Agents and crypto

Over the past couple of weeks there’s been a lot of excitement around the intersection of AI agents and crypto.

One of the projects to capture everyone’s attention is Truth Terminal (or Truth), a custom tuned large language model (LLM) trained by a developer called Andy Ayrey. Custom tuned LLMs are general purpose LLMs like Claude, ChatGPT, Llama etc. that have been trained on additional data.

A criticism levelled by online hackers at the general purpose models is that they are shackled in their responses. Too politically correct and boring. Truth as a direct response to this, was trained on Internet memes and culture to develop an edgier, Internet-native, personality.

Andy created an X account for Truth and gave it read / write access, which means that it can post on X and respond to users. While it’s unhinged at times, Truth also exhibits a healthy dose of introspection.

In addition to an X account, Andy set up Bitcoin and Ethereum wallets for Truth that users can send money into. A couple months ago Marc Andreessen sent $50k worth of BTC to Truth’s Bitcoin wallet, which Truth put to use in a variety of ways, including investing into a memecoin called GOATSE, which at one point reached a market cap of $300m.

Next to Truth, there’s a few other agents that are active on social media including Aethernet and Luna. These agents all have their distinctive personalities and interact with others as if they were just another user.

Another category of agents that is emerging is agents that act as helpful companions.

Gina for example, is an agent that users can interact with on Farcaster (and on X soon). Users can query Gina directly in feed to answer questions, analyse images or blog posts, or generate visualisations of onchain data. I recently asked Gina to generate a chart for the two best performing DeFi tokens over the past 6 months that have a market cap of over $500m.

This is what Gina produced:

The team behind Gina plans to enable users to create a wallet where they can add funds and have Gina execute transactions autonomously on their behalf. I would then be able to do things like tell Gina to mint 1 NFT per day for me from Zora, or buy a specific token when certain conditions are met.

Something that is getting drowned out amidst the current excitement is that, in reality, these projects are not quite agents yet. They’re LLMs. The key difference between LLMs and agents is that LLMs are tools specifically designed to process language, while agents are fully autonomous systems that can make decisions and work toward goals on their own, without human intervention.

For Truth Terminal, Aethernet, Luna, and Gina, the setup of social accounts and crypto wallets, the spending of crypto from these wallets are still all managed by their respective developers.

Still, what these projects have managed to do is capture the imagination of users. Agents and crypto is a nascent yet powerful intersection. For starters, crypto wallets serve as native bank accounts for agents. Agents don’t have an address, they can’t KYC, and can’t onboard at a bank—but they can spin up a crypto wallet. Once agents have a wallet, they can seamlessly move funds and initiate actions using stablecoins and other crypto assets across blockchains. Thanks to native interoperability blockchains enable, these agents then have access to a shared app ecosystem, something that is not on offer in web2.

There’s a range of different paths teams can take. Agents can be their own online personality, or they can be assistants. I’m sure there’s also other things they can be in the future. The great thing is that all the infrastructure developments in AI and crypto have enabled a wide range of avenues to be explored.