools, Uniswap lets anyone build on top of its app as though it were a protocol. You can imagine users leveraging hooks to customize their pools however they want, automatically raising fees when trading volume soars, automatically paying off LPs the longer they’ve provided liquidity, or even, who knows, automatically sending profits into Aave to reinvest on behalf of the pool.

The hooks are not just plugins, in the language of Chat-GPT, but user-generated plugins. By letting users build out Uniswap’s core product, Uniswap not only becomes a protocol, but becomes a frontend for all its users’ desires—which it can then package for other users in turn. It becomes, in many ways, the frontend for DeFi. Just as you might want to stake your $ETH through Eigenlayer so that you can restake it across a number of protocols, you might want to deposit your liquidity into certain hooked-out liquidity pools on Uniswap that can manage that liquidity across other DeFi services as well. Both Eigenlayer and Uniswap can become distributional networks to access other services. To create a proto-app is to create a gateway to all the other services in your field. 

And it’s here that I need to come clean about my own bias, my own agenda. The reason I have written this piece is because I, too, am working to make a proto-app. Our thesis at JokeRace is that contests are, effectively, just games for communities to make decisions—and that when these decisions are made onchain, they can be executed across any other onchain service as well. We can build, in effect, the frontend to every financial and social service available to communities onchain, all while funneling users to them. The ace team at Hats Protocol, for example, built a module on top of us that can give the winner of a contest a “Hat” (an onchain role), and we can imagine advanced options for apps to build on us in the future that we can incorporate into our frontend in turn: funding modules that pay out recipients based on a winning algorithm, or treasury management modules that allocate funds according to winning strategies.

But these modules raise a bigger point. Because proto-apps in crypto are apps built directly on financial rails, you can see why it’s so important for them to avoid operating through app stores that would compete for fees. Again, proto-apps are distributional networks that can take a cut from the services they feature. They are, in some ways, the new app stores themselves.

I’ll stop shilling here. And I’ll just say that the fact that these modules can be built by pretty much anyone and used for pretty much anything without seeking approval of an app store is why web apps and horizontalization are so key to building apps as protocols: proto-apps are choose-your-own-ending applications, or if you prefer, frameworks for user-generated apps. They let anyone build the tools to interact and transact with each other however they like. They are, if you prefer, metaapps, apps for building apps.

And that is the irony of the situation. At the moment that apps look like an apocalyptic investment category, has-beens of another era, we’re on the verge of users being able to develop their own apps within their favorite apps, Roblox-style. No longer need new app developments be constrained within the siloes of the incumbents—now anyone can build them, permissionlessly. It’s apps all the way down. Which is to say: it’s protocols all the way down too.

But there is a logic here too. The reason there have been no great apps in the past decade is because the great apps, as we knew them, were already claimed. The great apps as we don’t know them, however, have just begun.

That’s because the next great apps won’t look or operate like apps at all—they’ll look and operate like protocols. Or rather, these proto-apps will look and operate like nothing we’ve quite seen.

These proto-apps will look and operate like, well—they’ll look and operate like proto-apps.

David Phelps
Aug 28-Sep 7, 2023

With immense special thanks to Ben Basche, Ria Bhutoria, Gaby Goldberg, Spencer Graham, Tina Haibodi, Li Jin, Sean McCaffery, Nick Naraghi, Peter Pan, Kinjal Shah, Seyi Taylor, and Jesse Walden for feedback.

Leave a comment