When all applications look like protocols: Prototype applications rise in Web3

Original Title: The Proto-App Thesis

Original Author: DAVID PHELPS

Translated by: 深潮 TechFlow

I.

The following questions plague those who dare to build consumer applications:

1. When all evidence points to protocols, do we really believe that applications will accumulate value? In a world where protocols like Ethereum easily generate fees, prove the legitimacy of tokens on a secure base layer, and capture value from all applications built on them, can applications really make money?

2. Given that applications increasingly resemble the historical phenomena of 2008 to 2012, can new applications even establish sustainable users? After all, why have there been no widely adopted, lasting new applications in nearly a decade, or if we exclude TikTok, in over a decade?

The obvious answers to these two questions are so grim. Once our investors come to the understandable conclusion that applications cannot accumulate value, investment in applications will inevitably decline, and the hope of refuting this view will be very slim.

Let’s be fair to the other side: some investors might ask whether this negative sentiment means that applications have hit rock bottom? Perhaps one application thesis has reached its end, and another will rise? Of course, this is based on the assumption that the application cycle is indeed a cycle, and not just a fleeting moment in the decade of Web2.

However, while these investors may diligently provide evidence for a new application thesis, they are likely to do so only by making a better case for the protocols. Yes, our investors can cite examples like Friend.Tech, which had sales of about $1 million at its peak; but today, that number is close to $16,000, having dropped nearly 99% in two weeks. Similarly, our investors might argue that Friend.Tech's fees far exceed those of the protocol it is built on, Base; but if the application indeed heralds the arrival of more applications, then Base will inevitably gain total revenue from its suite of applications that exceeds any single application's earnings.

The argument seems clear: even for the strongest arguments for applications, they can essentially be counted towards the protocol layer.

Joel Monegro wrote the "Fat Protocol" piece during the peak of the application era in 2016, pointing out that "the market cap of protocols always grows faster than the value of the applications built on them because the success of the application layer drives further speculation on the protocol layer." Many have tried to refute the fat protocol thesis—including myself—but by 2023, it seems to have become an unspoken truth.

In fact, the fat protocol thesis provides us with an enticing answer to both questions: the reason no great new applications have emerged in the past decade is precisely because value has begun to accumulate at the protocol level.

Thus, rather than trying to prove that applications will somehow become better investments and revenue generators than protocols, we should seek to argue the true answers to our two key questions—can today’s applications accumulate value? Can today’s applications gain sustainable users?

Even asking whether value accumulates to applications or infrastructure assumes that they are mutually exclusive categories. If we ask why we have no significant new applications, we overlook the fact that we already have them, but they may not look like traditional applications.

So why don’t they?

Because the next wave of great social applications will look like protocols: Proto-apps.

II.

Why have there been no widely adopted, lasting applications since TikTok—only a few sporadic novelty applications that fade with time? This can be attributed to what we call the social application thesis, which has determined the fate of the major applications of the past decade, namely social applications. In short, the categories in the social space are limited, and the winners in each category have gained distributed network effects strong enough to fend off competitors.

In other words, our app stores are full, and we have exhausted the five main categories of online expression:

• Long-form video (YouTube);

• Short-form video (TikTok);

• Long-form text (Reddit);

• Short-form text (Twitter);

• Images (Instagram).

In the early stages of social applications, you could create a successful application by mixing these categories, such as Tumblr or Facebook, which combined images, short text, long text, and video. But that opportunity only existed in the years before social applications were deconstructed.

In fact, in the past five years, the only hope for launching a successful new application has been to carve out an entirely new category: for example, live audio (Clubhouse) or live photos (BeReal). While these real-time experiences are novel, they hinder sustained interaction, which is the foundation of successful social engagement.

In other words, there are no remaining categories, and there is no way to compete in the categories that currently dominate our daily lives. You could create a competitor that is ten times better than Twitter: ultimately, Twitter remains the place where publishers gain the widest audience reach, and readers gain the broadest access to publishers. They cannot get that from other applications unless they spend years retraining their social networks. Every second spent doing so is a second away from Twitter, where readers and authors spend so much time that they consume each other's attention, unable to seek elsewhere.

Distributed network effects (or the lack thereof) also help explain why there are two main categories of applications where newcomers can share market share without completely disrupting existing applications: messaging and dating.

It can be argued that messaging and dating applications follow the same pattern. Because they are not public markets for the flow between creators and their audiences, their network effects are weaker. Simply put, creators are not using messaging and dating applications to reach the broadest audience of strangers, so they lose nothing when switching applications to connect with different groups. But note that the main players here also have a decade-long history: Signal was founded in 2014, Telegram in 2013, Discord in 2012, Snapchat in 2011, WhatsApp in 2009; Bumble was founded in 2014, Tinder in 2012, Hinge in 2011, and so on.

Even with weaker network effects, the category is still too crowded for newcomers. As Chris Paik said in 2021, "The reason we haven’t seen new mobile-first social companies emerge is that we have entered the stage of the smartphone's half-life."

According to the social application thesis, we have reached saturation.

It is hard to say that the social application thesis is wrong; indeed, within the scope of this narrative, it seems irrefutable. But interestingly, the social application thesis overlooks the fact that in the past decade, an important new application has emerged for the first time. However, the magnitude of this application’s change is such that you might not even have noticed it as an application. I am, of course, referring to Chat-GPT.

In fact, to call Chat-GPT any kind of social application is somewhat of a stretch. Of course, you could make the argument. You could argue that Chat-GPT is just another way to maintain an endless conversation on a specific medium, involving everything you are interested in, just like other major social applications. The difference is that what you produce is not for the world to see, but just for yourself, and there is a good reason for that: you are chatting with a higher version of yourself, a version of you that has all the answers.

But even if we do not classify Chat-GPT as a social application, it tells us something—we take for granted that major applications will be social applications, and the rules of user-generated content and instant messaging applications will define the rules of all major applications.

So please bear with me. The uniqueness of Chat-GPT as a new type of social application has nothing to do with whether we classify it as a social application. You can say it is a new type of social application or not; it does not matter for the point I want to make, which is that we need a new paradigm to define the successful applications of the future. So let’s be bolder:

Chat-GPT is the first significant application of the past decade because it is the first example of a whole new category: Proto-apps.

To understand Proto-apps, we need to grasp what makes Chat-GPT an absolute exception to the rules of applications that defined the past decade.

III.

We can say that at its algorithmic core, Chat-GPT breaks the traditional paradigm of successful applications in three ways:

1. It launched as a web application. While Chat-GPT recently released the inevitable mobile application, unlike almost all other major applications of the past 15 years, it first launched as a web application. And there is good reason for that: by design, Chat-GPT only requires a web1 user experience, just a prompt bar and a response window. For Chat-GPT, the building blocks of web2 social, such as navigation bars, search, and information feeds, have magically disappeared. That’s why we hardly need a Chat-GPT app, just like we don’t need a Google app; embedding them in a browser makes more sense.

But more practically, Chat-GPT is a web application because launching a mobile application first is a bad idea. While common sense suggests that major applications should be mobile applications, downloading apps has increasingly become a major friction point that only the most popular services can overcome. More importantly, mobile applications have the advantages of push notifications, data collection, and interoperability with device assistants, but they come with a 30% app store commission. As I mentioned in my piece "Apple is a Nation," Apple’s vertical control over the app store, operating system, and phones gives it the hegemonic power to tax, but progressive web applications mark the first crack in this fortress, as applications can lower costs while still enabling notifications and data collection.

Of course, you only need to look at the history of crypto applications over the past five years to see that these applications have also succeeded as web applications.

2. It is horizontal. Since the so-called deconstruction of Craigslist, there has been a consensus that applications configured for specific industries and user cases are the winning applications. Last fall, when we launched a competition platform for hackathons, grants, bounties, prediction games, giveaways, governance, and more, my co-founder and I repeatedly encountered a common concern from venture capitalists: we should choose a use case and customize it based on its needs. While we might think that horizontal services (like Thumbtack or LinkedIn) typically win through distribution moats, economies of scale, and a broader user base, the venture capitalists have a point. Even Thumbtack or LinkedIn, compared to Craigslist, are relatively verticalized services: the user profiles of these applications may differ (many professions), but their use cases are fundamentally the same (finding jobs).

Over the past decade, people have essentially used almost all major applications for professional networking, casual chatting, or giant applications like Twitter while engaging in network chatting.

In contrast, Chat-GPT feels like a return to the web1 vision of a "store" like Craigslist-Amazon. It has no fixed use case; you can do anything you want within its interface—learn facts, plan your day, create content, get feedback, or just have a thought-provoking conversation partner listen to your worries. Part of the proto-app argument is not just that the application and protocol layers are increasingly merging; more simply, the key is that these applications are actually proto-apps, a return to the early web.

But unlike Amazon, which gradually evolved into a "store" through a highly specific use case (books), Chat-GPT is "everything" from the start because it offers digitally rich bits rather than a limited number of physical goods. The more it can do, the more reason you have to return—because Chat-GPT’s greatest breakthrough may be that it does not force you to define your use case category. As an AI, it responds case by case based on your thinking, so you no longer need to determine why you are there, as you did in web1 (selecting the right Craigslist or Yahoo category link) or web2 (opening the right application).

The leveling of web services is no longer about forcibly merging different services into the same user interface as it was in the Craigslist era; instead, it looks like a strategy to capture every possible use case of conversation with applications based on user needs and language.

3. It is merely a front end of a protocol and can integrate other services into its own front end. This is why Chat-GPT can become a horizontal key after a decade of verticalization. Unlike traditional applications that integrate front ends and back ends, Chat-GPT is essentially a front-end version, one of many fronts built on top of the OpenAI model. You can add other fronts on this engine: Dall·E is just a front end for generating images, while Chat-GPT is a front end for generating text for people to interact with.

Decoupling the back-end protocol from the front-end application means:

1) Applications are just a limited-friendly way to converse with the protocol, a window into a broader world; and

2) In theory, anyone can build more front-end applications on top of the protocol.

Of course, OpenAI can and does power many consumer applications. But OpenAI's major strategic decision is that it enables services to be built on top of its protocol—through its own front end in Chat-GPT. "Plugins" allow users to connect to services like OpenTable, Expedia, Instacart, Zapier, Wolfram, etc., so you can order groceries, travel, and (in the future) market trades through Chat-GPT. Ultimately, it can become a front end not just for the OpenAI model but for any web service. With a few lines of natural language text, you can execute commands on the internet. Unlike the traditional data siloed web2 model, it is a model built on composability—a service that can interoperate with any other service on the internet.

I have said that Chat-GPT is merely a front end of a protocol. But this also means that anyone can build on top of that protocol through the application, and similarly, it can become a front end for any service using that protocol. Or more broadly: Chat-GPT can become the front end for any transaction on the internet.

As a protocol application, its power lies not only in the fact that anything can be built on top of it, but also that it can, in turn, be built upon. It is the gateway to the internet.

IV.

We can look at some recent applications, such as Eco's Beam Wallet or Friend.Tech, to see that the proto-app argument is playing out beyond AI. They are web applications, or more specifically, progressive web applications (PWAs), that circumvent the tax collectors of the Apple App Store. They are front ends based on protocol smart contracts, namely the blockchain. The real question is whether they will allow others to build services on top of them (raffles, giveaways, community access, direct messaging, etc.) and integrate them into their front end. Because doing so, becoming a portal for other services, can arguably be the key to their horizontalization.

I have tried to avoid talking about cryptocurrency, but at this point, it has become clear: cryptocurrency is indeed what I have been talking about. In cryptocurrency, the underlying protocol is the blockchain itself, but any application built on top of it can also become its own protocol, as anyone can build on your application without permission, reading the transactional data in your application's open-source smart contracts in a composable way, and then executing actions on their own applications. I believe this is even a significant advantage over a centralized AI protocol, which can read and write any data in natural language but keeps that data locked within its own application, not allowing others to freely read it for their own applications. As my co-founder Sean said, in contrast, blockchain is simply an open API for any service to trigger actions on other services.

This means that in reality, every application is a protocol that can charge fees on the blockchain through progressive web applications (PWAs) without having to pay commissions through the app store. Anyone can build services on top of your application, and you can integrate services into your front end. Every application in cryptocurrency can follow the same strategy, which is also why Chat-GPT has achieved great success.

This is also why, despite the widespread belief that applications can actually make money by merging with the protocol layer to allow others to build on top of them. By allowing others to profit on top of them—through hooks, funding, collective purchasing, collective investing, or even advertising—proto-apps can safely take a cut. When their users make money, they make money too.

However, since the consumer space in both Web2 and Web3 is still deeply influenced by outdated arguments that ceased to operate a decade ago, only a few consumer services (like Lens Protocol) actually want to build with composability as a core feature, and these remain the only proto-apps so far.

For example, Uniswap v4 can be said to be the flagship of proto-apps (partly because there are no other options). By introducing "Hooks" as an on-chain API that allows users to write their own code to automatically manage liquidity pools, Uniswap enables anyone to build on top of its application as if it were a protocol. You can imagine users customizing their liquidity pools according to their needs using hooks, automatically raising fees when trading volume spikes, automatically paying fees to LPs the longer they provide liquidity, and even automatically sending profits to Aave for reinvestment on behalf of the pool.

These Hooks are not just plugins; in Chat-GPT's language, they are user-generated plugins. By allowing users to build the core product of Uniswap, Uniswap not only becomes a protocol but also the front end of all its users' wishes, which can then be packaged for other users. In many ways, it becomes the front end of DeFi. Just as you might want to stake your $ETH through Eigenlayer to be able to re-stake it across multiple protocols, you might want to deposit liquidity into certain liquidity pools on Uniswap that can also manage that liquidity in other DeFi services. Eigenlayer and Uniswap can both become distribution networks for accessing other services. Creating proto-apps is creating a gateway to all other services in the domain.

I just want to say that these modules can be built by almost anyone and can be used for almost any purpose without seeking approval from the app store, which is why web applications and horizontalization are so important for building applications as protocols: proto-apps are applications that choose their own destinies, or if you will, frameworks for user-generated applications. They allow anyone to build tools to interact and transact in any way they like. If you will, they are meta-applications, applications for building applications.

This is the irony of the situation. Currently, applications look like a doomsday investment category, a relic of another era, while we are on the verge of users being able to develop their own applications within the applications they like, just like Roblox. New application development is no longer constrained by the barriers of existing applications; now anyone can build them without permission. Applications are everywhere. That is to say: protocols are everywhere.

But there is also a logic here. The reason no outstanding applications have emerged in the past decade is that the outstanding applications we know have been monopolized. However, the outstanding applications we do not know are just beginning.

This is because the next great application will not look or operate like an application but will look and operate like a protocol. Or more accurately, these proto-apps will operate in ways we have never seen before.