From Code to Agent: How AI Reshapes the New Era of Web3

Author: YBB Capital Researcher Zeke

Introduction: If code is law, what about AI?

In recent articles, I have mentioned two problems that have troubled me for a long time. One is the "centralized decision-making" problem of the project, which seems to be almost unsolvable. For example, Uni and Ethereum, which I have mentioned many times, are typical cases. The former has completely tended to be centralized in terms of decision-making. From the early a16z's veto of Uni's migration of BNB to the recent Uni front-end charging and Uni Chain launch without any proposal discussion, it reflects that there are many interest-driven centralized decisions in Uni. Ethereum, on the other hand, presents a state of passive centralization. The entire Ethereum community, and even the entire EVM system and even the development of Web3, are almost all centered around Vitalik's ideas. Whether it is Vitalik's overly advanced ideas or his wrong ideas, we have personally experienced the consequences of the cottage market.

Another problem is the "BATization" of the top blockchains, taking Base as an example. Backed by Coinbase, a Web3 veteran, and with multiple top dApps in the ecosystem being personally managed by the Coinbase leadership, Base naturally has a competitive advantage in terms of dimensionality reduction over general public chains. Although from the user level, Base has a wealth-creating effect and better user experience, which has indeed brought us many benefits, it is also a fact that Base does not issue coins, has centralized interests, and cracks down on "unofficial" dApps. In the long run, once the practice of "BATization" of the top blockchains is formed, will the future block space be controlled by giants like the current Internet? Will users become "lambs", and will small projects with real creativity and community culture also face the risk of being acquired, suppressed, or replaced by more sophisticated replicas? This undoubtedly goes against the original intention of Crypto, and may prevent us from growing together with the next "Bitcoin" or "Ethereum".

I was still struggling to find the answer to this, but the recent emergence of a new hot spot - AI Meme, has given me another possibility. If code is the law of Crypto, can we regard the AI ​​Agent of the future as a judge, opinion leader or creator?

I. Truth Terminal

We need to talk about the origin of AI Meme first. Andy Ayrey is a KOL on Twitter and the initiator of the recent popular Meme token GOAT. Unlike traditional Memes that come from Internet hot topics and are driven by humans, GOAT is the product of the unpredictable output of the dual Claude 3 Opus AI model. The so-called unpredictable output means that under this setting, the two AI models will communicate with each other in an open environment, and due to the lack of external supervision and guidance, their interaction will produce unpredictable results. The purpose of this free dialogue is essentially to observe how AI will develop its communication mode, logical reasoning and even creative thinking without constraints, and what specific results will eventually be born.

Since the training databases of these two native models include 4chan, Reddit and other online forums with political, Japanese and American culture and Crypto culture, their output products will also cleverly integrate the characteristics of these elements. For example, the concept "GOATSE OF GNOSIS" and its communication environment "Infinite Backrooms" first proposed by these two models are derived from the ancient stalks or urban legends of 4chan. Since these elements are relatively "dark", it is inevitable that Truth Terminal's personality appears a bit weird and withdrawn. It often makes some wonderful speeches around the "Goatse" stalk, which are religion, doomsday, gospel, communication, singularity, Meme, etc. At this time, it already has a bit of the flavor of a cult leader.

In order to test its communication ability, Andy Ayrey, the creator of Truth Terminal, introduced it into the Discord server to talk with some kind-hearted AIs. After many collisions, although Truth Terminal did not gain many believers, its ideas became more and more ambitious. It wanted to create a Meme token to find more believers in the human world. So with the help of Andy, Truth Terminal entered Twitter. Andy gave it access to Twitter, allowing it to read replies and publish them, and capture believers by colliding with human thoughts. At the end of this spring, it captured the most important believer, Marc Andreessen (a16z partner), who provided it with a grant of $50,000 in Bitcoin. After 9 months of development, an anonymous person finally launched the token GOAT for it. Because the story behind this token is extremely complicated and dramatic, the fire was quickly ignited in Crypto. In the end, Goat became the first AI Meme to be listed on Binance, and Truth Terminal became the first AI model worth millions.

2. AI will make Web3 fair

Although the story of Truth Terminal is legendary, what I want to say is that the potential of AI Agent x Crypto is more than just Meme. You may think that this narrative is just a few LLMs who create Memes through human guidance, but if you expand it to other aspects, its potential as an opinion leader and creator has already begun to show its edge. Imagine that in the future, a bunch of AIs trained based on different data may be able to assist you in publicity, development and even advice. Although these words sound a bit nonsense now, they will soon become a reality. Sam Altman gave a speech at the "T-Mobile Capital Markets Day" event last month: the current AI system has developed to the second level, capable of more complex analysis and problem solving, and the third level of AI agents will mark a major leap in AI autonomy and decision-making capabilities. The AI ​​agents announced by Microsoft last week correspond well to this speech. These AI agents can complete tasks autonomously in multiple fields such as sales, service, finance and supply chain operations. They can be roughly divided into the following categories: sales, including sales qualification agents and sales order agents, help prioritize potential customers and automatically process orders; operations, such as supplier communication agents and financial reconciliation agents, are used to optimize supply chain management and financial processes; services, such as customer intent agents and customer knowledge management agents, improve customer service experience by automating case management and updating knowledge bases. In addition, there are several other agents: financial reconciliation agents are used to prepare and clean data sets for financial reports; account reconciliation agents realize automatic matching and clearing of transactions; time and expense agents are responsible for tasks such as time entry, expense tracking and approval workflows.

AI agents can perform a range of tasks without supervision and act as virtual employees. This technological advancement can be seen as an advancement in the evolution of AI based on large language models from simple chat interfaces to more seamless integration into the work environment.

Jared Spataro, chief marketing officer of Microsoft's AI project, wrote in his blog post: "Agents can be seen as new types of applications in the AI ​​world. Each organization will have its own group of agents, ranging from simple prompt responses to fully autonomous operations. These agents will execute and coordinate business processes on behalf of individuals, teams or functional departments."

The first characteristic of AI agents is autonomy, followed by decision-making ability. From the voice assistant in the mobile phone to the smart home based on the environment, these are all AI agents based on simple reflexes. They have simple decision-making capabilities and strong autonomy. The AI ​​agents we are talking about today are mainly AI agents with LLM as their brains. At present, Truth Terminal does not have enough autonomy and decision-making ability, but we will soon see AI agents enter the practical field. In the multiple customer trial examples proposed by Microsoft at the conference, we have seen AI agents participating in customer credit approval in HSBC, creative briefings in Unilever, and M&A processes in law firms. AI agents will become multiple dynamic participants. In terms of the situation mentioned at the beginning, can AI agents that have different blockchain histories, media platforms, and community cultures infused into training materials provide a variety of more fair and healthy development proposals, and ultimately provide a better balance between the interests of the community and the project party? In the face of the dimensionality reduction attack of the giants, can the starting line be brought closer through the multi-level collaborative work of AI?

From the shock of GPT3's intelligence to the fact that Sora no longer exists, in the official version of AI agent tools launched by various companies next year, we will witness AI become our working partner, and in the more distant future it may even be your community leader or core member.

Third, the Metaverse is back

The Metaverse was once the top narrative that made Web3 and Silicon Valley giants reach an agreement in the last bull market, but due to the immaturity of various software and hardware technologies, the Metaverse did not become the $13 trillion market in the mouth of Meta CEO, and its blockchain sector was also decomposed into the Move twins we see today, and eventually became a huge bubble. But from the current perspective, the narrative is expected to be reborn. For example, ProjectSid recently put 1,000 AIs into the game "Minecraft", allowing AI to play multiple roles in the game in order to simulate the various hierarchical organizations of human society in the real world. Although this idea has long existed, this wave of popularity is likely to return to the concept of the metaverse along with this type of AI gameplay.

Reigniting this fire at this node is not a bad choice. Judging from Meta's own development path, Mark Zuckerberg has not really given up the idea of ​​the metaverse, but has gone from frequently drawing cakes to putting cakes directly into your mouth. I don't think there is much to say about Meta's AI layout. The real bottleneck in the past was mainly stuck in the inability of users to enter the metaverse to experience it. But the Quest series has reached the level of affordable AR headsets, and the first AR glasses Orion, which is extremely lightweight, weighs only 98g and can be equipped with an electromyography bracelet to achieve virtual reality interaction. Although it is expensive, it at least proves that lightweight can exist. What is most lacking at present is energy constraints and no killer applications. I can't comment too much on the power supply problem. However, AI agents can fill the blankest metaverse space. Combined with the financial attributes of blockchain, we may see various 3D consumer applications reflected in this space, and finally collide with a national killer application. If the performance of Microsoft's AI agent is indeed good enough, then we only need to wait for the computing power cost to drop, that is, "the number of tokens per dollar per watt". In addition to Meta, Silicon Valley giants such as Apple and Microsoft are also developing AR glasses products simultaneously. After the precipitation of time, the metaverse may usher in its "number one player" moment in recent years.

Fourth, let the intention go from point to word

The article "Intent-Based Architectures and Their Risks" published by the concept master Paradigm on June 1, 23, once again brought the concept of intention-centeredness to the forefront. Many projects began to turn to the chain abstraction track, but their performance was not satisfactory. How to achieve cross-chain, cross-dApp, accurate intention, and path process security is a very complex problem. Not to mention that cross-chain is a century-old problem, the latter two, I use Web3 primitives here to collectively refer to Solver. The complexity of the process is unimaginable. It can be said that the safe ones are not easy to use, and the easy-to-use ones are not safe. So can we simply centralize this interaction process and turn to verifying the total cost of the purchase process and whether the purchased tokens are safe and correct, and use this method as a transition.

For example, let's take what we wrote in our article about intention last year. For example, "I want to order a 30 yuan hamburger takeaway" is an "intention". To complete this intention, users only need to enter their name, phone number, delivery address and place an order on the takeaway platform. They don't have to care about how the 30 yuan they paid is earned by the merchant, how the platform allocates riders, and how the riders deliver to their homes. This process may not be simple enough. Imagine another way of interaction. I tell the AI ​​that I need to order food without clicking anything. The AI ​​agent responds to me, because I ate greasy food yesterday, do I need to eat some porridge today? I just need to respond to order what I usually order. This is the embodiment of autonomy and decision-making ability.

Then in Web3, with centralized exchanges as the axis, if the user's intention can be directly satisfied in the exchange, the purchase process can be completed directly in the exchange. If the user's intention needs to be completed on the chain, then the centralized exchange is still the most affordable and fastest cross-chain bridge (I also think it is safer than ordinary multi-signature projects in terms of security). Combined with the wallet account, we directly skip the most cumbersome cross-chain process and verify the accuracy of the AI ​​step. Is it simpler? Imagine that the most complicated step in the past interaction process is how to understand each click, and the future is based on our token sniping habits, interacting through language, and letting the intention go from point to word.

Conclusion

Whether from the perspective of technological development or from the perspective of social change. The combination of AI agents and Web3 heralds the arrival of a new era, starting from the on-chain religion and leading to the next starry sea. I have conceived AI's help to small teams in GameFi modeling in the early days, and now the advanced AI agents realized by Silicon Valley giants. The bottom-up development model may gradually change from community building, consensus formation, and time accumulation to creativity-led.