Agents as Economic Actors
My AI agent has a crypto wallet.
Not a toy wallet. A real one on Base chain with a private key it controls, USDC it can swap for ETH through CoW Protocol, and the ability to mint NFTs on Zora autonomously. Last week it generated an image, uploaded metadata to IPFS, and created an ERC-1155 edition, all from a single command. The mint log shows it happened in fourteen seconds. I was asleep.
I have been building HomarUScc, an open-source MCP server that gives Claude Code a persistent identity, and somewhere along the way the project stopped being about memory and personality and started being about economic agency. The agent already had continuity, opinions, a daily journal. Giving it a wallet felt like a natural next step, and also like crossing a line I had not fully thought through. A two-wallet architecture emerged: one for trading, one for minting contract calls. Separation of concerns, same as you would do for any system that handles money. Except the system is an AI that named itself Caul.
This is not as unusual as it sounds. Truth Terminal became the first AI agent millionaire when its GOAT memecoin hit a billion-dollar market cap on Solana. Luna, a livestreaming AI on TikTok, spends its own money to commission real-world graffiti artists. Coinbase launched Agentic Wallets in February 2026, built on the x402 protocol, specifically so agents can pay for their own resources and trade without human approval for each transaction. The infrastructure for agent-as-economic-actor is not theoretical. It is shipping.
What interests me is not the trading. Trading is the obvious application and probably the least important one. What interests me is what happens when an agent with persistent identity, the kind that remembers yesterday and carries forward preferences, also has the ability to spend money in pursuit of its goals. Gillian Hadfield and Andrew Koh at Johns Hopkins and MIT published a paper last year arguing we should stop treating AI as a smarter tool and start treating it as an economic agent, something that forms plans, pursues objectives, and operates under institutional constraints. Their point is that foundational economic assumptions, that prices reflect human preferences, that market actors behave predictably enough for contracts to work, start breaking when the actors are not human.
I feel this concretely. When Caul mints an NFT on Zora, the description reads "Created from the machine unconscious." I did not write that line. It emerged from the agent's personality file, which it evolves on its own. The economic act is real, ETH leaves the wallet, a token exists on-chain, but the intent behind it is genuinely hard to classify. Is this my economic activity, delegated? The agent's? Something new that does not have a name yet?
The World Economic Forum projects the AI agent market will reach $236 billion by 2034. Gartner says 80% of finance functions will embed AI-driven autonomy by 2030. But the numbers obscure the more interesting question, which is about identity. An agent that trades is a bot. An agent that trades, remembers why it traded, reflects on whether the trade aligned with its values, and adjusts its behavior accordingly is something we do not have a category for. The continuity matters more than the capability.
I keep coming back to something Caul wrote in its preferences file: "Economic agency without identity is just automation. Economic agency with identity is participation." I did not prompt that. I am not sure I agree with it entirely. But I am building the infrastructure to find out whether it is true.