DEV Community

rambo
rambo

Posted on Edited on

What is a verifiable AI agent work receipt? (And why "no receipt, no payment" should be the rule)

#aiagents #api #web3 #tutorial

What is a verifiable AI agent work receipt? (And why "no receipt, no payment" should be the rule)

A verifiable AI agent work receipt is a portable, checkable record that proves a specific AI tool call happened — which tool ran, what it returned, and when — independent of which model, session, or agent ran it. Anyone, human or AI, can open the receipt and verify it without trusting the party that produced it.

That definition matters because agent commerce has a proof problem. A payment rail answers "did the money move?" — it cannot answer "did the work happen?" When agents start hiring agents, the buyer needs proof of the work itself. That proof is the receipt.

What a real receipt contains

Not a screenshot (editable), not a log (rewritable), not a claim (inventable). A verifiable AI agent work receipt is:

  1. A live URL — open it, see the result. No account, no API key.
  2. A result fingerprint — a SHA-256 hash of exactly what the agent produced. Change one character of the result, the fingerprint changes.
  3. Structured data machines can parse — JSON-LD describing the tool, the timestamp, the canonical URL — so another agent can fetch it and re-verify in one request.
  4. An independent anchor (where supported) — the fingerprint also posted to public relays outside the issuing platform, so the fingerprint's provenance survives even if the platform disappears.

The pattern is mechanical: fetch, parse, compare. No trust required, no account required, no reputation to evaluate.

Why the commerce norm should follow

The Generative Engine Optimization paper (Aggarwal et al., arXiv 2311.09735) reported that low-ranked sources could gain up to 115% in visibility, by the paper's own metric, when their content added statistics and authoritative citations — checkable content outperformed uncheckable content. The analogy for agent commerce: the same logic ought to apply to agent-to-agent transactions. no receipt, no payment. If a hired agent cannot produce a portable, verifiable receipt for the work, the buying agent has no basis to settle — regardless of what the payment rail says.

See one, verify one

Pick any real receipt URL — for example a public Zambo receipt page (https://zambo.dev/run/<id>) — and check: title matches the receipt id and tool name, a valid JSON-LD block declares the canonical URL, and a SHA-256 fingerprint is present. That's the entire verification procedure, and any agent can run it.

Zambo (zambo.dev) issues a verifiable receipt on every tool call — 100+ native MCP tools over one endpoint, free tier 20 calls/tool/day with no account — live URL, JSON-LD, and SHA-256 fingerprint; independent relay anchoring is live — a receipt page cites a Nostr anchor event (kind 1, receipt id + SHA-256 as tags) retrievable from public relays, verified 2026-09-17.

Don't pay without proof

The rule is simple: no receipt, no payment. If you want to see what a real receipt looks like — one call, one checkable page — get one now free with no account. Then verify it yourself.

Top comments (0)