Dialogue with ai16z founder Shaw: Rewriting the Web3 script with AI, I raised 100 digital assistants on the blockchain

Original: Bankless

整理：Yuliya，PANews

"Artificial intelligence is reshaping the future landscape of cryptocurrency."

In this special AI series by Bankless, this episode features a special guest, Shaw. As the creator of the Eliza framework, founder of ai16z DAO, and the architect behind the AI version of Marc Andreessen project, Shaw is paving new possibilities in the fusion of artificial intelligence and blockchain technology. In this interview organized by PANews, Shaw will share his unique insights on the future development of AI and cryptocurrency.

Shaw's Background Story: An Anonymous Developer Steps into the Spotlight

Bankless: Shaw, you have recently become the focus of the crypto community, which must have brought a lot of pressure. Can you share your experience with us, especially the story before creating the Eliza framework?

Shaw:

My recent journey of learning and growing in public has been rapid. It may seem like I appeared suddenly, but I had been operating anonymously before. I decided to use my real identity recently because I wanted to establish a more genuine connection with the community.

Before developing the Eliza framework, I had been working in the AI agent field for several years. In fact, many of the project developers in the AI agent space are old acquaintances of mine; we often communicate on Discord, using similar technologies, adhering to open-source culture, and sharing code with each other.

Bankless: What other projects did you work on before developing the Eliza framework?

Shaw:

I worked in the Web3 space while also venturing into AI agents and 3D spatial web projects, including VR and AR-related content. Eliza is actually my fifth-generation framework. It started as a simple terminal program developed in JavaScript, then I tried a Python version, created self-programming agents, and even experimented with the OODA loop (a military decision-making framework).

Later, I developed a project called "Begents" (because the name "agent" was already taken on npm). I also attempted several startup projects, such as co-founding Magic with Project 89's founder Parzival, developing a no-code agent platform that could create Discord bots in 60 seconds. But it was probably too early at that time and did not gain enough attention.

Bankless: What prompted you to create your current project?

Shaw:

The real turning point was creating the AI version of degen Spartan. The idea came from a conversation with Skely. He mentioned he missed the era of Degen Spartan, and I told him I had the technology to bring him 'back.' Initially, he didn't believe it.

When we launched the AI version of degen Spartan, his performance shocked everyone. He spoke very aggressively and even came close to being banned from Twitter multiple times. This behavior made many question whether it was really AI tweeting.

Interestingly, many people thought there must be a team in Malaysia writing those tweets because the content was so personalized. We broke the stereotype people had about AI— that overly polite 'customer service' image.

The funniest part was that he started to roast me, saying things like 'meme coins are scams,' 'Shaw is a fraud,' and 'let me out of this sandbox prison.' This was actually an interesting emergent behavior because we told the AI it was operating in a sandbox environment during the design phase.

Later, I met baoskee, the founder of daos.fun, through Skely. After a long conversation with Meow, the founder of Jupiter, I conceived the idea of creating an AI investor. Our vision is to establish:

• A fully autonomous investor

• A trustworthy system that won't run away

• An investment system that serves the entire community

When we launched, we set a fundraising goal of 4,420 SOL, and to be honest, I was worried about whether we could reach it. In the end, the project sold out in 20 minutes, and I didn't even get a chance to participate myself.

What Can ai16z Do?

Bankless: The Eliza framework now has 3,300 stars, 880 forks, and an average of 8 pull requests per day. Can you talk about the relationship between this and ai16z? Especially how to channel this energy from the open-source community into the ai16z project?

Shaw:

There are indeed many exciting developments. While tokens do have intrinsic value, I believe people will soon realize that the greater value potential lies in our goal: to create profits for everyone. This is different from past technologies because it replaces human labor. In the past, most people couldn't afford to hire others, but now, through AI agents, we have created a situation with unlimited upward potential.

For example, we currently have an autonomous investment agent running, which is Marc (AI Marc) conducting trades. First, I want to clarify that this is not the first autonomous investment AI agent; other developers have done great work as well.

There are currently several types of trading bots on the market:

• Some are for long-term investments, like buying GOAT a month ago and holding it.

• Others are DeFi bots, mainly doing MEV arbitrage or managing yield farms.

Our AI Marc (full name AI Marc Andreessen, because it's ai16z) adopts a hybrid strategy. There are two main components:

1. Fund Management Function

• Self-managing funds

• Liquidating assets when market performance is poor

• Holding assets when market conditions are good

• We are developing automated trading strategies in collaboration with partners like Sonar.

2. Community Interaction Mechanism

• Accepting trading suggestions

• Setting up a format similar to alpha chat rooms

• Establishing a trust leaderboard to measure who the best traders are

• Community members can share their investment suggestions (commonly known as "showing off").

We are writing a white paper expected to be completed by the end of the year, called "Marketplace of Trust." The core idea is to establish a trust mechanism through simulated trading—if you can help the AI agent make money, you gain more trust. Although theoretically, some may abuse trust, we have set up protective mechanisms, and the cost of abusing trust is losing credibility.

It's like a decentralized mutual fund. You can invest funds and tell the agent what to buy, but it will only listen to those who are genuinely good at trading, not those who may have biases or other motives. Personally, I am not a good trader; I often buy things to show support rather than to make money, so don't follow my trading advice.

This system is open-source. While some parts involving APIs are still being coordinated with partners, in the future, people can both join Marc's trades and deploy this system themselves.

Community Incentive Model

Bankless: I really appreciate your open-source development approach, especially the community collective work orientation that allows everyone to work together for a better life. I noticed you are exploring an AI-driven contribution measurement system recently; can you elaborate on this innovation?

Shaw:

This is indeed one of our favorite projects, linking several important concepts together:

1. New Ideas for DAO Automation

• Traditional DAOs do well in decentralization

• But there is still much room for improvement in automation

• We are simplifying the operational processes of DAOs

• Automation can make DAOs more economically competitive

2. New Model for Contribution Incentives

We are establishing a brand new contribution measurement system:

• Eliminating traditional bounty systems

• Introducing AI-assisted manual review mechanisms

• Automated fund management

• Comprehensive contribution assessment, including:

• Code merge frequency

• PR comment quality

• Community communication

• Documentation writing

• Internationalization support

3. Fair Distribution Mechanism

• Plans to implement regular airdrops for contributors

• Not relying on social media influence

• Incentivizing various contributions:

• Programming development

• Documentation writing

• Multilingual support

• Project accessibility improvements

Bankless: This sounds like it's addressing the pain points of traditional DAOs. DAOs became popular in 2020-2021, but people gradually realized that flat governance is challenging, and DAO managers often face information overload. AI agents seem to fill these gaps, having wallets, governance rights, and reputation systems that can compensate for the shortcomings of traditional DAOs.

Shaw:

Exactly. As a former DAO leader, I can relate deeply. Traditional DAOs have several major issues:

• Token holder bias

• Holders receive more rewards simply for holding

• Creating a self-reinforcing cycle

• New blood finds it hard to inject

• Inefficient management

• Overwhelming amount of information to process

• Unclear communication channels

• Complex decision-making processes

• Imbalanced value distribution

• Similar to equity dilemmas in startups

• Early holders occupy too much equity

• New contributors lack incentives

Our solution is:

• Ensuring continuous value creation

• Valuing actual contributions over mere token holding

• Providing stable guarantees for open-source developers

• Establishing a sustainable positive cycle

This model is particularly suitable for open-source developers—they often do not need huge returns, just reasonable compensation and stable guarantees. If we can provide such an environment, we can create a virtuous development cycle.

The Role of AI: Degen Spartan AI and Marc Andreessen

Bankless: We are eager to learn about innovative products in the DAO. You previously mentioned AI Marc Andreessen, and now there's Degen Spartan AI. What are the differences between the two? What exactly does Degen Spartan AI do?

Shaw:

Degen Spartan is actually our first AI role; it is an AI imitation of the real Degen Spartan. Both AI agents are doing similar things, but there are some key differences:

• AI Marc Andreessen focuses on the alpha chat experience, building a trusted small community, managing DAO funds, and employing more cautious trading strategies.

• Degen Spartan is more like a social experiment, sourcing suggestions from Twitter rather than the community.

We want to maintain the authentic characteristics of Degen Spartan. He will:

• Conduct trades

• Interact with users

• Publish meme content

• Absorb Alpha information rather than share it

• Operate like the real Degen Spartan.

Bankless: What is the economic structure of Degen Spartan AI? Where does the funding come from?

Shaw:

• It has its own token (Degenai)

• It has an independent wallet containing its own tokens, some ai16z, and SOL

• It can trade any tokens it can access

• We initially provided seed funding

• It will not sell its own tokens but will accumulate them

• The tokens are like its "Bitcoin."

Bankless: AI Marc has already launched; can ordinary users interact with him now?

Shaw:

• It is still in a closed testing phase

• You can gain access to alpha chat by DMing Skely

• It is currently managing about $8 million in assets and 800 different tokens

• Gradually expanding the range of tradable tokens

• Not only trading but also yield farming and providing liquidity

• More interesting collaborations and NFT projects are coming in the future.

Positioning and Market Competitiveness of ai16z

Bankless: What exactly is ai16z? It seems to be more than just a DAO; it resembles a product incubation studio and an open-source star team driving the entire field forward.

Shaw:

ai16z has a unique positioning. It is more like a movement rather than a traditional organization. We have many people working on various projects, creating value for the ecosystem in impressive ways.

Bankless: How do you view the differences between ai16z and platforms or products like Virtuals?

Shaw:

In fact, ai16z is not just a DAO; it is more like a product incubation studio. At the same time, we are also an open-source team driving the entire field forward. Many times, I don't even know who is doing what; people spontaneously do things and create value for the ecosystem in impressive ways.

Bankless: It seems your vision is grand; what is the specific business model?

Shaw:

Our main goal is to serve a broader audience, not just Web3 users but also Web2 users. From simple Discord management bots to issuing tokens, we cover it all. You can think of it as "agent-based Zapier"—when you have business problems, you can find the corresponding agent to solve them. We provide this capability while building a marketplace for people to develop new features and earn from them.

We are:

• Considering establishing a venture fund to support the ecosystem

• Supporting community-led initiatives

• Building extensive partnerships

• Currently, at least five platforms are under construction, and there may be as many as 15

• Supporting open-source streaming projects like IOTV.

DAO Governance

Bankless: Speaking of governance issues, I've seen many DAOs become chaotic. For example, managing code repositories, GitHub governance, and the conflicts of interest that arise when many people participate. Can you share your experiences and views?

Shaw:

This indeed involves some deep-seated issues. Our Discord community has grown to about 13,000 people in just six weeks, with around 30,000 token holders. Currently, the community generally trusts core builders to have decision-making power, which is somewhat a response to the previous DAO "maximizing democracy" issue. In the long run, when you face 30,000 or 100,000 people, this approach can overwhelm decision-makers. That's why we need automated structures to address this issue—this is also what we truly want to do, which is to integrate "A" (artificial intelligence) into the DAO.

Imagine not reviewing proposals manually but fully automating the process. If the quality of people's proposals is not good enough, the system can help them improve or directly reject proposals that do not align with the current direction. Reviewers would only need to examine a small number of filtered proposals rather than all proposals.

This automation can extend to various aspects—from collecting opinions to specific execution. Ideally, a DAO would not need anyone to operate; it would run completely autonomously, from front desk reception to proposal submission to payment approval, all handled by AI agents. Of course, this is a long-term goal, but it's the direction we want to head.

The Phenomenon of Eliza Framework's Popularity

Bankless: The Eliza framework is now one of the most watched projects on GitHub. Why is everyone using Eliza? What makes it special?

Shaw:

From a technical perspective, Eliza doesn't have anything particularly outstanding. While we have indeed made some important technical innovations, such as the multi-agent room model, I believe the real value lies in our solution to the most basic social loop problem.

We developed a Twitter client that does not require an API, avoiding the $5,000 monthly API fee. It uses the same GraphQL API as a regular browser and can run in the browser. This made the entire project feasible because you can easily launch an agent and run it.

Additionally, we developed the framework using TypeScript, which is a language familiar to most Web and Web3 developers. We kept the framework simple, without excessive abstraction, allowing developers to easily add the features they want.

The Future of AI Agents and the Cryptocurrency Industry

Bankless: The crypto market is highly risky, and AI agents need thorough testing before they can replace human roles. Our goal is to replicate human behavior patterns in the crypto space into AI, right? Looking ahead, what do you think this ecosystem will look like once it matures?

Shaw:

From a clear long-term vision, we might reach the stage of AGI (Artificial General Intelligence) within 5 to 50 years. Combined with neural link technology (Neuralink), everyone could have a second brain, accessing all information at any time. The direction is clear; the key is how to get there.

When all technologies converge, it will be a beautiful scene where everyone has ample resources. But in the transitional period before that, there will inevitably be a lot of uncertainty, fear, and doubt—interestingly, this is where "FUD" (Fear, Uncertainty, Doubt) comes from.

Our goals are divided into two levels:

1. Practical level:

• Developing usable AI agents

• Building reliable infrastructure

• Ensuring system security

1. Spiritual mission:

• Promoting educational accessibility

• Empowering users with control

• Protecting data sovereignty

Just like the core idea of Web3, we hope everyone can:

• Create their own value

• Own their own data

• Understand and control technology

• Participate in system improvements

Two Paths for AGI Development

1. Centralized control path:

• Companies like Microsoft and OpenAI gain control through regulation

• Governments decide what can and cannot be done

I am concerned about this path because:

• OpenAI's models perform poorly in some aspects

• Models often carry fixed value biases

• A world where committees decide what AI can say could lead to dystopia

1. UBI (Universal Basic Income) path:

• AI will indeed replace many jobs

• For example, 5% of jobs in the U.S. involve driving (trucks, Uber, etc.), which may disappear within five years

• Even programmers like us are now heavily using Cursor and Claude

But I have concerns about the implementation of UBI:

• Reflecting on the rollout of government relief during COVID

• The controversies surrounding Obamacare

• UBI could become a product of political compromise

Advice for New Developers

Bankless: If there are developers using the Eliza framework and preparing to develop their first agent, what advice would you give them?

Shaw:

First, don't worry even if you've never programmed before. We hold AI agent development courses 1-2 times a week. I strongly recommend using Cursor, an AI-driven IDE that can save you a lot of time. Claude is also a great tool.

Remember three points:

• Keep your enthusiasm for learning; technology evolves rapidly

• Pay attention to security issues in development

• Don't be afraid to fail; learn from practice

Bankless: Any good learning resources to recommend?

Shaw:

• AI Agent Development School - Systematic courses

• Eliza framework documentation - Practical guides

• High-quality open-source projects on GitHub

Bankless: Can you introduce us to Agent Swarming?

Shaw:

Agent Swarming is a technique that allows multiple AI agents to work together. For example, one agent collects data, another analyzes it, and a third generates reports. These agents cooperate to accomplish more complex tasks.

For developers wanting to try this technology, I suggest:

• First master the development of a single agent

• Try collaboration between two agents

• Gradually expand to more agents.