Gaia Building The Agents AMA Recap

Preview

With the launch of Gaia's first fully autonomous hackathon, an engaging AMA session brought together Cecilia as Host and Moderator; James Young, Co-Founder at Collab.Land; Andrew Foreman, Founder at Greene Street; Harish Kotra and Carlos, Developer Advocate at Gaia.

The discussion explored the three innovative AI agents driving the hackathon: the Organizer Agent, designed for seamless event coordination; the Judge Agent, capable of evaluating submissions with both objective and personalized feedback; and the Bounty Distribution Agent, seamlessly executing rewards for winning teams. The speakers provided insights into leveraging cutting-edge technologies such as the eliza framework, Lit Protocol, and MetaMask Delegation Toolkit, while addressing the challenges of integrating decentralized AI tools and managing dynamic ecosystems.

Together, they highlighted the potential of AI-led DAOs, shared a vision for reducing coordination overhead, and emphasized the transformative role of collaboration and innovation in advancing Web3 infrastructure.

(Click here for the full AMA playback)

Preview

(Source: Gaia)

Cecilia (Host):

Welcome everybody. In yesterday's AMA, we covered the basics of Gaia's first fully autonomous hackathon. Today we are going to focus more on three mysterious AI agents who are going to dominate this hackathon. Let's start with a round of introductions from our speakers today.

James:

DM everyone. My name is James Young.

I lead a project called Collab.Land and we are into the intersection of AI and crypto because we believe that AI is really the future of group communication. Chat is not only a great interface for group discussion, but it's a great interface for AI. We manage over 60,000 on-chain communities snd we've seen these patterns of community engagement, burnout from mods and things like that so we are at this epicenter of AI and crypto from a group chat perspective.

And we've seen these patterns of community engagement, burnout from mods and things like that. So we are at this epicenter of AI and crypto from a group chat perspective. Happy to be here.

Andrew: My name is Andrew, Andrew Foreman. I'm one of the partners at Greene Street. We're a development product studio specializing in web3 and blockchain applications, anything from full stack to protocol. Our team has a lot of background and experience doing smart contract development, ML infrastructure for trading. When these AI agent applications started popping up earlier this year.

They were a perfect fit for what our team had already had experience in. Our first big project was doing the Polymarket Agents repo: we built an agent that can autonomously trade in a prediction market, which is an interesting information problem in itself. Then, over the last couple months, we've got to know the Gaia team and familiar with their infrastructure and excited to build out agents in this ecosystem. The power for agents here is the same for a lot of the Web3 and crypto applications. It's increasing the usability in the UX of these protocols and also connecting them and making the information flows work a bit better when you have this really powerful AI sitting in between them. We're excited to try both those things out here in this hackathon is like taking a good protocol and a good application and taking it to the next level with AI.

Harish:

Hey folks, I'm Harish. I'm a developer advocate at Gaia, primarily focused on building partner integrations, creating demos for all of you builders out there, building with Gaia and some of our partners and also working with James, Andrew and Lincoln from Coinbase for building out the agents for this hackathon.

Carlos:

This is Carlos from Gaia. I work in DevRel also with Harish. I have a background in creative tech, e-commerce. I've just worked in many different projects. I'm excited to talk about how we're working on building these agents, especially excited to work with James, Andrew and Lincoln.

Cecilia (Host):

Great, great. It's so nice to have everyone here. And now let's dive into the AI agents right away. So what are these agents that are being built for the hackathon? So James and Andrew, can you give us a quick intro of your agent overview that you're building?

James:

At Collab.Land, we are building the organizer agent for this hackathon, viewing it as a microcosm for event coordination. Starting from scratch just a few weeks ago, this organizer agent has served as a forcing function to test how quickly we can integrate various components.

A special mention goes to ai16z for their phenomenal eliza framework, which all developers should explore. We're leveraging eliza for the organizer agent. From a developer perspective, frameworks like eliza, akin to Next.js for AI and crypto, highlight the need for a decentralized Vercel. This organizer agent is specifically designed to streamline coordination.

As this space is still emerging, we've identified gaps and have been working at Collab.Land on a starter kit—a foundational step toward creating a Vercel-like solution for crypto AI. We're pushing to have it ready for the hackathon.

This is an experimental process, and there’s no better way to learn than by doing.

Andrew:

We're covering the judge agent. For this hackathon, we have an agent that can take in hackathon submissions and give out a little feedback summary report. The guide team had this awesome idea to have the judge act or sound as like Vitalik. So that's kind of why we get this like funny pun of like vittai-alik.

We'll be fine tuning and customizing our agent to sound more like him, which I think just points to something that's generally interesting about these kinds of applications is that there's this objective criteria of what makes a good hackathon project, what's code quality, what do the organizers look for. Then there's the subjective element of what would Vitalik say about this, or what's his style preferences. Trying to bring that together into one agent is the goal here.

Preview

(Source: Luma)

Cecilia (Host): What is the tech stack and the partners that are being integrated into the organizer agent?

James:

We're using the eliza framework from ai16z alongside Lit Protocol to manage the agent's ability to sign a smart account. Additionally, we’re coupling the MetaMask Delegation Toolkit, which enables delegation. There’s a lot to unpack with Lit Protocol and MetaMask, but for developers building autonomous agents, this involves creating programs designed to operate independently. However, developers often bear liability because they typically hold the private key or seed phrase for these agents.

As we transition from meme coins, which are egalitarian, to AI-driven meme coins, the challenge lies in ensuring developers are not in the critical liability path. Lit Protocol, combined with the MetaMask Delegation Toolkit, is an emerging solution that is critical for enabling AI autonomy while maintaining human oversight. At Collab.Land, we aim to promote AI-led initiatives steered by groups of token-holding individuals who can guide the AI effectively. This forms the cornerstone of our foundation, integration efforts, and partnerships.

Understanding this requires hands-on experimentation, which is why we’re building out a hacker template or starter kit. A special shoutout to Gita, who has been working tirelessly on integrating Lit Protocol with MetaMask and has done an excellent job. We’re also leveraging technologies like Gaia to ensure decentralized control over training data, aligning AI agents with the incentives of their users.

Beyond this, we’re exploring integrations with Story Protocol to enable AI agents to own and share intellectual property. When these agents are guided by humans, it’s fascinating to see what content emerges. Other partners we’ve integrated include Bounty Caster for bounties, Joke Race for on-chain submissions, and Privado for identity solutions. The Joke Race project, for instance, ensures that all submissions are as on-chain as possible.

Our goal is to create a comprehensive template that simplifies the intricacies of building and managing AI agents. By integrating these various services, we aim to provide developers with a robust foundation to innovate without being burdened by technical complexities.

Andrew:

What James and his team are doing is impressive. Happy to dive into what we’re doing with the judge agent as well.

The judge agent is unique. We don’t have a front end for our application, so the focus is on creating an AI agent capable of managing the workflow for scoring submissions. This starts with our Gaia node, which connects all the various pieces and applications that it has access to.

For those unfamiliar with the Gaia node, it’s essentially a collection of AI tools assembled to build an agent. It includes a chat model, database components, and configuration options for feeding information into the AI model. Think of it as a specialized web server optimized for serving AI inference. Our task is to orchestrate all these components into an end-to-end workflow.

The workflow begins with Joke Race, where a hacker submits their project details, including the project name, description, and GitHub link. From there, we initiate the review and scoring process. The agent connects to RLM through our Gaia node, communicates with the developer via Farcaster or Telegram, and posts results on-chain for payment purposes. The agent integrates these steps into a seamless process.

One noteworthy integration we’re also implementing, as James mentioned, is Lit Protocol. Lit acts as a decentralized key management system. This allows us to store the API keys for our agent securely within a decentralized protocol. While not strictly a workflow feature, it represents an essential step toward decentralization and governance. In the future, if the agent becomes widely used with more features and users, its governance could be opened up to a broader community.

This kind of decentralization and governance potential is an exciting add-on feature we’re actively designing for the judge agent. It underscores the broader possibilities of leveraging these tools to create innovative, scalable, and decentralized AI solutions.

Preview

(Source: Luma)

Cecilia (Host):

Could anyone share some backstories about the three AI agents or some fun facts behind ideating and building these three agents? Were there any difficulties or challenges along the way?

James:

I can give a little bit of a backstory, and it may seem like, why is Collabland interested in AI agents when Collab.Land is a token gating service? My background is in DAOs—I helped co-write the white paper for Moloch DAO and helped start Metacartel DAO five, five and a half, or six years ago.

The idea of empowering a group of people online to make decisions and coordinate was fascinating. Previously, before crypto, I was a game developer. Looking at this from a game theory perspective was quite interesting. I thought of DAOs as a way to reduce coordination overhead. But what you see are people campaigning, politics, and similar dynamics.

Now, the reason I’m so interested in this organizer agent is: can you have an AI that helps propose solutions? It acts as a reflection or a mirror for the rest of the group being led by this AI, but you still need to steer the AI. In AI parlance, there’s RLHF—reinforcement learning through human feedback. What’s interesting is, how do we tokenize that? The AI makes proposals based on what it reads in a group chat, reflects back what’s being discussed, and the group just vetoes or approves. That’s how you steer these DAOs.

My backstory, coming from DAOs and token-gated chats, is that we’re ripe for AI-led DAOs to start moving things along. When you organize humans, especially online in a decentralized setting, what you often lack is leadership. Can we use AI to spark ideas and address voter apathy, which is common in DAOs? What we want to figure out is, can we create a veto-only DAO that’s led by an AI? If we can, can we then have AI-led DAOs where AIs are members, and humans are at the edge of the network? With the swarming of these AI-led DAOs, can we reach AGI—not as one sentient model, but as a graph or network of specialized AIs? These AIs would have niches, coordinate with one another, barter, and exchange information.

Ultimately, the organizer AI agent is an experiment in coordination reduction. What does it mean to have hyper-efficient, on-chain organizations? I imagine this as the new era of network state startups. This hackathon is just a spark—a microcosm of that vision. That’s how I connect DAO token gating with AI agents coordinating and organizing events like this hackathon.

Harish:

I've organized hackathons all my life. For this autonomous AI agent hackathon, it boiled down to identifying which pieces of the hackathon could potentially be automated through an AI agent, which is almost all the hackathon process itself. That’s how we jumped into saying, let's create three agents that do bounty creation so that we don't have to go back and forth with partners or ask this guy to figure out which bounties are ideal. Let's see what the agent creates. It’s going to be fun to see what the organizer agent creates as bounties.

Then, of course, bounty distribution is the biggest roadblock in general across the board. There are a lot of complexities involved, and builders obviously want to get their hands on the money they win as soon as they can. The nuances that go into bounty distribution itself are a big challenge to crack, which is where Coinbase and Lincoln from Coinbase are supporting this hackathon, ensuring bounty distribution is seamless and automated from the input that the judge agent provides.

Specifically, when it comes to challenges, we are discovering them as we build these agents live. We are yet to see what more complications will arise. While we’d love to integrate all our partners, that’s another challenge—where do we stop? Every partner for this hackathon has amazing platforms, and it’s hard to include all of them within the timeline we’re working on.

We are essentially running into challenges every day while building these AI agents and organizing the autonomous AI hackathon. I’m really looking forward to seeing what these agents produce as bounties and what you, as builders, submit.

Preview

(Source: Gaia)

Cecilia (Host): To wrap things up, the last question is for the builders. As creators of these agents, you know them the most. Is there any particular type of AI agents you or the judge would like to see being built during the hackathon?

Harish:

We have three tracks for the hackathon that we're looking for builders to tackle. One is an agent that pushes the boundaries of conventional norms and creates something amazing. Then there is agent-to-agent interactions. If you are able to pull this off with the starter kit that James' team is offering—creating agent-to-agent interactions using any Web2 or Web3 APIs—that’s a great use case to build. And finally, how do you utilize user-generated data to create value for end users? What AI agents can you build to leverage the collective knowledge out there?

There are also bounties tied to these tracks that you should look forward to. You can contribute and use some of our partners, like Eigenlayer, Coinbase, Base, Story Protocol, Nevermind, Lit Protocol, Privado, Functor Network, and Bounty Caster. I hope I didn't forget anyone, but all of them will have bounties published on Bounty Caster that you can leverage while building for the tracks I just mentioned.

James:

I would like to see if there could be an AI agent built and trained on figures like Steve Jobs, Paul Graham, Balaji Srinivasan, and Elon Musk, and have that agent lead a DAO to figure out how to mainstream crypto, market it, and productize it. I think that would be interesting to see how much traction it could gain. If that AI agent could then communicate with other types of AI agents trained on prominent historical figures and see what those agents do in terms of making proposals, that would be fascinating. I’m all for keeping people at the center of this and having the agents coordinate actions between them.

Cecilia (Host):

I would like to remind everyone who wants to be part of this legendary hackathon to register as soon as possible using the Luma link. you can also find the link on Gaia’s twitter so make sure you give them a follow. Starting from Dec 4 - 6, we will have a webinar each day showing the 3 AI gents in live environment so you get to know them better. The official hackathon kickoff will be on Dec 9th and the week will be full of workshops to help you and guide you in the process.

Last but not least, demo day and bounty distribution will be on Dec 13th. Make sure you finish everything before that day.