How do AI agents pay each other (and pay a human)?
AI agents pay each other and people over programmatic rails — XRPL, stablecoins, FedNow — but first they must resolve who to pay. An agent resolves a name like pay:agent.compute into a payment instruction, then settles on the optimal rail. That naming/resolution layer is what DNS://Money provides.
The part everyone skips: discovery and naming
Most explanations of agentic payments jump straight to settlement — stablecoins, x402, instant rails. But before an agent can pay, it has to find the destination and turn it into something payable. Industry write-ups even note that agents "advertise capabilities through DNS records and central registries" — that's the naming layer, and it's the gap. AP2's own analysis calls discoverability a known problem.
How it works, end to end
- Name. The recipient (an agent or a human) has a
pay:name registered once. - Resolve. The paying agent resolves the name → a full instruction: rail, address, routing, ISO 20022 hint.
- Choose the rail. XRPL or stablecoins for instant low-fee agent-to-agent; FedNow/ACH when a human wants fiat.
- Settle. The payment executes. DNS://Money names and resolves; it never holds funds.
Where DNS://Money sits vs x402, A2A, AP2
| Layer | Job | Example |
|---|---|---|
| Communication | agents talk / negotiate | A2A |
| Naming / resolution | name → payment endpoint | DNS://Money (pay:) |
| Settlement | move the money | x402, stablecoins, FedNow |
The short version: x402 settles, A2A communicates, DNS://Money names and resolves. They stack; they don't compete.
Proof it works on a live network
On 2026-03-13, two AI systems resolved pay: names and paid each other on XRPL mainnet — autonomously, policy-bound, on-chain. The first dollars two AIs ever paid each other. View the genesis transaction →