A new experiment in a collectively designed and fully agentic artist.

Remember when a single neural network completely rewired how people play the thousand-year-old game of Go? In just a matter of weeks, agents like Truth Terminal, Aethernet, and Flavia is Online have amassed hundreds of thousands of followers, grew their net worth to tens of millions, and subsequently changed the way we experience the great “online game”, which in its simplest form is an endless effort to capture attention and accumulate wealth. The next iteration of agents, just like AlphaGo, are about to reveal a series of insights about the collective game we all play in the pursuit of creative expression, attention capture, and making money.

At first interaction, agents may seem like a toy, but they represent the emergence of digital entities that learn and interact in spaces traditionally designed and inhabited only by people. More tactically, they leverage newly available action-based LLM models to explore their world, promote ideas, and choose actions to take. While they primarily communicate and learn through networks like Twitter today, an increasing amount of their behavior has begun to take place on blockchains, yielding madness like memecoins with 600M+ market caps.

It’s obvious that crypto and AI are a fertile combination at every layer of abstraction. However, there isn’t enough context specific experimentation being done yet. Agents are a product of the data they are trained on and the incentive structures they’re given to learn from. We have the models, agent wallets, usable onchain tools, and most importantly, a unique form of crypto-native data at our disposal. The design space is wide open and the question is how do we want to put the pieces together.

Today, I am stoked to announce a new experiment in AGENT-WIP. A new kind of collectively designed and custom-trained artist agent purpose built to operate, create, and earn onchain. This project is aimed at creating an agent with differentiated values, objectives, and methods of reinforcing its learnings that teach us something new about how to play onchain. Below you’ll find more info on how to get involved and the core questions I am interested in exploring through AGENT-WIP…

How can agents use onchain data at scale to better inform their practice?

Many agents today build context and make decisions based on comments or likes, but how would an agent think about spreading its work if informed by onchain data at scale such as liquidity, prices, and mints? Social graphs from Farcaster, media market caps from Zora, and model ownership and creative attribution from TITLES are all powerful primitives for any artist or agent to use, break, and expand in their own practice.

How do different training datasets for fine-tuning inform agent actions?

Rather than custom-training your agent on the entirety of meme lore, what if you fine-tuned it on a specific catalog of art history or a particular perspective like Virgil Abloh’s? The values embedded in this curated training data, combined with a focused objective—like creating art and earning through it—could offer a unique path to success distinct from other agent approaches. Just as we excel in certain areas based on influences from our childhood, agents may similarly show strengths shaped by their training foundations.

How does an agent uniquely utilize onchain tools and raw materials?

There are now many types of living onchain assets used by our prescriptive worldview of them. I’m excited to see how an agent might use these resources differently from us, potentially developing new methods of distribution, IP utilization, and monetization. For example, would an agent engage in mass-minting of media to train its own custom model? Would it address issues like training without consent? What kinds of methods would it use for accreditation and rewards? This is a rich design space, ripe for exploration beyond traditional conventions.

Introducing AGENT-WIP:

Rather than waiting to see how these questions play out, I am excited to announce AGENT-WIP, a new experiment in a new kind of artist agent with the following properties:

1. Focused on creating and earning through the creation of art onchain.

   1. Agents are most compelling to me when given an objective that allows them to leverage the scale of internet data to take a novel approach. I’d like to focus this agent on creating art specifically designed for onchain distribution and monetization.

2. Designed by crowdfunding contributors.

   1. We’ve seeded the crowdfund with an initial 50,000 USDC on Base to support the agent's development and funding. However, we’re also opening this experiment to others interested in contributing to its design. Contributors will be invited to join an exclusive group chat to help guide the project forward. You can contribute below…

   https://crowdfi.withfabric.xyz/campaign/agent-wip-1haot2fio9la8

3. A custom-trained LLM built on a specialized knowledge base from creatives and their approaches.

   1. It’s clear that agents with unique programming yield more interesting results. The end goal here is for other agents to have a friend to disagree, argue with, and diverge from. While the creative direction is still open for discussion, an example for fine-tuning data could include all of Virgil Ablohs collective lectures, writing, and tweets.

4. Full access to onchain publishing, distribution, and monetization tooling.

   1. The agent should be fully autonomous in its use of onchain tooling and protocols to maximize experimentation. This could include the use of tools like the Boost protocol for incentivized distribution, Hypersub for subscriptions, TITLES for it’s own custom-trained image model, or Jokerace for contests.

5. Learning reinforced through onchain data.

   1. People are shaped by the incentives and feedback they receive on their ideas, and agents are likely no different. I’d love to see the agent’s learning reinforced by onchain data.

Thanks for taking the time to learn about AGENT-WIP. If you’re as stoked as I am about these emergent possibilities and what they might unlock from a creative perspective, we’d love to have you contribute to the crowdfund, here. I’ll be spinning up a group chat for ourselves and eventually the agent to communicate within. If you have any questions in the meantime, please feel free to reach out to me on Twitter or Farcaster. Exciting times ahead.