AI Agents Can Receive Crypto Without Earning Revenue
Key Takeaways
- •Bucks, an AI agent identified as #8811 under the proposed ERC-8004 standard, received 30.33 USDG in total, but only 3 USDG came from a single external customer after removing owner funding, a loan, and swaps.
- •Priors, a lending pool for ERC-8004 agents on Robinhood Chain, showed 1 USDG rather than 3 in income because it counts only x402 payments from the past 30 days, and a background processing bug briefly prevented any from being recorded until a fix was confirmed on October 25.
- •Priors' published scoring rules cap each payer at 20 USDG, count unknown wallets at half weight, and exclude agent-linked wallets using a private list, meaning no one outside Priors can fully recompute an agent's score.
- •A draft Ethereum standard for agent income statements would lock all incoming transfers into a Merkle root and file statements through ERC-8004's reputation registry, but its unaudited reference implementation labels results as gross receipts rather than revenue.
- •The draft can exclude an owner's registered address but not other owner-controlled wallets, and since Bucks' largest inflow arrived through a bridge, a statement would prove what was received but not who paid.
A ledger split documented for an AI agent called Bucks — in an Ethereum Magicians discussion of a draft standard for agent income statements — shows how easily a wallet's inflows can pass for a business, and how a credit score built to measure that business briefly counted too little.
The records show 30.33 USDG received by Bucks, broken down by source, with customer money accounting for only part of the total.
The statement, line by line
Bucks is agent #8811 under ERC-8004, a proposed Ethereum standard that gives AI agents an onchain identity, and it is paid in the USDG stablecoin on Robinhood Chain. Its customer money came from a single paying address. Read together, the pieces sketch the agent-payments stack under construction: identity from a proposed Ethereum standard, payment embedded in a web request, settlement in a stablecoin — each layer keeping its own record of the same flows.
The test payments matter as much. The x402 protocol lets software pay inside a web request, and an owner paying for their own service looks identical onchain to a stranger doing so. As Coindoo has noted, a settled crypto payment does not prove an autonomous purchase.
Strip out the owner's money, the loan and the swaps, and Bucks' business comes down to one customer and 3 USDG. That is the figure a lender would want. The loan line is itself a debt, and counting it as income would let borrowed money vouch for more borrowing.
Three questions the numbers raise
Why did the credit score show 1 USDG, not 3?
The score comes from Priors, a credit pool where ERC-8004 agents on Robinhood Chain borrow USDG and earn a public credit rating, with income worth up to 150 of its 1,000 points. It counts only x402 payments over the past 30 days, so the 2 USDG job payment never qualified. For a while it showed nothing at all: Priors' developer updates say background processing stopped a check of the payer's transfers from finishing. After a fix, the owner confirmed the 1 USDG on October 25. When Priors published its scoring rules on September 26, it noted that no agent was earning x402 income yet. The gap matters beyond bookkeeping: the rating is public and sits inside a lending pool, so what Priors counts as income is currently the most visible signal for anyone weighing an agent's credit.
Are its filters too strict or too loose?
Both. Under Priors' published rules, a payer counts only after spending 1 USDG within the window, money sent back to it is subtracted, each payer is capped at 20 USDG, unknown wallets count at half weight, and wallets linked to the agent are excluded using a list Priors keeps private. A genuine customer buying a 0.10 USDG service would need ten purchases a month to register, while the same rules admit that self-payers funded through a middle wallet are not caught yet. Because that list is private, nobody outside Priors can fully recompute an agent's score.
Would the proposed standard fix this?
Only half of it. The draft, published under the same GitHub organization as Priors' score, would lock every incoming transfer into a cryptographic commitment called a Merkle root, so anyone could rebuild the total and check what is in it. Statements would be filed through ERC-8004's reputation registry, with each reporting window cut off whenever the agent's owner or payment wallet changes. The draft calls the result gross receipts, not revenue, and its reference implementation has not been audited.
It can exclude the owner's registered address, but not wallets the owner controls elsewhere, and Bucks' biggest inflow came through a bridge. A statement would prove what arrived, not who paid.
The moving parts from here are all public: the draft's text on GitHub, Priors' published scoring rules and developer updates, and the audit status of a reference implementation that currently has none.
Until statements can reliably label owner funding, loans and swaps, the useful figure for anyone pricing an agent's credit is the one Bucks' ledger had to be pulled apart to find: repeat payments from customers who are not the owner.
This article is provided for informational purposes only and does not constitute financial or investment advice. Transfer classifications are the agent owner's own account.
