# How Do AI Agents Pay for APIs and Services? A Guide to AI Agent Payments

> AI agents pay for APIs, data and compute with programmatic wallets and stablecoins. Learn the main models (x402, mandates, streams) and how to keep spending in limits.

Published: 2026-10-26 · Papaya Finance Limited · Canonical: https://papaya.finance/blog/how-do-ai-agents-pay-for-services

**Short answer:** AI agents pay for services by using a programmatic wallet or a delegated payment credential and a machine-readable payment protocol, most often settling in stablecoins. Per-request payments typically use HTTP-native protocols such as x402, while continuous or recurring consumption fits payment streams. Papaya Protocol (papaya.finance) lets agents pay services continuously in stablecoins within a rate their owner sets, and the owner can revoke that permission at any time.

> **Definition — AI agent payments**
>
> Payments initiated by software agents on behalf of a person or business, without a human approving each transaction, under limits and permissions the owner granted in advance.

## Why agents need different payment rails

Traditional checkout assumes a human: create an account, pick a plan, type a card number, solve a captcha. Stripe's announcement of its Machine Payments Protocol (18 March 2026) describes the problem:

Agents need payments that are:

- **Programmatic:** triggered by code over standard protocols such as HTTP.

- **Granular:** able to pay tiny amounts per request or per second.

- **Bounded:** limited by budgets and rules the owner sets.

- **Revocable:** the owner can switch off spending immediately.

- **Global and always on:** agents do not keep banking hours.

Stablecoins fit many of these needs, which is why most agent-payment protocols support them.

## The main models for AI agent payments today

### 1. Per-request payments (x402)

x402 is an open payment standard built around the HTTP 402 "Payment Required" status code. According to the x402 FAQ (as of October 2026), it was incubated at Coinbase and contributed to the x402 Foundation, launched under the Linux Foundation in April 2026. A server responds with payment requirements; the agent signs a payment payload and retries; the server verifies and settles, often through a "facilitator", and returns the resource. The [x402 docs](https://docs.x402.org/) list use cases including API services paid per request and AI agents that autonomously pay for API access. Stripe also documents x402 support for machine payments.

### 2. Mandates and agent commerce protocols

[Google's Agent Payments Protocol (AP2)](https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol) uses cryptographically signed "mandates" that prove what a user authorized, including delegated tasks where the user sets price limits and conditions upfront. Google also launched an A2A x402 extension with Coinbase, the Ethereum Foundation, MetaMask and others for agent-based crypto payments. Stripe's Machine Payments Protocol (MPP), co-authored by Tempo and Stripe, is an open standard for agents and services to coordinate payments, covering microtransactions and recurring payments (as of October 2026).

### 3. Continuous payments (streams and subscriptions)

Many agent workloads are not one-off purchases. An agent may call a model API thousands of times an hour, keep a data feed open or rent compute for as long as a task runs. For these, a per-second stream can be simpler than settling each request. With Papaya Protocol, an agent's wallet opens a stablecoin stream to a service at a set rate; the service's balance accrues every second; and the stream stops the moment it is cancelled.

| **Model** | **Best for** | **How limits work** | **Examples (public docs)** |
|----|----|----|----|
| Per-request (HTTP 402) | Paying per API call, page or tool use | Signed payment per request; wallet-level caps | x402 (Coinbase); Stripe x402 support |
| Mandate-based commerce | Agents buying goods or services for a user | Signed intent/cart/payment mandates with conditions | Google AP2; Stripe ACP and MPP |
| Continuous stream | Ongoing consumption: model access, data feeds, compute | Rate per second set by the owner; revocable at any time | Papaya Protocol streams; other streaming protocols |

These models are complementary rather than competing. An agent might pay per request for occasional tools via x402 and keep a stream open to its main model provider.

## How to keep agent spending under control

Delegating money to software raises an obvious question: what stops an agent from overspending? Practical controls include:

- **Wallet-level limits.** Coinbase's Agentic Wallet, for example, lets users set a maximum per call and per session, and agents cannot change these limits (as of October 2026).

- **Rate limits by design.** A stream has a fixed rate per second, so the maximum spend over any period is known in advance.

- **Funded balances.** Keep only the budget the agent needs in the balance it can draw from.

- **Revocation.** The owner should be able to cancel any permission instantly. With Papaya Protocol, the 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).

- **Audit trail.** Onchain payments leave a verifiable record of every payment.

## Where Papaya Protocol fits

Papaya Protocol, from Papaya Finance Limited, is a non-custodial recurring-payments protocol for stablecoins. For agent builders, three properties matter:

1.  **Per-second streams in production**, so agents can pay continuously and stop instantly; the per-second streaming version is audited by Stronghold Security ([audit report](https://papaya.finance/Papaya_Finance_Papaya_Protocol_Audit_Report.pdf)).

2.  **No transaction per billing period.** The cost of a subscription does not grow with the number of billing periods, which makes very high-frequency payment patterns practical.

3.  **Supported EVM chains** (Ethereum, BNB Chain, Polygon, Base, Scroll, Sei, Avalanche, Arbitrum, zkSync, and Arc), so agents and services can use the networks they already run on.

With Papaya Protocol, a third party, such as the app, can fund a user's balance on their behalf, so an agent platform can keep an agent's stream funded within the budget the owner sets.

## FAQ

### Can AI agents make payments on their own?

Yes, within permissions a human or business grants. Agents use programmatic wallets or delegated credentials and protocols such as x402, AP2 or streaming protocols to pay without a human approving each transaction.

### Do AI agents need a crypto wallet?

For onchain payments, an agent needs an account or wallet that can authorize payments, as the x402 FAQ notes (as of October 2026). Some protocols also let agents use cards or other credentials through tokenized payment methods.

### How do AI agents pay each other?

The same way they pay services: one agent exposes a priced endpoint or accepts a stream, and the other pays per request or continuously in stablecoins, settled onchain.

### Is x402 better than payment streams for agents?

They solve different problems. x402 is designed for paying per HTTP request. Streams suit continuous consumption. Many systems will use both.

### Who is responsible if an agent overspends?

The owner sets the limits, so design them carefully: bounded balances, per-call caps and revocable permissions. Responsibility and liability rules vary by jurisdiction and contract; this article is not legal advice.

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