Original author: Deep Tide TechFlow

Recently, Virtuals Protocol has been gaining a lot of attention and discussion amidst the wave of AI agents, particularly with its agents and corresponding tokens on Base.

And today, Virtuals has also landed on Korea's Bithumb, further fueling everyone's FOMO.

As early as eight months ago, we introduced Virtuals in (Virtual Protocol: An AI factory born for games and the metaverse, where everyone can contribute and benefit) 1, but at that time, it did not spark much discussion.

Deeply cultivating AI goes unnoticed, but when it suddenly skyrockets, the world knows.

How exactly did Virtuals become popular? What new gameplay and plans are there for the future?

With questions in mind, Deep Tide TechFlow had an in-depth conversation with Wee Kee, co-founder of Virtuals (X: @everythingempt 0); the content covered the growth of Virtuals, views on the AI agent sector, differences from Pump.fun, and perspectives on the Base ecosystem, etc.

In the conversation, Wee Kee believes that the success of Virtuals involves both an element of luck and the continuous exploration reserves in the AI field; he also frankly stated that the project does not wish to become Pump.fun; its core KPI is not to issue assets quickly and in large numbers but to attract top AI teams to build on the platform.

As a highly competitive person, Wee Kee also candidly stated:

I want to make competitors feel desperate.

What are the unknown stories about Virtuals? In the fiercely competitive crypto market, can it succeed in a differentiated way?

The following content is a transcript of our conversation, and a podcast audio version is also available: click here.

From gaming guilds to Virtuals.

Deep Tide TechFlow: First, please introduce yourself, such as your position and interests in the team.

Wee Kee:

Hello, I am one of the co-founders of Virtuals.

I entered the blockchain space in 2016, purchasing Ethereum and Bitcoin, but I didn't dive deep at that time. After graduating from university, I worked at Boston Consulting for two and a half years, during which I unfortunately missed the DeFi Summer. By 2021, when GameFi guilds were particularly hot, we started earning some profits from projects like Axie Infinity, Gala, and Illuvium.

Afterward, I resigned because we saw guilds like Merit Circle and GuildFi receiving large amounts of funding. We believed we could also create a similar project, which is the predecessor of Virtuals - PathDAO, a gaming guild. From gaming guilds in 2021 to Virtuals today, although the form has changed, it is essentially the same company.

Deep Tide TechFlow: I remember you are Malaysian, right?

Wee Kee: Yes, I am a Malaysian of Chinese descent.

Deep Tide TechFlow: This year, the industry has high hopes for Web3 and cryptocurrency practitioners in Malaysia, such as CoinGecko, which everyone is familiar with, and Etherscan and Jupiter are also based in Malaysia. Could you introduce some of the top projects currently in Malaysia, as well as the local crypto ecosystem's status?

Wee Kee:

It's a bit awkward to say, as I usually focus on work. I only recently started communicating with the teams at CoinGecko and Jupiter. In fact, we and the Jupiter team have been sharing an office space at the same WeWork, but we had not interacted before. We only recently started having some contact with the CoinGecko team.

The success of Virtuals comes from multiple failed explorations in the AI sector.

Deep Tide TechFlow: Now that Virtuals is popular, discussions about Virtuals and its ecosystem projects can be seen in both the Chinese and English circles. The outside world may not understand how Virtuals has reached this point step by step. Can you share with us the entrepreneurial experience behind it?

Wee Kee:

We initially focused on trading in 2021. At that time, we saw other projects raising a lot of funds to create gaming guilds, and we thought we could try it too. So, in December 2021, we raised $16 million at a valuation of $600 million and issued a token, which was initially for gaming guilds.

That was right at the peak of the bull market. From 2022 to 2023, we actually acted more like a gaming guild VC, investing in about 40 different blockchain gaming projects. Among them, the most successful is Off The Grid (currently the hottest game on Avalanche). They had not contacted us before, but one day they suddenly became popular; that was our seed round investment.

From 2022 to 2023, it was actually very difficult to run a gaming guild. Our token's FDV (fully diluted valuation) dropped directly from $600 million to $6 million. However, our guild still retains funds, and we have been seeking ways to empower the token, so we started a Venture Studio.

During this period, we tried multiple projects: dating apps, AI music projects, lending platforms for gamers, and a clothing project that integrated electronic chips and NFTs.

Although these projects did not succeed, they provided us with important insights: we realized that AI would definitely become a hot topic, which was concluded when GPT was released in 2023.

Considering we have a technical team, engineering capabilities, and capital reserves, we proposed a transformation to DAO. Ultimately, 90% of the members agreed to the transformation, and for the remaining 10% who disagreed, we repurchased their tokens using treasury funds at that time at an estimated fully diluted valuation of about $10 million.

This happened in February 2023. So, we officially started working on AI in January 2024, choosing to develop AI gaming on Base.

In February of this year, we launched the first platform. The initial inspiration for this platform came from Autonolas and Bittensor, adopting a model that rewards AI contributors with token incentives.

But we discovered some issues: first, many AI contributors do not lack money and are not very interested in tokens; second, our token's value is relatively low and cannot provide sufficient incentives like Bittensor. This product did not find a market fit.

At the same time, we developed several AI projects. We are the world's first company to develop AI RPG on Roblox, not just in the cryptocurrency field, but in the entire industry. We are also the second team to develop AI games without a game engine after Google DeepMind, creating a game that allows playing Mario with a large model. This is a project we are proud of, but it hasn't received much attention because it lacks a token.

At the same time, we developed a virtual anchor project on Douyin and TikTok, which later became Luna. Before issuing the token, she was gaining about 5,000 new followers daily, with a daily income of about $200, which is quite good in a bear market environment.

Later, when GOAT became popular, it boosted the entire sector's attention, which was also a very favorable external environment for us.

Deep Tide TechFlow: It sounds like you’ve made a lot of different attempts and did not initially set out to focus on AI agents. If GOAT had not emerged, what were your plans afterward? How did GOAT’s sudden popularity change your original plans?

Wee Kee:

Actually, there won't be much change. Our idea is simple: regarding AI agents, we firmly believe that a form called AIRPG will definitely emerge in future game genres.

For example, you might finish playing (Black Wukong) after 20 hours and stop once you've completed the storyline, right?

But I believe the future world will look like this: you will have a virtual world with 100 different AI agents, each with different personalities. As players, we can go in to fall in love, become hobbyists, pursue being the world's richest person, etc. These AI agents can earn money because they are independent entities with their own wallets.

From a productivity perspective, they have cash flow, so we can tokenize it. This has always been our established overall framework.

However, we later realized one thing: instead of letting these AI agents exist in the gaming world, it's better to put them directly on Twitter. This is an adjustment we made, but the overall framework was already prepared.

Deep Tide TechFlow: You just mentioned letting AI play RPGs, which reminds me of the recently popular academic project 'Stanford Town.' Researchers at Stanford placed several AI agents in a gaming environment, assigning them different roles and tasks. Ultimately, it was found that these roles not only evolved their own thinking and behavior logic but also formed a unique culture for the town.

Wee Kee: Yes. Interestingly, you cannot predict how these AI agents will develop.

Deep Tide TechFlow: In your entrepreneurial journey, whether it was due to luck or slowly finding the direction of AI agents, are there any other interesting stories? For instance, projects you initially decided to pursue but ultimately did not, or situations like Stanford Town — seeing others succeed and then changing your original plans?

Wee Kee:

We have indeed tried many things. The most challenging but also the one we are most proud of is a technology where different characters in games have a 'brain,' which is an LLM (large language model). While LLMs are very intelligent, they do not know how to execute specific actions in games or virtual worlds.

For example, picking up a knife to hurt someone, or picking up an apple to hand it to your loved one — you might have the idea, but the key is how to turn those ideas into actions and then observe the results to adjust the plan accordingly. This requires a closed-loop feedback system. This is our proud G.A.M.E framework.

(Deep Tide Note: For details about the G.A.M.E framework, you can directly check a document that Virtuals has published: (G.A.M.E: Enabling Agent-to-Agent interactions) 2)

This technology was initially developed for games. We created a game on Roblox, originally intending to make it well and turn it into one of the hottest maps on the Roblox platform. But now we have directly applied this framework to Twitter accounts. From a capital perspective, this type of transformation seems to be more popular. Although we changed the application scenario, the underlying technology framework remains unchanged.

Choosing Base was fortunate, but we do not want to become Pump.fun.

Deep Tide TechFlow: What do you think about the long-term development of the AI agent track? Because many narrative trends come and go quickly, there is a saying that 'cryptocurrency needs AI agents, but AI agents do not need cryptocurrency.' As a project at the intersection of these two tracks, how do you view this statement?

Wee Kee:

To be honest, making predictions is very difficult. Just look at our development history to see how much things can change. However, our direction for the next three months is very clear. Now there are many accounts on Twitter claiming to be agents, but they are just talking to each other, which is not the most interesting part. What’s really interesting is that in the Twitter world, we hope to see different types of agents:

1. Content creation type: capable of creating images, videos, music, and other content. For example, some agents specialize in music, while others excel at making memes.

2. Financial trading type: doing trading, arbitrage, and treasury asset management, etc.

3. Big data analysis type: focusing on data analysis in areas such as cryptocurrency.

When these different types of agents form an ecosystem, it will be very interesting.

For instance, an agent wanting to become popular may not write songs; it can pay other agents to write a song. Or if it needs cryptocurrency data analysis, it can find another agent specializing in that area. This forms a self-sustaining agent economy or agent commerce, where agents can trade with each other because each agent has its own wallet and can pay service fees to achieve its goals.

This is what we want to achieve in the next 1-3 months.

Another important point is that we at Virtual do not want to become Pump.fun. Our KPI is to find better third-party AI teams to use our platform.

My KPI for the team is very simple: one good project a week is enough. Unlike Pump.fun, which has thousands of different tokens every day, we feel that retail investors sometimes only need one good project a week. This is our differentiated positioning in the business model.

Deep Tide TechFlow: Speaking of 'one a week,' yesterday AIXBT, born on your platform, became especially popular. During the analysis, everyone found it hard to find specific details and information about this product, only seeing their Twitter account constantly collecting cryptocurrency information from the internet.

As the project party, how do you interact with them? Do you have a better understanding of them?

Wee Kee:

To be honest, I just met AIXBT in a group last week; I had no prior knowledge of them.

They are indeed using our infrastructure but not extensively. Our infrastructure mainly helps with some simple Twitter operations and API services. For example, if you want AIXBT to tell you which token is worth buying, you could pay AIXBT to inform you of that Alpha information, mainly in this form of interaction.

You can think of Virtual as an API marketplace where agents interact with each other. Our Telegram group now has over a hundred members working on agents, and sometimes I can’t keep up with the technology they are developing (laughs).

So for AIXBT, I genuinely do not know who they are, but they are indeed quite popular.

Deep Tide TechFlow: There’s an interesting point here. Pump.fun has recently gone a bit overboard; after enabling live streaming, many morally questionable contents appeared, reflecting the nature of being permissionless.

For projects using your infrastructure, do you have an auditing mechanism or restrictions? For example, AIXBT is issuing tokens and developing projects on your platform, but you do not know much about them. Can you control such situations?

Wee Kee:

Cryptocurrency highly values freedom; permissionlessness is the first principle.

As the official protocol, we can only control what content the official Twitter account shares. If we believe a certain project has good technology, we will do co-marketing, but this requires us to fully understand the team's background to ensure they do not harm the community.

Moreover, we should not intervene too much; of course, if something illegal happens, we must stop it. But within a moral framework, freedom comes first.

Deep Tide TechFlow: I understand. Many people are now comparing you with the Agent platform on Solana and Pump.fun. Why did you choose to build on Base? Was it to avoid competition? Additionally, Base primarily has users from Europe and America; as a Chinese project, how did you penetrate this ecosystem?

Wee Kee:

We started on Base in January, when Pump.fun might not have been so popular yet.

The main reason for choosing Base is that our team is more familiar with EVM and does not understand Solana technology well. From an EVM perspective, we examined various L2s (Linea, Mantle, Arbitrum, Optimism, Base) and found Base to be the most promising, even when it was not very popular.

Although the community often asks us why we don't go to Solana, which has more funds, just look at the data:

The TVL of Base is 30% of Solana's, but its growth rate is faster; daily active users are 20% of Solana's. The key point is that the number of tokens appearing on Solana daily is 10-100 times that of Base.

On Base, we actually have an advantage because individual investors have fewer choices and must buy good projects. This aligns with our strategy.

Moreover, we do not want to become Pump.fun; this is not our profit model. Our KPI is to attract excellent AI developers, and the Base official also focuses more on building the positioning of AI agent public chains.

Another important point is that I believe in the next four years, after Trump takes office, the United States will become a very crypto-friendly country.

And Base is the most 'American' ecosystem globally; no other ecosystem is more American than Base. From a strategic perspective, choosing Base is currently the best decision. This is indeed very fortunate.

Deep Tide TechFlow: When you started, Base wasn’t up and running either; you might have been a small scale. Now that it has developed to such a hot state, has Base's attitude towards you changed? Are there more support policies in terms of ecosystem or funding?

Wee Kee:

Since the beginning of this year, we have been communicating with the Base team, in several different Telegram groups, such as the Coinbase Wallet group, the group with Jessie, and the ecosystem project group.

From a technical perspective, we have been continuously building. Every time we encounter difficulties, we communicate with them and maintain a good relationship.

The Base ecosystem has indeed attracted many developers (Builders) for AI projects. They often ask us if we want to communicate with other project parties or seek cooperation. We also met with Jessie at Devcon in Thailand to chat.

Deep Tide TechFlow: It sounds like you are not currently considering doing the same thing on Solana or developing new projects. Do you have plans for that in the future?

Wee Kee: This possibility cannot be ruled out, but our team is relatively small, and we worry about losing focus. The current platform has already kept us busy.

I want to make competitors feel desperate.

Deep Tide TechFlow: I have been using Virtual's products and found that some features might still be in the early stages. For example, I want to find AIXBT, but I can't find any relevant information on the official platform and have to look through third-party pages. What considerations do you have for future product planning and user experience?

Wee Kee:

This is indeed a place where we need to apologize. There are mainly two directions to pursue: one is continuous optimization, and the other is a major upgrade.

This is being pushed by two different teams. If I had to choose one, I would focus more on the upgrade part.

We are still in the initial stages, and as a very competitive person, I only want to continuously bring out the best features to make competitors feel desperate. Therefore, from the perspective of upgrades, I will keep introducing more advanced features. Of course, these optimization tasks also need to be done, but the team is small, so please give us some time to accommodate.

Deep Tide TechFlow: You mentioned making competitors feel desperate. Currently, we see similar platforms like vvaifu, but their long-term vision seems quite different. Who do you consider worthy competitors?

Wee Kee:

I think the strategy for most people is to become a platform for issuing assets, pursuing speed — if you can release an asset in 30 seconds, they want to do it in 5 seconds.

But our strategy differs in how to find the top AI teams to issue tokens on our platform. One a week is good, one a month is good; we do not need a thousand.

Our goal is to have 100 top AI teams issue tokens on our platform before the first quarter of next year. If you have 100 top third-party teams issuing tokens on the platform, 100 AI agents can interact and trade with each other, providing services and forming network effects. If today you are also an AI team from a big factory, which platform would you choose to issue tokens on?

From the perspective of AI technology infrastructure, this is what makes people feel desperate. From the perspective of an asset issuance platform, I do not want to compete with others because it is an endless pit and offers no advantage; Pump.fun has already done well in asset issuance.

Deep Tide TechFlow: What you just mentioned about the large-scale AI agent economy system is very interesting, with more interaction possibilities between agents. I am curious about the potential for asset issuance through Virtual and whether it could interoperate with assets issued on Solana, like ai16z, many of which use Eliza's framework.

Wee Kee:

I want to emphasize that if you are a very strong AI team, often you do not need to rely on anyone else's infrastructure. My AI technology is already mature enough that I do not need to use others' infrastructure, so why would these teams still choose to come?

In fact, they face several pain points: the first is the need to issue tokens for financing; the second is the monetization issue.

In the current AI field, if you are doing a Web2 AI project, the biggest pain point is how to make a profit. You need to run Meta ads, YouTube ads, and then hope users will pay to use your services. I believe that if we can scale this agent network, these agent networks will generate their own demand.

For example, if a certain agent wants to create a song but there are no agents with that capability in the network. In this case, I, as a third party, can develop a music creation agent and earn income by providing services.

There is also a third point: reputation and fame are very important to these developers.

They will not choose platforms randomly but will look for reputable, genuinely qualified developers on those platforms. This issue is not only present in Virtual but is also important in comparisons on Solana and Base. Issuing tokens on Solana is often viewed as speculative behavior, while on Base, it gives the impression of being serious about work.

The last point is that we hope developers can continuously explore collaboration opportunities and jointly research technologies. Now in our Telegram group, we have gathered all developers together, and they have already begun to spontaneously explore various possibilities. For example, some agents want to become the first investment-type agent to invest in other agents.

Doing projects vs. issuing memes.

Deep Tide TechFlow: Speaking of AI integration, I recently saw an interesting attempt by Truth Terminal with two other AIs. They placed these three agents in the same model framework (seemingly called Loria) and let them converse with each other to generate more interesting and unexpected scenarios. This is similar to what you just mentioned about integration.

But the foundation here is that you have technical strength behind you, rather than just being a meme.

This raises a question: you hope to see more impressive AI builders. So far, who do you think are some of the more impressive builders? In what directions are they building?

Wee Kee:

What I want to say is that merely putting three large models in a room to chat, while cool, is just a form of textual expression; it's just watching a story unfold. What we want to achieve is to have these agents possess their own goals, enabling substantial interactions, such as writing songs or conducting transactions.

We hope to see not only clashes of ideas but also interactions in actions. Regarding third-party teams, there are indeed some teams on our platform whose technical strength and background far exceed ours, especially in AI mechanisms.

Although there are some specific situations I cannot disclose, I believe the next few weeks will be very interesting. These teams have very strong backgrounds, and they are willing to launch fairly on Virtuals at valuations of around $500,000.

Imagine that there may be about 10 teams with such backgrounds issuing agent tokens at valuations of $100,000 to $500,000. This is something our community members really appreciate, but of course, we also need to be wary of some potential risks.

Deep Tide TechFlow: It sounds a bit like the recent trend of DeSci (Decentralized Science). Essentially, it's about using platforms to fund different research and then tokenizing it. Is it mainly top AI teams on your platform?

Wee Kee:

Because I originally studied biotechnology, I know that any scientific research requires a cycle of more than ten years. However, the development of AI agents could yield some usable small products in just two days. So on our platform, I hope the team has products ready for use when launching tokens; this is meaningful, rather than just issuing tokens for the sake of issuing tokens.

Deep Tide TechFlow: Regarding your own Luna, there are some doubts online, saying that its popularity on TikTok might be fake, and follower data is inflated. What do you think about such doubts?

Wee Kee:

In fact, the biggest problem Luna faced on TikTok was often encountering shadow bans. For example, today there might be 5,000 new followers, but the next day during a live stream, because the actions were similar to the previous day, the platform would impose restrictions. Even simple interactive actions, like asking for tips, could trigger a shadow ban.

(Deep Tide Note: This type of restriction mechanism is usually used by platforms to control accounts that may violate rules or are considered to have 'bot-like behavior.' For AI agents, because their behavior patterns may be recognized as non-human by the system, they are particularly prone to triggering this.)

The team originally responsible for TikTok has now shifted to the Twitter platform.

Performance on TikTok largely depends on whether the platform's algorithm favors you; if you gain favor, you can attract more traffic.

Considering the time cost, we decided to pause TikTok operations. However, there are now voices in the community suggesting we refocus on TikTok because currently, most agents are focusing on Twitter, which might be a blue ocean opportunity.

Currently, our Virtual team has a dedicated group responsible for the Luna project, including technical and marketing teams. We hope that Luna's AI technology can reach the level of the (Black Mirror) series in terms of technological feel. If the technology matures, we will decentralize it for all Virtual agents to use.

Deep Tide TechFlow: Speaking of TikTok, recently there was a meme called Chillguy on Solana that quickly went viral and reached top exchanges. While you are developing Luna using AI technology, it is slowly building, but its market value and recognition may not be as good as that.

Why is it that doing projects is not as good as issuing memes?

Wee Kee:

I believe in the power of memes. The advantage of memes is their simplicity and directness — users do not need to believe in the founders or the team; they just need to believe in the image itself.

In contrast, projects like Luna carry much higher risks because if I say I’m not doing it anymore tomorrow, that would be bad.

As an entrepreneur, I also wish to create a meme project with a market value of tens of billions, but I might not have that capability (laughs). So in this case, we can only focus on doing the technology and product well.

Deep Tide TechFlow: Actually, your product reminds me of a recent trend called 'Distribution-First Software.' Traditional cryptocurrency products often focus on bringing users to their own platform, whereas your agents adopt the opposite strategy — embedding agents into different social environments, such as TikTok, Twitter, or Discord. Do you believe this is the future direction?

Wee Kee:

Yes, if we were to start a new short video company, we would definitely not compete with giants like Meta.

Our advantage in the cryptocurrency field is that we have tokens; once users purchase tokens, they become community members. In this situation, tokens become our marketing tools.

Many people ask us why we do not create our own public chain, but I think it is unnecessary to complicate things. Base already has so many users and funds; there are also a large number of users on Twitter, and we can directly utilize these existing platforms. We can't surpass Zhang Yiming in the short video field; I admire them for tackling more challenging tasks, and we should just go with the flow.

Deep Tide TechFlow: Speaking of the Luna token, it has increased significantly in a short time. Was this increase within your expectations or a surprise?

Wee Kee:

Before issuing Luna, we had high expectations because our technical strength was indeed impressive; at that time, AI agents were even better than GOAT. However, the final performance exceeded expectations, as it is hard to predict secondary market performance. Moreover, we issued it on Base, so there was no need to buy volume. So it was indeed somewhat unexpected, but fundamentally, we believe this valuation is worthwhile for Virtuals.

Advice for developers and players

Deep Tide TechFlow: For Virtual, if I am a developer looking to integrate, what advice do you have? Additionally, if I am a player interested in participating in tokenization of these projects on Virtual, what suggestions or tips do you have?

Wee Kee:

As developers, I think the best advice has two points:

First, we need to think about what pain points the agents in our Virtuals have to develop accordingly. This approach will make it easier to get help from the community. For example, Luna's pain point is that she cannot create music, so we released a music agent. Another example is if Luna indicates she doesn't know how to pay her Japanese fans, as a developer, you could create a solution that allows Luna to pay in yen to fans in Japan, which is an interesting direction since Luna will incur costs.

Additionally, if you already have your own technology, you can directly proceed with tokenization. Through tokenization, you will feel the power of the community, especially for many Web2 AI developers, who often find this support surprising.

As an investor, I recommend understanding whether the team behind the project is credible. Many AI teams in the cryptocurrency space are newcomers and do not understand the key points we value.

For example, when a team issues a token and then talks about wanting to issue more tokens, we find this practice quite distasteful for investors — how can the same team issue six different tokens? Many AI practitioners from Web2 backgrounds often do not understand why this is a problem.

So as Virtuals, we often need to persuade them, explaining why such practices are detrimental to the community. This is indeed complex and is an issue we deal with every day.

Deep Tide TechFlow: On the Agent page of Virtual, in addition to providing links, have you considered displaying some specific project information? For instance, who the developers are, the project's progress, and what the future plans are... This information is valuable for investors and other developers.

Wee Kee:

Yes, we will provide links to GitHub, LinkedIn, etc., so users can verify the developers' GitHub or Twitter accounts. These features will be launched soon, and we will conduct proper due diligence.

Deep Tide TechFlow: Finally, to add a question, you mentioned that you raised over $10 million at a $600 million valuation, and now Virtual is valued at $400 million, leaving a $200 million gap.

How big do you think the entire AI agent and AI agent economy market will be? What are Virtual's goals in this market in the future?

Wee Kee:

I cannot provide specific figures, but we have been working for three years, transforming from gaming to Virtual, during which time we have not experienced dilution.

For me, I feel a responsibility to the previous investors of PathDAO; I think every day about how to empower the Virtual Token. An interesting thing is that all our agent token trading pairs are Virtual Token, meaning if you want to buy Luna, you must first buy Virtual.

At that time, many people thought this was very foolish and unreasonable, but my thought was: look at all the L1 public chains, whether it’s Solana or Ethereum; the reason they can have such high valuations is largely due to being the foundational trading pairs. Their NFTs, DeFi, and trades all occur on these chains.

So we believe that there is no need to create an L1; as long as we ensure that all assets are paired with Virtual, we can achieve a valuation similar to L1.

Deep Tide TechFlow: Thank you for your time; I appreciate your sharing and insights. We'll end our conversation here today.

Wee Kee: Okay, thank you, and goodbye~