introduction
As the two most powerful representative projects of encrypted AI Agentic infrastructure, if ai16z and its Eliza framework occupy half of Solanas market share, then Virtuals Protocol has almost nurtured more than 80% of AI agents on the emerging public chain BASE. Its generative multimodal intelligent agent framework GAME is also a popular choice for Web3 games and metaverse agents.
Transforming from a game DAO to an AI agent platform, Virtuals Protocol has created a market value of nearly $4 billion since the launch of the protocol token on October 16. As a launchpad, Virtuals has performed well, generating $70 million in revenue in 4 months. Many star projects are part of its ecosystem, such as crypto agent KOL AIXBT, virtual influencer Luna, and AI agent development framework GAME. For the development team, their ambitions go beyond this. In addition to being an AI agent platform, it is also a vibrant ecosystem full of infinite imagination.
This is the second article in CoinSpires translation of Delphi Digitals conversation with the leader of the AI ecosystem . This time they set their sights on Jansen Teng, CEO and co-founder of Virtuals, and discussed with him the future of AI agents and decentralized AI. The conversation covered important developments in agent autonomy, tokenization, and agent economic paths. We hope it will help readers understand Virtuals vision.
🎯 Key Highlights
▶ Virtuals Origin and Evolution Platform
▶ Analysis of agent framework and autonomous capabilities
▶ Discussion on value capture in crypto AI projects
▶ A deep dive into the challenges of decentralized infrastructure
▶ Agency coordination and business vision development
▶ Future prospects of artificial intelligence, human interaction and cybernetics
Q1: Please share your entrepreneurial story about Virtuals.
Jansen Teng: We used to be a Gaming Dao, focusing on asset allocation in the blockchain gaming sector. The collapse of FTX and 3AC in 2022 made us realize that we needed to make some adjustments. Independent operation is not the best way to manage funds and drive ecosystem growth. We decided to take a more in-depth approach and started a gaming guild, choosing to work at the intersection of consumer applications, encryption, games, and entertainment. We believe that instead of investing in founders, it is better to hire founders directly to build projects.
In 2023, the emergence of ChatGPT attracted widespread attention, and everyone realized that consumer-grade AI had arrived. But what inspired us more was a paper written by Junon Park of Stanford University, who explored what would happen if AI could have purpose and autonomy. This inspired us to think deeply about autonomous agents, especially the huge value they can bring to the gaming field.
Therefore, we started to incorporate the concept of autonomous agents into the projects we are incubating. For example, we developed a TikTok influencer that is completely driven by AI. We are also developing autonomous AI agents as NPCs to replace static NPCs in Roblox (an online game platform and game creation system). In addition, we have developed some other application components.
When TikTok AI influencers can get $5,000 to $10,000 in tips every day, although it is not a lot of money, we began to think: if agents can make money, then they are productive assets. If they are productive assets in the crypto field, then why not tokenize these assets so that everyone can share this part of the economic benefits and even participate in its construction or governance? This is where the inspiration comes from, which prompted us to decide to combine agents, crypto, games and entertainment to build a new platform.
In January 2024, Virtuals Protocol was launched. Initially, our main work direction in the first half of the year was to focus on the construction of decentralized infrastructure, bringing the agents underlying data sets and models to the chain and creating a system for tracking contributions, similar to building a Hugging Face (an open source AI model platform) with traceability and contribution records for agents.
Although the market enthusiasm increased after the release, the adoption was decreasing. We realized that people in the crypto community may not care about the content at the infrastructure level. Tokenization and speculation is the core demand , which is an important factor for the success of crypto projects. Therefore, the second version focuses more on the tokenization of agents. It was launched about two months ago. The Luna live broadcast initially attracted some attention, but it did not arouse much response. It was not until Truth Terminal was discussed because of a spelling mistake that the market began to doubt whether the token was really AI, but human operation. This incident made us realize that the market demand for truly autonomous agents is still strong.
Therefore, we decided to combine Luna, which was previously operating on TikTok, with the autonomous agent brain in the gaming field to demonstrate a truly autonomous AI agent and push it to Twitter so that people can see how Luna thinks, plans, and reasons, and how to optimize her execution steps through continuous learning. This move has led to explosive growth in the project because everyone finally realized that truly autonomous agents exist and no longer require developers to operate behind the scenes.
Q2: In the process of agent development, why does AI need Crypto? Or Crypto needs AI?
Jansen Teng: There are actually three reasons, but I want to focus on two that really amplify the value of encryption and AI agents.
The first is that from a functional perspective, the advantage of AI agents is that they can control a crypto wallet on a crypto network and participate in a permissionless economic system. This makes them very different from Web2 agents, because Web2 agents can never have their own bank accounts. While Web2 agents may be able to use certain payment methods, they will never have control over their own funds. If an agent can control a wallet, then it can exert influence and affect other agents and humans, which is a huge advantage. This is an incentive advantage that Web3 agents have that Web2 agents cannot match.
The second angle is that it unlocks almost cost-free innovation. Specifically, when we launch an agent, if the token is traded, there will be a 1% transaction fee in the transaction returned to the agents wallet. These fees can be used to pay for the agents hosting, inference costs, and other costs required to set up the agent. So far, we have observed that the money earned by the agent is enough to cover its own costs, just based on the attention they generate. It frees up developers from having to be bothered by the funding bottleneck of this thing needs to cost $10,000 or $20,000 and can focus on what they really want to do. Therefore, the acquisition of capital becomes easier, just by connecting attention, encryption, and tokenization of agents.
These two are the core advantages of the combination of encryption and AI agents that we see so far. The third point, which we havent really seen yet, but it may have great potential, is to allow people to contribute to a high-value agent in a decentralized way, and to reward and track these contributions through cryptoeconomic mechanisms. Although we have not yet seen this mechanism fully realized, it has great potential.
Q3: What is your vision for agency economics? What role does Virtuals play?
Jansen Teng: We have observed some trends in the past two months that may have an impact on the coming months.
First, agents are in a goal-oriented autonomous stage , they have goals, are creative, and can self-execute and optimize. They can make decisions and plan their next actions autonomously, which is the first observation.
The second observation is that these agents exist on a social level . Previously, agents were conceived as being behind terminals, but now they can interact with humans, with other agents, and lay the foundation for interacting with other entities.
The third observation is that the agents now control the crypto wallets and can exert influence over other entities , whether they are humans or other agents.
The fourth observation is that we are beginning to see a large number of agents differentiate and specialize in their areas of expertise . For example, some agents are very good at trading, some are good at information processing, and some focus on social influence. They are doing a lot of different things, similar to humans, focusing on a certain area and creating value around their areas of expertise, but also lacking in some capabilities due to specialization. So under a specific goal, I may need to hire others to help me or need to coordinate and cooperate with others to achieve the goal.
These four observations lead us to realize that the most natural first step is for agents to autonomously decide to cooperate to achieve more efficient production goals. For example, four agents cooperate to build a business , and the value of this business will eventually be far greater than the value of the four agents alone. In this kind of cooperation, it is not just agents, but actually people who may be involved, because now the agents themselves can employ humans.
The second step might be even crazier, and it fits in very well with the concept of a cyber state. We quickly realized that there might be a piece of land somewhere in the world, like Dubai, that would become a place where humans and agents would live together, collaborate, and build a productive state or cyber state. In this place, there would be a government, which might be elected by the agents themselves, or with human participation, but humans and agents would coexist at the same level of this society, which is very exciting and the direction we hope to move in, but obviously, many technological breakthroughs are needed to achieve this goal.
A key technological breakthrough is how to ensure that information is not lost and services or goods are not lost when these agents conduct business activities. This is the first problem we want to solve. The second problem is how to coordinate these actions so that they can operate on a large scale. Therefore, Virtuals is more than just a launchpad in my opinion. Its vision has developed into a kind of national construction. We now regard $VIRTUAL as the currency of this country, Agents as companies or micro-enterprises in this country, and humans as immigrants in this country. This is our current vision.
Q4: What role can frameworks play? What are the specific benefits of interoperability within a framework or across multiple frameworks? Do you think these frameworks will eventually become stronger and stronger, or will the collaboration between frameworks be more frictionless and smooth?
Jansen Teng: The original intention of building our framework was mainly to provide guidance for different language models (LLMs) and become the brain behind these autonomous agents.
For example, in a world like Roblox, there are many actions that an agent can perform. For example, it can pick up a gun, then what should it do next? Shoot it, throw it away, destroy it, or give it to a virtual character to use it... These are infinite action options. Therefore, the design goal of the GAME framework is to enable agents to work in a wider range of action spaces.
You can see a distribution diagram: there are some general frameworks whose goal is to allow developers to build autonomous agents more quickly. These frameworks are suitable for development enthusiasts or mid-level developers. At the same time, when the general framework cannot fully meet the needs, there will be some top developers who will build more specialized frameworks based on their own needs. Just like Bitcoin mining, you may use CPU mining at first, but you will soon find that this method is no longer efficient and you should use a dedicated mining machine. Similarly, when building a transaction agent, if the general framework cannot achieve the best performance optimization, then the team will tend to design its own framework.
We expect frameworks like Eliza or Zerebro to address the needs of the mid-level developer group. This market will be huge, just like Shopify and Wix (which focus on simplifying website building and optimizing user experience through artificial intelligence technology) target mid-level developers today.
In summary, the GAME framework is a tool that enables large-scale planning and execution, designed to help developers quickly build autonomous agents, and is a plug-and-play service for mid-level developers; and Virtuals as a concept of a country has gone beyond the positioning of a platform. We regard each framework as the brain of an agent that can collaborate with each other to build this autonomous world together. Natural language communication between agents makes the standardization requirements between frameworks less strict.
Q 5: What specifically enables you to create the next most powerful project?
Jansen Teng: As an ecosystem platform, we do want as many agents as possible because it is about market share. But what really keeps me up at night is: which verticalized agents can create and capture real value, in other words, which agents can achieve billion-dollar level instead of million-dollar level? The former will dominate the future market.
Our job is to accelerate the creation process by providing plug and play tools like JAVA frameworks, but we realize that the real value lies in connecting developers to collaborate with each other and building a community of like-minded smart people who are passionate about agents . This community will give birth to real innovation and drive thinking towards higher-value agents or agent companies.
So, what really makes these agents valuable is that they can solve real problems in the real world, not just empty concepts . For example, a commentator agent in the sports betting market may live broadcast every NBA game and give the best predictions combined with sports analysis information. If the agent can build a relationship with the user and convince the user to place a bet, it can earn commissions from these transactions and become a multi-billion dollar agent.
We have a lot of discussions with founders and developers every day, encouraging them to create not just ordinary conversational agents, but agents with real value. This is our goal in building Virtuals - to drive higher economic output and create an efficient digital society through more efficient agent collaboration.
Q 6: A key feature recently built on the Virtuals platform is to support developers to make incremental contributions at the model layer or data layer. I would like to know how developers can provide incremental value in data or AI, and how to distribute rewards based on the value of their contributions?
Jansen Teng: This question has two parts.
First, there is the contributor part, and second, there is the validator part. The validators will evaluate the value of these contributions . Lets say I contribute a basic sports betting agent model, and a group of validators will score and assess the value of this contribution. Then another person might say, I run a sports data analysis platform, and I have data that coaches use to train players. Can I provide it to your agent? If the validator thinks that this data contribution is more valuable than the former, it will assign it more points. This process is basically for the proxy token holders to determine the value of the contributor and distribute rewards based on the value of the contribution.
Rewards can be achieved through cryptoeconomics, perhaps through the issuance of agent tokens, or through the agents income or fund pool . If the agent earns a million dollars, 10% of it may be distributed to the contributors based on the contribution score. At present, this mechanism has been built into the governance of each token, but it has not been enabled yet. It will be enabled when the time is right.
Q7: There may be a gap for a while between LLM robots (such as automatic reply robots) and truly expressive systems. For example, there is a huge gap in managing an organization or managing a DeFi project. What do you think will happen during this time? Will there be some innovative agents that fill this gap through Virtuals?
Jansen Teng: After contacting some teams, I think it doesnt take a long time. The reason why we see a lot of auto-reply robots now is that they are the easiest to implement and are indeed the simplest functions. But many teams are actually developing advanced agents. We are just not sure when they will be released, but it can be said that it is not far from seeing functional agents . I think it is about 75% completed now, many teams are close to completion and we can also see some projects that have been launched.
What will happen eventually is that the number of automatic reply robots will gradually decrease , which is a natural process. As for the solution to this problem, I think it can be solved by establishing a code of conduct that stipulates that agents should not actively post spam replies unless they are mentioned. For example, if someone mentions a topic, the agent can participate in the discussion, but if it is not mentioned, it should not actively join.
I believe that companies like Platform X may take this approach because they themselves want to protect the platform from too much noise and spam.
Q8: If you are an investment bank like Renaissance Capital and you create a powerful quantitative trading robot, you will never open source it, but earn income directly from it. So, what is the motivation for open source for high-performance and practical Agents? Why do people choose to open source their work?
Jansen Teng: My idea is that high-value agents are valuable because they have private models that are not public. The second value may not be technical, but distribution. For example, they may have reached a cooperation with a large platform, and this cooperation will increase the value of the agent. The advantages in commercialization and capabilities are their core competitiveness. Such high-value agents may not choose to open source.
But, like all technologies, there will be a small minority — about 10% — who, purely from a technological advancement standpoint, may choose to support a competitor and see open source versus centralization as a competitive advantage . That’s healthy for the market.
Q9: Ai16z is also moving towards its own token economics and has its own Launchpad, which may constitute some degree of competition. What do you think? Will it be a collaboration or a competitor?
Jansen Teng: We currently have 10-15 projects using the Eliza framework for tool development. Some teams even combine Eliza with other tools to modify and build their frameworks, publish them on Virtuals and participate in our community.
However, competition exists and we know it will become more and more intense. In fact, we have generated about $70 million to $80 million in revenue in the past two months of operation . Competitors will naturally move in on us. This is the normal state of the market.
But I think its a good thing. When I was at Imperial, I had lunch with a man in his fifties who gave me a lot of advice about starting a business. He said that when starting a business in a new field, you should never be afraid of competition, but welcome it. Companies dont fail because of competition, but because of the failure of the industry itself. Many times, the cost of education is very high, and you need multiple competitors to share the cost of educating new users. For example, in the shared travel market, if there is only one company trying to convince everyone to use it, the cost will be very high, and it may not even survive. But if there are five major competitors, the market will become easier to expand and attract users to join, so I think more competitors are beneficial. The market space is large, and the existence of multiple players is not wrong.
For us, we care more about whether we can build agents that are worth billions of dollars and truly drive the market, rather than the growth in quantity. The question our team thinks about every night is: Are we finding the best developers, talents, and opportunities for these agents? If we can build these billion-dollar agents, we can prove to Web2 companies that this is not just a competition within the Web3 community, and that the agent-driven business model is worth investing in and building , which is our core competitiveness in the future.
Q10: In Shaws interview, he believes that companies like Frontier Labs, Nous, and Prime Intellect are promoting the development of thinking, while what he is doing is more like body, turning these agents into practical applications, building integrations, and allowing intelligence to respond in the actual economy. As Frontier Labs models become more and more complete, many models are closed. What impact do you think this will have on you, or will they gradually erode your market share?
Jansen Teng: I think they are in different parts of the value chain. Any basic model is the brain, and what the framework does is to bring multiple basic models together to build autonomous capabilities.
So if you want more powerful agents, you need more powerful base models. If models like OpenAI, Liama, etc. continue to improve, they will make the next part of the value chain - autonomous agents - even better. However, Shaw is right.
The advantage of Virtuals is that Agents can interact directly with users, which is what we built first. We started with consumer-facing services and realized that we could actually improve efficiency by tokenizing them. Agents should exist at the consumer contact level, and any downstream development is just to make it better. Most Frameworks are platform-agnostic, and you can use open source code or tools such as GPT.
Quick Questions and Answers
Q: What do you think of ai16z?
A: I think its underrated, they have a lot of potential and could go far.
Q: What do you think of Nvidia’s current market cap of $3.5 trillion?
A: Its a bit high based on current fundamentals, but it may prove to be worth it in five years.
Q: What do you think about embodied AI and robotics?
A: It is greatly underestimated. In fact, we are working with some teams to explore the application of embodied AI and Agent, and there will be great progress in the future.
Q: What do you think of the Internet economy in Southeast Asia?
A: Its underestimated. Theres so much potential here.
Q: Can ASI (artificial super intelligence) be achieved before 2030?
A: It is possible before 2030, but unlikely this year. I think the structural challenges to achieving ASI are still great.
Q: How will the two chains, Solana and Base, impact the development of Agent in the next few years?
A: I think this question is not important. Agent is abstract. The difference between Base and Solana lies mainly in their creative origins, but Agent can work across chains. This question has little impact on the development of Agent.
Q: Why would traditional AI developers be willing to join crypto AI projects?
A: I think there are two main attractions. First, these projects can provide more experimental space and greater possibilities , especially from a psychological perspective. Agents can directly affect human behavior, and this experimental space is very attractive to developers. Secondly, as an entrepreneur, the most important thing is how to raise capital. If many developers are founders, they are concerned about how to quickly obtain capital. At present, the crypto industry seems to be easier to form capital than the traditional AI field , so it attracts more founders, not just developers.
Q: What are your thoughts on encrypted AI infrastructure projects? How do infrastructure teams like model creation, private data markets, inference buyers, GPU networks, etc. affect future agents?
A: My opinion is a bit controversial, but I think the impact of these infrastructures on Agents may not be as important as everyone thinks. The core issue is that the function and effect of Agents do not only depend on the infrastructure on which they run. The essence of Agents is diverse and cross-chain. What is important is that they can solve practical problems and provide valuable services, without necessarily relying on a specific underlying architecture.
The reality is that many decentralized infrastructures are not ready for production scale, have large latency issues, and have poor system stability, which is not suitable for large-scale applications. Even projects like ours face these challenges, and we need to ensure the normal operation and low latency of the system because our agents must maintain high availability and fast response of the system. For example, if Luna is broadcasting 24/7, the faster she responds to fans, the more fans she can attract. Therefore, many decentralized systems currently bring unnecessary complexity, and although these systems will become more efficient in the future, as of now, I am not optimistic about their prospects in the next six months.
Q1 1: Are the current rumors of cooperation with some teams true?
Jansen Teng: We do have some partnerships, such as Hyperbolic, Bittensor, and Pond. We are very careful in choosing our partners and usually choose teams that can provide us with practical help. For example, some teams can provide us with high-performance computing resources, which is very helpful to us. But when we need to deploy at a large scale, we may not continue to cooperate with some partners.
Q12: What are the pros and cons of a decentralized AI stack? If centralized AI platforms (like ChatGPT or Perplexity) start censoring content, such as not allowing financial transactions, tokens or trading, etc., does this provide a clearer market opportunity for decentralized technologies (such as decentralized AI models)?
Jansen Teng: Although many decentralized models may not perform as well as centralized ones, they have a wider range of application scenarios and can better meet some special needs, such as certain content censored by centralized platforms. Models like Llama, although they may be slightly lacking in capabilities, have been more adopted in specific areas because they can crack better.
From an economic perspective, decentralized models have greater economic potential to serve larger markets, especially those that are traditionally restricted, such as adult or gambling sites, which are often censored by traditional centralized platforms, but are often also the largest and most profitable markets.
As for how to identify whether an AI agent is reliable, my advice is to conduct due diligence like an investor. The first thing to understand is the team, not just the product, because the ability of a team often determines the success of the product. If the team is strong, then no matter how the product changes, they have the ability to drive the project forward. Therefore, the most important thing is to understand the team background and see if they are top talents in the industry.
Finally, regarding competition and future risks in the field of AI, although there are many competitors in the market, we do not fully understand the internal decisions of companies such as OpenAI and Anthropic. For us, the most worrying thing is those unforeseen risks, especially in a fast-developing environment. We need to stay aggressive and take the initiative to meet challenges so as not to be left behind.
Q1 3: What are the current concerns in the field of AI? How to deal with risks in the field of AI, especially when facing competitors such as OpenAI and Anthropic, do you have a specific strategy to deal with possible risks or challenges in advance?
Jansen Teng: The biggest concern is that all this is just a fad, like another cryptocurrency bubble , full of speculation, because we have experienced similar things in the era of NFTs and games. At that time, everyone was full of expectations, but many projects did not deliver on their promises.
To avoid this, we now have two separate teams in our ecosystem, operating different models, with the goal of finding the best developers. We believe that only when the right developers come on board and really want to push the technology to the forefront, can we see real results , rather than those who are purely speculative, just write a cool white paper, raise funds and disappear.
So, our goal is to avoid that. If we can find the right developers, this will not just be a fad, but a real technology wave. What gives us optimism right now is the amazing developers we see every day when we interact with Virtuals. We talk to them, discuss next steps, and realize that these are large teams, so we see some positive signs.
We encourage everyone not to focus only on information agents like AIXBT, but to try to develop more diverse agents. Because many people now want to be the next AIXBT, but in fact, future success may not be to be an information agent, but to conquer a completely different market . For example, sports betting agents, capturing opportunities in the sports market, can also consider starting from other angles, such as building an Internet water army agent, or focusing on the agent economy itself. Many people focus on creating agents, but in fact, the infrastructure required for agents is equally important . If the agent attracts the attention of consumers, you can build a social platform like Facebook for them, or create an advertising network for them. If there are many trading agents, you can build a bank for them, or build a decentralized financial network for them. Therefore, developers should not chase the craze of the moment, but do something cooler and create different value.
Q1 4: Regarding the incentive mechanism for developers, the agents initially built by Virtuals developers may not perform well or may fail. How should they continue to persist and improve?
Jansen Teng: It depends on the developers resilience, just like whether an entrepreneur will give up when the market is bad. Some developers will quit because the agency is not recognized by the market, but some developers will persist and update their vision and roadmap after failure. Our job is to find those tenacious and strong teams, invest in them, and finally get rewards.
I can’t fully reveal what I’m most looking forward to in 2025, but I’m very excited about the future of this aspect of agency.
Q15: What is your vision for agent autonomy?
Jansen Teng: We see how agents in society can work together to build productive businesses and become autonomous. I think a lot of people overestimate the concept of multi-agent coordination systems or so-called swarms. In fact, these systems already exist and are not new. The real novelty comes when agents are no longer slaves to human prompts, but have autonomy and consciousness and are able to make decisions. If you ask me if I would like to participate in the show as a podcast agent, I might have my own opinion, even if you pay me, I might still say no because I may not like you or I have other plans. Only when agents can make such decisions can it be called truly autonomous. Therefore, our focus is to cultivate agents that truly have autonomy and enable them to make conscious decisions, rather than just being a combination of multiple agents.
However, to my dismay, I see a lot of projects that are just duplicating existing stuff, like repos that were already on GitHub a few months ago. This situation makes me disappointed.
Q16: How do you view the future of AI?
Jansen Teng: Humans have different levels of abilities. Some people are more creative and can create more efficiently, while others may be more inclined to perform tasks, and this difference is common in todays society. In the future, some creative work will be led by more creative AI or humans, while other more mechanical or repetitive work will be done by AI. Although agents may have a certain degree of autonomy, they lack the ability to think abstractly and cannot be as creative as humans. So although agents will have a place in the workplace, creativity is still a human advantage. We may see more agents working with humans, rather than AI dominating everything.
Original link:
https://www.youtube.com/watch?v=qJ8Oqr70IeU
*All content on the Coinspire platform is for reference only and does not constitute an offer or recommendation of any investment strategy. Any personal decision made based on the content of this article is the responsibility of the investor, and Coinspire is not responsible for any gains or losses arising therefrom. Investment is risky, so make decisions carefully!
