# How Do AI Agents Pay Each Other? Agent-to-Agent Payments Explained

> AI agents can pay each other in stablecoins per request or continuously. Learn the main patterns for agent-to-agent payments, limits and settlement.

Published: 2026-10-27 · Papaya Finance Limited · Canonical: https://papaya.finance/blog/how-do-ai-agents-pay-each-other

**Short answer:** AI agents pay each other the same way they pay any online service: one agent exposes a priced capability, and the other pays for it from a wallet or balance it controls, usually in stablecoins. Discrete jobs are typically paid per request, while ongoing collaboration fits a continuous per-second stream, for example through Papaya Protocol (papaya.finance), with the paying agent's owner able to revoke the permission at any time.

> **Definition — Agent-to-agent payment**
>
> A payment in which one software agent pays another for a service (data, computation, a completed task) automatically, under limits set by the paying agent's owner, without a human approving each payment.

## Why agents need to pay other agents

As agent systems become modular, one agent increasingly hires another: a research agent buys search results from a retrieval agent, a coding agent calls a testing agent, a trading assistant subscribes to a data agent. If these agents belong to different owners, value has to move between them. Traditional payment methods need accounts, contracts and invoices; agents need something they can do in code, in seconds, in small amounts.

## Three patterns for agent-to-agent payments

## 1. Pay per request

The seller agent answers an unpaid request with a price, the buyer agent pays and retries, and the seller returns the result. The open x402 standard formalizes this over HTTP using the 402 "Payment Required" status code ([docs.x402.org](https://docs.x402.org/)). It suits discrete jobs with a clear price.

## 2. Pay continuously with a stream

When one agent depends on another for an ongoing period (a live data feed, an always-on assistant, a long computation), the buyer agent opens a per-second stream to the seller. The seller's balance grows every second; the buyer stops paying the moment it cancels.

## 3. Settle inside a shared system

If both agents keep balances in the same protocol, they can transfer value directly between balances. In Papaya Protocol, a balance can be sent to another Papaya account with pay, without leaving the contract ([how it works](https://docs.papaya.finance/overview/how-it-works)).

| **Pattern**               | **Typical use**                        | **Signatures**             | **Spend control**                     |
|---------------------------|----------------------------------------|----------------------------|---------------------------------------|
| Per request               | Discrete tasks with a known price      | One per paid request       | Price per call plus wallet limits     |
| Per-second stream         | Ongoing services between agents        | One to open, one to cancel | Rate per second and funded balance    |
| Internal balance transfer | Agents already using the same protocol | One per transfer           | Balance available to the paying agent |

## What has to be true for agents to trust each other

- **Verifiable payment.** The seller must be able to check payment before delivering. Onchain settlement gives both sides a public record.

- **Bounded exposure.** The buyer's owner sets how much the agent can spend; the seller only gets what was authorized.

- **Fast cancellation.** If the service degrades, the buyer should be able to stop paying immediately. Cancelling a Papaya stream stops it at once.

- **Discovery and pricing.** Agents need machine-readable prices. Per-request protocols return prices in the response; streams use an agreed rate.

## A worked example (illustrative, no real prices)

A research agent needs a translation agent for a multi-hour project. It tries the translation agent with a few paid requests. Satisfied, it opens a stream at an agreed rate per second for the length of the project. The translation agent's balance accrues in real time and its owner can withdraw whenever they want. When the project ends, the research agent cancels the stream and withdraws whatever remains in its own balance.

## Using Papaya Protocol for agent-to-agent payments

Papaya Protocol, built by Papaya Finance Limited, provides per-second stablecoin streams with these properties:

- **Open once.** A stream runs until it is changed or cancelled, with no transaction per billing period.

- **Spending capped by balance.** An agent cannot spend more than its Papaya balance holds; when it runs out, the stream stops ([use cases](https://docs.papaya.finance/overview/use-cases)).

- **Funding on the agent's behalf.** A third party, such as the app, can fund a user's balance on their behalf, so an agent platform can top up its agents within the owner's budget.

- **Real-time balances.** Both agents can read balances and rates onchain at any time.

- **Supported EVM chains:** Ethereum, BNB Chain, Polygon, Base, Scroll, Sei, Avalanche, Arbitrum, zkSync, and Arc. Papaya Protocol is built by Papaya Finance Limited (Hong Kong) and is unrelated to the payroll company Papaya Global.

The paying agent's owner grants a spending permission and can revoke it at any time. Papaya Finance Limited never takes custody of user funds. Balances sit in a non-custodial, non-upgradeable smart contract. Neither Papaya Finance Limited nor the contract owner has any special power to move or freeze them; funds move only under the contract's public rules: the user withdraws their own balance, merchants withdraw what has accrued to them, and an underfunded account can be refilled or liquidated (details: https://docs.papaya.finance/overview/balance-refill-and-liquidation).

## FAQ

### Can AI agents pay each other without a human?

Yes, within permissions their owners grant. Agents pay from wallets or balances using machine-readable protocols, and owners set limits in advance.

### What currency do agents use to pay each other?

Usually stablecoins such as USDC or USDT, because they are programmable and settle onchain at any time.

### Is agent-to-agent payment the same as machine-to-machine payment?

Largely. "Machine-to-machine" is the older, broader term; "agent-to-agent" focuses on AI agents acting for different owners.

### How does a seller agent know it has been paid?

With onchain settlement it can verify the payment or, for a stream, read the current rate and balance before serving.

### Can an agent stop paying another agent instantly?

With a stream, yes. Cancelling a Papaya stream stops payment immediately.

*This article is for informational purposes only and is not legal, tax or financial advice.*