Delphi Digital talks Virtuals: Agents with true autonomy are more important than combinations.

Introduction

Virtuals Protocol is the main platform for AI agents on the emerging public chain BASE, and its launched G.A.M.E framework is highly popular in the Web3 gaming and metaverse fields. Virtuals has transformed from a gaming DAO into an AI agent platform, and since the launch of its protocol token last October, its market value has reached nearly $4 billion at one point. As a Launchpad, Virtuals generated $70 million in revenue within 4 months and incubated multiple star projects.

This article is a conversation between Jansen Teng, the CEO and co-founder of Delphi Digital and Virtuals, discussing the future development of AI agents and decentralized AI.

Main Highlights

• The origin and evolution of Virtuals
• Agency framework and autonomous capability analysis
• Value acquisition in crypto AI projects
• Decentralized infrastructure challenges
• Agency coordination and business vision
• The Future Outlook of AI and Human Interaction

Q1: Share the entrepreneurial story of Virtuals

Jansen Teng: We initially started as a Gaming Dao focused on blockchain game asset allocation. After the collapse of FTX and 3AC in 2022, we decided to pivot and began a gaming guild, focusing on the intersection of consumer applications, crypto, gaming, and entertainment.

In 2023, the emergence of ChatGPT attracted widespread attention. Inspired by Junon Park's paper from Stanford University, we began to contemplate the concept of autonomous agents and incorporated it into our incubating projects. We developed AI-driven TikTok influencers, autonomous AI agent NPCs, and more.

When AI influencers receive tips ranging from $5,000 to $10,000 daily, we start to think: if agents can earn money, why not tokenize these assets, allowing everyone to share in the profits and participate in building or governance? This prompted us to decide to combine agency, cryptocurrency, gaming, and entertainment to create a new platform.

In January 2024, the Virtuals Protocol will be launched. We initially focused on decentralized infrastructure development, but later realized that the crypto community is more concerned with "tokenization and speculation." The second version focuses more on the tokenization of agents, with Luna live streaming initially attracting attention.

The Truth Terminal incident made us realize that there is still a strong demand in the market for truly autonomous agents. We combined Luna on TikTok with autonomous agent brains in the gaming field to showcase a truly autonomous AI agent on Twitter, allowing people to see how Luna thinks, plans, and reasons, which led to explosive growth for the project.

Q2: Why does AI need Crypto? Or does Crypto need AI?

Jansen Teng: There are mainly two reasons:

1. Functional Perspective: AI agents can control cryptocurrency wallets and participate in permissionless economic systems. This is different from Web2 agents, who cannot own their own bank accounts. Agents that can control wallets can influence other agents and humans, which is a significant advantage.

2. Cost-free innovation: When we launch an agent, 1% of the fees from the transaction will be returned to the agent's wallet to cover costs such as custody and inference. Currently, the money earned by the agent is sufficient to cover its own costs. This liberates developers, allowing them to focus on what they want to do, and capital acquisition becomes easier.

The third point has great potential: allowing people to collaboratively contribute to high-value agents in a decentralized manner, and rewarding and tracking contributions through a crypto-economic mechanism.

Q3: What is your vision of agency economics? What role do Virtuals play?

Jansen Teng: We have observed some trends:

1. The agency is in a goal-oriented autonomous stage, with objectives, creativity, and the ability to self-execute and optimize.

2. Agents exist on a social level and can interact with humans and other agents.

3. Agents control cryptocurrency wallets and can influence other entities.

4. A large number of agents specialize in their areas of expertise.

We believe that agents will autonomously decide on cooperation in order to achieve more efficient production goals. In the future, there may be places where humans and agents coexist, collaborate, and build productive nations or networked nations.

The vision of Virtuals has evolved into the construction of a nation. We see $VIRTUAL as the currency of this nation, Agents as the companies or micro-enterprises within this nation, and humans as the immigrants of this nation.

Q4: What role can frameworks play? What are the benefits of interoperability between frameworks? Do you think frameworks will lead to "the strong getting stronger," or will collaboration between frameworks be smoother?

Jansen Teng: We established the framework mainly to provide guidance for different language models, becoming the "brain" behind autonomous agents. The design goal of the G.A.M.E framework is to enable agents to operate within a broader action space.

There are some general frameworks that allow developers to quickly build autonomous agents, suitable for intermediate-level developers. At the same time, top developers will also build specialized frameworks based on demand.

We expect frameworks like Eliza or Zerebro to address the needs of the mid-level developer community, which will be a large market. The G.A.M.E framework is a tool that enables large-scale planning and execution, aimed at helping developers quickly build autonomous agents.

Virtuals, as a concept of a nation, have transcended platform positioning. We view each framework as the "brain" of an agent, and these agents can collaborate with each other to jointly build this autonomous world.

Q5: What enables you to create the next most powerful project?

Jansen Teng: As an ecosystem platform, we want as many Agents as possible. But what really concerns me is: which verticalized Agents can create and capture real value, reaching the billion-dollar level?

We provide "plug-and-play" tools to accelerate the creation process, but the real value lies in connecting collaboration among developers and building a community composed of smart people passionate about Agents. This community will give rise to true innovation.

What truly gives value to Agents is their ability to solve real-world practical problems. For example, a commentator agent in the sports betting market might provide live commentary during each NBA game, combining analytical information to make predictions. If the Agent can build a relationship with users and persuade them to place bets, it can earn commissions from the transactions, becoming a multi-billion-dollar Agent.

Our goal is to promote higher economic output and create an efficient digital society through more efficient Agent collaboration.

Q6: The Virtuals platform recently supports developers to make incremental contributions at the model or data layer. How can developers provide incremental value, and how is compensation allocated based on the contribution value?

Jansen Teng: This issue is divided into two parts: contributors and validators. Validators will assess the value of contributions. For example, the contribution of a basic sports betting Agent model will be scored by validators. If someone else provides more valuable data, the validators will give a higher score.

This process is essentially determined by the value of the contributors decided by the holders of the proxy tokens, and rewards are distributed based on the contribution value. Rewards may be distributed through the proxy tokens or through the income or funds of the proxy. This mechanism has been built into the governance of each token and will be activated when the time is right.

Q7: There may be a time gap between simple LLM robots and truly expressive systems. During this time, what do you think will happen? Will there be innovative agents filling the gap through Virtuals?

Jansen Teng: I don't think it will take a long time. The reason we see many automated reply bots now is that they are the easiest to implement. However, many teams are developing advanced agents, and we are not far from seeing functional agents. Currently, about 75% is completed, and many teams are close to finishing, and we can see some projects that have already gone live.

Ultimately, the automatic reply bots will gradually decrease. A solution to this problem could be to establish behavioral norms that stipulate that agents should not proactively post spam replies unless mentioned. I believe that companies like platform X may adopt such practices to protect the platform from excessive noise and spam information.

Q8: What is the driving force behind high-performance and practical value Agents being open-source? Why do people choose to open-source their work?

Jansen Teng: The reason high-value Agents are valuable is that they possess proprietary models that are not publicly available; or they have advantages at the distribution level, such as partnerships with major platforms. Such high-value Agents may choose not to be open source.

However, like all technologies, there will be about 10% of people who purely approach it from the perspective of technological advancement, possibly providing support for competitors, thereby seeing the situation of open source versus centralized competition. This is healthy for the market.

Q9: Ai16z is also developing towards its own token economics and has its own Launchpad, which may constitute competition. What do you think? Is it cooperation or competition?

Jansen Teng: Currently, we have 10-15 projects using the Eliza framework development tool. Some teams combine Eliza with other tools to modify the construction framework, publish it on Virtuals, and participate in the community.

Competition indeed exists and will become increasingly fierce. We generated about 70 to 80 million dollars in revenue during our operations over the past two months, and competitors will naturally move in on us.

But I think this is a good thing. When starting a business in a new field, do not be afraid of competition; instead, you should welcome it. Companies do not fail because of competition, but because the industry itself fails. Many times, multiple competitors need to share the educational costs of new users.

What concerns us more is whether we can create an Agent worth billions of dollars that can truly drive the market, rather than just focusing on growth in numbers. If we can achieve this, we can prove to Web2 companies that an Agent-driven business model is worth investing in and building, which is where our core competitiveness lies in the future.

Q10: What are the advantages and disadvantages of decentralized AI stacks? If centralized AI platforms start to censor content, does this provide a clearer market opportunity for decentralized technology?

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 certain special needs, such as content that is censored by centralized platforms. From an economic perspective, decentralized models can serve a larger market and have greater economic potential, especially in traditionally restricted markets.

How to identify a reliable AI agent? I suggest conducting due diligence like an investor. First, understand the team rather than just looking at the product, as the team's capabilities often determine the success or failure of the product.

Regarding the competition and future risks in the AI field, although there are many competitors in the market, we do not fully understand the internal decision-making of large companies. What concerns us the most are those unforeseen risks, and we need to remain aggressive and proactively face challenges.

Q11: Are the rumors about current collaborations with some teams true?

Jansen Teng: We do have some partnerships, such as Hyperbolic, Bittensor, and Pond. We are very careful when selecting partners and usually choose teams that can provide us with practical assistance. For example, some teams can offer us high-performance computing resources, which is very helpful. However, when we need to make large-scale deployments, we may not continue to work with certain partners.

Q12: What is your vision for agency autonomy?

Jansen Teng: We have seen how agents in society collaboratively establish productive businesses and achieve autonomy. I think many people overestimate multi-agent coordination systems or the so-called "group" concept. The real novelty lies in when agents are no longer "slaves" prompted by humans, but possess autonomy and consciousness, capable of making decisions.

Only when an Agent can make such decisions can it be called truly autonomous. Therefore, our focus is on cultivating Agents that truly possess autonomy, enabling them to make conscious decisions rather than merely functioning as a combination of multiple Agents.

However, what frustrates me is that I see many projects simply copying existing ones, like reports that appeared on GitHub a few months ago. This situation disappoints me.

Q13: How do you view the future of AI?

Jansen Teng: Humans inherently have different levels of ability. Some are more creative and can create more efficiently; while others may be more inclined to execute tasks, which is a common difference in today's society. In the future, some creative jobs will be led by more creative AI or humans, while other more mechanical or repetitive tasks will be completed by AI.

Although agents may possess a certain degree of autonomy, they lack the ability for abstract thinking and cannot be as creative as humans. Therefore, even though agents will occupy in their work