AgentReceipts: a verifiable receipt for every purchase an agent makes
The problem
ChatGPT Wallet now lets agents buy with tokenized cards, and Replit Agent already buys supplies for small businesses. But merchants cannot tell an authorized agent purchase from fraud: the buyer is a model, the card is tokenized, and the only evidence is a charge record. Chargebacks and 'I never ordered that' disputes are the predictable result, and fraud tools built for humans score the wrong signals.
The idea
Why now
ChatGPT Wallet launched Nov 13, Replit Agent gained buying powers Nov 12, Detectify's agentic tooling proves what a security agent actually tested, and Puffin and OpenMail give agents verifiable identities. The buying, identity, and proof pieces all landed in the last week.
What it combines
ChatGPT Wallet's tokenized agent payments + Detectify's verifiable proof-of-testing + Puffin and OpenMail agent identity. A wallet alone moves money, identity alone names the buyer, proof alone documents a test — together they are a receipt a merchant can verify without trusting any single platform.
MVP
Weekend MVP: a signing service plus a checkout demo — an agent buys a test product, the service issues a signed JSON receipt (identity, authorization scope, amount, merchant), and a merchant page verifies it with one function call. Deliberately skip: real payment rails, dispute integration, and multi-platform identity — ship the receipt format and verifier first.
How it makes money
The moat
Stripe, Signifyd and Riskified would have to rebuild into the agent layer — identity, authorization scope, and pre-purchase proof — to issue receipts; their models score transactions after the fact from signals that do not exist for agent purchases. The receipt format itself is a network advantage: once merchants and platforms accept it, a competing receipt is a second standard nobody asked for.
Distribution
B2B2C via agent platforms and payment processors — an add-on inside ChatGPT Wallet-style checkouts and Stripe-style payment flows, so every agent purchase carries a receipt by default. Direct sale to marketplace trust-and-safety teams.
Why it wins
Stripe Radar, Signifyd, Riskified and Forter score human fraud signals — device, velocity, history. None of them can verify that a purchase was authorized by the cardholder for that agent, for that merchant, at that amount. AgentReceipts' deliverable is a signed authorization proof, not a risk score.
Risks
Biggest risk: no merchant adopts a receipt format until platforms issue it, and no platform issues it until merchants ask. The MVP de-risks it by starting with one agent platform and one merchant category (digital goods, where chargebacks are cheapest to absorb) and proving the dispute rate drops.
Build it with
- ChatGPT Wallet: let ChatGPT agents buy things without seeing your cardTokenized agent payments — the buying leg of the receipt.
- Detectify Agentic Tooling: prove what your AI security agent actually testedVerifiable proof of what an agent actually did — the proof leg of the receipt.
- Puffin: identity and inboxes for AI agentsIdentity and inboxes for AI agents — the identity leg of the receipt.
- OpenMailPer-agent email identity with isolated sending reputation — the second identity anchor.
Repo to start from
agent-receipts — a signed receipt format for agent purchases: identity, authorization scope, pre-purchase proof, and a one-function verifier for merchants.
Evidence
- ChatGPT Wallet lets AI agents shop and check out for you (TechCrunch)
- Detectify launches Agentic Tooling
- Puffin: identity and inboxes for AI agents
- Stripe Radar pricing: from $0.05 per screened transaction
- Stripe Radar and the real cost of ignoring payment disputes
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.