Skip to content
All guides

AI agents can now carry a budget. Write the spending rules first.

Mastercard opened early access to Agent Pay for Machines on Oct 20, 2026, letting AI agents pay for digital services inside a spending limit a business sets. What it is, what it is not yet, and the four rules a small business can write this week.

On October 20, 2026, Mastercard opened early access to Agent Pay for Machines, a way for AI agents to pay for digital services on behalf of a business while staying inside a budget the business approves. It is early access for developers, not a setting you can switch on today. But it shows where AI agents are heading, and the rules it depends on are ones a small business can write now.

What Mastercard announced

Mastercard showed the system at Cloudflare Connect and opened early access with APIs, tools and documentation for developers. It follows pilots Mastercard ran with Cloudflare, Ant International, AIsa, Base, BVNK, Coinbase and Hype.

The problem Mastercard describes is familiar. Paid online services usually need an account, an API key, prepaid credits or a billing relationship with each provider. That works when an agent uses one or two services. It gets harder when the agent needs to pick services as a task unfolds.

Mastercard's own summary of the hard part is worth repeating: "Payments aren't the hard part. Authority is."

How it works, in plain words

  1. 01

    You set the job and the limit.

    A business gives an agent a task and a spending limit.

  2. 02

    The service asks to be paid.

    When the agent reaches a paid service, the service answers with "402 Payment Required," a standard web response that means payment is needed to continue.

  3. 03

    The agent signs a promise to pay.

    The agent creates a digitally signed payment voucher, which Mastercard calls "a secure promise to pay," within its approved limit.

  4. 04

    The provider checks it and delivers.

    The provider verifies the voucher and the work continues.

  5. 05

    Payments settle later.

    Providers can combine vouchers and redeem them later, so the agent keeps working without a new account or invoice for every service.

In Mastercard's pilot demo, a marketing agency called Hype gave an agent a budget to plan a sample campaign. The agent bought social media research from one provider and campaign images from another. Mastercard sums it up as "Nine payments. Two independent merchants. One approved budget." Those are Mastercard's demo figures, not a measured result from a real customer rollout.

Figure: How Agent Pay for Machines works, an illustration (not a product screenshot) from our guide. Steps per Mastercard, Oct 20, 2026.

Why it matters to a small business

Today, most small businesses give software a card on file and hope the monthly bill makes sense. AI agents change that. An agent that can choose its own tools can also choose to spend, many times, in small amounts, faster than anyone reads an invoice.

Systems like this one are built to keep that inside a limit. The limit still has to come from you. If nobody decides what an agent may buy, how much, and who approves more, the safest payment system in the world will still spend money you did not mean to spend.

What it is not, yet

It is not something a small business can turn on today. Early access is for developers who build agents and paid services, and the pilots ran with payment and technology companies plus one marketing agency. It is also not a reason to hand an agent your company card. The point of the design is the opposite: limited authority, for a defined task, with a record of each payment.

When tools built on it reach you, they will most likely arrive inside software you already use, such as an assistant, a marketing platform or an accounting tool. That is when your rules need to be ready.

Four rules to write before any agent can spend

  1. 01

    A budget per task, not an open card.

    Give each agent a limit for each job, not access to a card with a high ceiling. Start small and raise it on evidence.

  2. 02

    A list of what it may buy, and from whom.

    Name the kinds of services an agent may pay for and, where you can, the providers. Anything off the list waits for a person.

  3. 03

    A person who approves anything above the limit.

    Pick one named person per agent. If that person is out, the agent waits.

  4. 04

    A weekly look at every agent payment.

    Put agent spending somewhere you already review, with the task, the provider and the amount on one line.

Figure: Four limits to set before any agent spends, an illustration (not a product screenshot) from our guide. Our suggestions, not Mastercard's.

This week

  1. 01

    1.

    List every AI tool in your business that can already spend money or start a paid plan on its own, including trials that auto-renew.

  2. 02

    2.

    For each one, write the limit, the allowed purchases and the approver on one page.

  3. 03

    3.

    Turn off automatic upgrades and top-ups where you can, and send billing alerts to a person who reads them.

  4. 04

    4.

    Ask your bank, card issuer and main software vendors how they plan to handle payments made by AI agents.

Figure: A four-question readiness check, an illustration (not a product screenshot) from our guide. Quote per Mastercard, Oct 20, 2026.

Our take

Agent payments are coming in through developers first, which gives owners a short window to set the rules before the tools arrive. Write the spending policy now; it costs an afternoon.

No partnership with Mastercard is implied here.

Sources

What happens when AI agents start paying each other? Watch Mastercard Agent Pay for Machines in action (Mastercard, October 20, 2026)

Mastercard Debuts Tool to Help AI Agents Buy Digital Services (PYMNTS, October 20, 2026)

Know someone who should read this?

Send it to a colleague, or hand it to your AI assistant and let it tailor the advice to your business.

Send to a personHand it to your AI

Want this working in your business?

Tell us what you are trying to fix and we will point you at the smallest useful next step.

Talk to us