Can an AI agent buy something on its own and prove it owns it?
I built a store that sells icons to AI agents with no human approving the purchase. There are three tiers: 1,000 free, 500 standard at $0.002 and 50 animated at $0.010. An agent asks for whatever it needs, gets back a mix, and is charged only for the ones that cost money. The price had to be low enough that buying beats generating one, and the payment had to leave a public record, because that record is the proof of ownership.
- The agent requests icons. Telby batches them into one request for efficiency.
- Telby prices it. Only the paid icons count, so a basket of six came to $0.014 with three of them free. Watermarked previews cost nothing, so the agent can judge before paying.
- The agent signs one transfer. Its key never leaves it, and nothing is charged until this point.
- Settlement waits for a ledger close. The testnet ledger closes approximately every 5 seconds, which is where almost all of the waiting goes.
- The files are released. Each SVG carries the transaction that bought it.
- The purchase is recorded. A second transaction stores a hash of the icon list, which is the part anyone else can check.
end to end. This is a testnet constraint.
five icons in 7.5s instead of 39s, for the same money
given funds, a catalog and no instructions
The part I did not expect
Two of the four test agents bought nothing. One decided the free icons were good enough. The other wanted the icon and thought the price was fair, but had never been told it was allowed to spend money. Tests three and four were identical except for one line giving the agent a budget, and that line decided whether it bought anything. Payment code does not fix that.