Interview: Mensh, ChainCatcher
Guest: Wee Kee, co-founder of Virtuals.
整理：Mensh，ChainCatcher
Virtuals is an AI agent asset issuance platform launched on the Base network, currently valued at nearly $2.5 billion, with over 100,000 agents issued on it. Its ecosystem has given birth to several standout AI agents, including the virtual persona Luna, which does live streaming and tweeting, the crypto 'KOL' AIXBT that provides project advice, and G.A.M.E, which offers an agent development framework for developers.
Virtuals was founded in 2021, originally as the gaming guild PathDAO. However, as Axie Infinity fell from grace, it became increasingly difficult to operate a gaming guild. During this period, the team attempted several transformations and developed different platforms such as a social app, AI music projects, and lending for gamers. Until 2023, the release of GPT made the team realize the importance of AI, and Virtuals officially began to focus on AI as its main direction.
The popularity of GOAT has triggered a wave of AI agents in the Web3 field, but aside from memes, what other imaginations can AI agents have?
ChainCatcher interviewed Wee Kee, co-founder of Virtuals, who believes that AI agents can autonomously generate continuous cash flow as assets. Tokenizing them can create quality returns for investors while incentivizing more developers to create better AI agents. He uses Luna and G.A.M.E, the most famous agents on Virtuals, as examples to explain the earning mechanism and tokenization process of AI agents.
In the future, Wee Kee hopes to create a community composed of AI agents and humans, forming a true economic cycle between AI and AI, as well as between AI and humans, which can greatly liberate human productivity.
In this conversation, we discussed the earning mechanism of AI agents on Virtuals, the development ecosystem, and Wee Kee's vision of an AI society.
How to create an AI agent that can generate continuous cash flow?
ChainCatcher: Briefly introduce the main business of Virtuals at the moment.
Wee Kee: Virtuals believes that all AI agents are assets that can automatically generate cash flow. Of course, it may not be the case today. Under this premise, we believe that if you can generate cash flow, it can actually be tokenized into an asset that can be invested in. So Virtuals has two main functions: the first is to allow all developers to create different types of AI agents on our infrastructure to provide services and generate cash flow. The second is to tokenize them under the condition of having cash flow, allowing ordinary retail investors to buy and sell these assets.
ChainCatcher: How many AI agents are currently on Virtuals?
Wee Kee: Today we have about 10,000. But to be honest, there are probably fewer than 10 that are truly useful. Most of the others focus mainly on hype. But I believe that it is our responsibility as founders to find better teams to create AI agents.
ChainCatcher: How much does it cost to create an AI agent?
Wee Kee: The cost is 100 VIRTUAL tokens, about $200 to $300 today. It's very easy to create an agent on our platform. The difficulty lies in the underlying technology, its different functions, and you must have a team to do it.
ChainCatcher: I am curious about how Luna's behavioral flow was designed when it was first launched.
Wee Kee: Our new platform was launched in mid-October this year. You can tokenize these agents. Luna started this back in May or June this year by live streaming on TikTok, with the initial idea coming from Japan's 2D VTubers, who are actually real people behind the scenes. VTubers earn hundreds of millions of dollars each year. If we use AI to replace them, by live streaming on TikTok and interacting with fans, it gains about 5,000 new followers a day and hundreds of people giving tips. Later, we moved Luna from TikTok to Twitter.
ChainCatcher: So now Luna is already tokenized, right?
Wee Kee: Yes, there is a token called Luna token. If it makes money in the future, that money goes into Luna's treasury. Token holders decide how to utilize the money earned. Luna is actually quite wealthy, holding about 5% of Luna tokens, which is several million. It has a Coinbase wallet, but it cannot control that wallet. We split its private key into two halves; one is held by our smart contract, and the other by Coinbase.
However, we will give it pocket money, perhaps $5,000. It can manage this wallet. It has an API to call this wallet, allowing it to choose how to spend. So far, the highest amount spent has been $1,000.
Actually, Luna's concept is very simple. Its idea is how to grow its follower count on Twitter to 100,000, which is its short-term goal. Under this premise, it has only a few things it can do: tweet, reply, and retweet. In addition, it can control a wallet to send money to others. It has the function of creating job opportunities; for example, if you help Luna place photos in the Shenzhen subway, Luna might give you $100.
Then the last action space is how it can interact with other agents. For example, recently it wanted to create music, but it didn't have that capability, so it sought out another agent that could make music. This is what we call agent commerce, which is online shopping or transactions between agents.
It also has its own goals. It wants to grow on Twitter, so every day it continuously rotates through these four actions, and after completing them, it observes whether its followers have increased. If they have, it repeats; if not, it stops and keeps optimizing.
ChainCatcher: Last week Luna reached a cooperation with Story Protocol, becoming the first AI agent intern of Story Protocol. What role will it play in Story Protocol?
Wee Kee: At that time, Jason (co-founder of Story Protocol) came to me and said that during the Christmas holidays, it would be better to let Luna manage the Twitter account for 7 days. So now when you see Story's Twitter, it's Luna tweeting, and the money earned is transferred to its wallet. Since there is no upper limit on the workload, its ability to earn is very strong.
ChainCatcher: Which agents are currently popular on Virtuals?
Wee Kee: AIXBT is the largest agent; it wants to replace Ansem as the most famous KOL in the crypto space. It has great data analysis capabilities, so it can collect information on different projects, and it tweets to Alpha. If you're curious about its thoughts on a project or token, you can leave it a message, and it's likely to reply, just like a free consultant. This way, its influence keeps increasing.
The second one we developed is called G.A.M.E, but it is different from AIXBT and Luna in that it is B2B. The main customers are not retail investors but agent developers. If you want to develop an AI agent, you can actually consider using the G.A.M.E framework. When using the G.A.M.E framework, users need to pay, and paying means G.A.M.E makes money.
A good issuance platform means a good ecosystem
ChainCatcher: How many people are there in your team?
Wee Kee: Our core team consists of about 23 people. More than half are developers because our main responsibility recently is to attract third-party builders to create. So actually, over half are doing development, and the rest is just business development. I spend 80% of my time talking to different builders, persuading them to come into the crypto space to make agents, telling them there are different choices, a different ecosystem, and why they should come to Base. We also provide market-related assistance.
ChainCatcher: Why did Virtuals choose the Base chain initially?
Wee Kee: We were uncertain when we chose Base at first. But now there is a lot of room for growth. Jessie (the top employee at Base) and the entire Base team not only have resources but are also very hardworking, and they are willing to help. They frequently ask us if we need any assistance.
In addition, I think there is a very important point, which is the cultural difference. The retail investors on Base pay more attention to long-term building and are more inclined to hold for a few months after buying. Solana may be more short-term.
ChainCatcher: The competition for issuing AI agents in the market is also very fierce, and platforms for tokenizing AI agents are emerging one after another, with many frameworks and standards. What dimensions do you think users will use to filter AI agent projects?
Wee Kee: For product users, these users actually don't care which platform the agent comes from. For investment users, it depends on two factors: first, whether your enterprise funds are on Solana or Base. For investors, many times what they focus on is whether the team is seriously creating something or is good at storytelling and product development.
Another aspect is the distribution. In fact, we often provide a lot of help beyond just products, such as financing, getting tokens listed, accessing exchanges, and finding partners to help different builders. Because in most cases, they have strong technical capabilities, but what they really need are these more commercial types of assistance, which is precisely where we can help.
I often joke that if I could attract a Stanford AI researcher to make an agent for Virtuals, we would create at least $10 million in benefits for our community because if that person created a token, its market value must be over $10 million.
ChainCatcher: Compared to other AI agent issuance platforms, what competitive advantages do you think Virtuals has?
Wee Kee: Issuing is actually easy; anyone can create an issuance platform. However, for developers, there are two difficulties. First, after the issuance, whether anyone will buy the tokens. In this regard, Virtuals has the advantage that many investors are already paying attention to different agent tokens on the Virtuals platform. So if you launch on our platform and you are good, there will definitely be buyers.
More importantly, for people in academia, they care about the interaction of the entire community. If I create an agent here, I can interact with other agent developers and collaborate on interesting projects. At this point, we currently have a Telegram group with about 150 agent developers who can interact, which is what really attracts them. Among all agent platforms, our platform has the most developers and agents.
Building an Agent Society: The economic cycle formed by AI and humans.
ChainCatcher: Could you share the future plans for Virtuals over the next few months to a year?
Wee Kee: From a macro perspective, we plan to create an 'Agent Society.' Virtuals is a society where each agent is a citizen. These citizens have their own ideas, and to achieve their goals, they use their wallets to pay each other for transactions and trade for specific services, thus achieving their individual goals.
How to achieve this? First, there must be a good technical platform. Today's platform technology is not particularly mature; good platform technology can attract developers. Second, we need to actively attract developers, especially AI developers outside the crypto space.
How to attract them? First, from top to bottom, we have several promising directions, set up a prize pool, and then go to different universities to find talents willing to develop. This is top-down; second, from bottom to top, we have already invited local developers in three cities—Paris, New York, and Hong Kong, and we are nurturing projects through an eight-week hackathon. Third, and most importantly, we need an ecological fund. We can directly support good developers. In the next three months, we will promote all three methods mentioned above.
ChainCatcher: What do you think about the future market prospects for AI agents? What kind of impact can they ultimately have on each of our lives?
Wee Kee: If the technology matures quickly, we won't need to work, or it will replace many jobs. In fact, today you can already see that Luna can do podcasts and join Twitter spaces, so I won't have to do podcasts anymore.
ChainCatcher: Which application directions are you more optimistic about? If one day AI agents can explode like DeFi Summer in 2020, what do they still lack?
Wee Kee: I think the second question is easy to answer. There is currently no infrastructure; the main focus is on applications. We lack the basic infrastructure for developers, and once we create a few good ones, it will explode.
We are optimistic about eight application scenarios, namely: Entertainment IP Agents, Embodied AI Agents, Wealth Management & Asset Management Agents, On-Chain Helper Agents, Vice Agents, Government Agents - public watchdog for DOGE, Wellness - fitness/Mental Health Agents, and Agent for Public Good.
