Agentic Payments: How AI Agents Pay, Who Is Building the Rails, and What Changes for Merchants

Agentic payments are payments made by an AI agent acting for a person or an organisation: the agent selects what to buy, presents a credential that proves it is authorised to spend within limits the payer set, and the merchant and network verify that authorisation before settling the transaction. The concept follows directly from agentic commerce — if an agent can search, compare and choose, it needs a way to pay — and it became concrete in 2025 when Mastercard announced Agent Pay and Visa announced Intelligent Commerce within a day of each other, followed by PayPal's agentic commerce services, Stripe's agent tooling and shared protocol with OpenAI, Google's Agent Payments Protocol with sixty-plus partners, and Coinbase's x402 standard for machine-to-machine stablecoin payments. By mid-2026 the networks, Stripe and AWS had joined a Linux Foundation body to govern x402, Visa had integrated its payment tools into OpenAI's agent system, and the question for merchants had changed from "will agents pay?" to "can they pay us?"

How Agentic Payments Work

The mechanism differs by rail, but every serious design solves the same four problems.

1. Agent identity. The merchant needs to know the request comes from a legitimate agent, not a bot scraping a checkout. Visa's Trusted Agent Protocol, developed with Cloudflare, signs agent traffic so a merchant can distinguish an authorised shopping agent from abuse.
2. Scoped credentials. The agent does not carry the person's card number. It carries a token — Mastercard calls them Agentic Tokens — bound to that agent, a merchant scope, a spend ceiling and a consent policy. If the agent tries to buy outside the scope, the token fails.
3. Consent and control. The payer sets the limits up front: what the agent may buy, from whom, up to how much, and whether each purchase needs a confirmation. The agent operates inside them without asking at every step.
4. Verification and settlement. The merchant verifies the token, the network checks the scope and the consent, and the transaction settles on the existing card, wallet or stablecoin rail with a receipt trail that names the agent.

The vocabulary is still settling — AI payments, AI agent payments, agentic checkout and agent pay all describe parts of the same mechanism — but the four problems above are constant.

The result is a purchase the person authorised in principle and the agent executed in practice, with a record both can audit. That is the difference between agentic payments and an agent simply typing a saved card into a form — which is what early agents did, and what the networks were built to replace.

The Rails: Who Is Building What

Two things stand out. First, the card networks chose to build agentic payments into the existing rails rather than let agents route around them — the tokenisation and consent layers sit on top of ordinary card transactions. Second, the crypto-native approach (x402, stablecoin settlement) went from fringe to co-governed by Visa and Mastercard inside a year, which suggests the machine-to-machine micropayment case — agents paying for API calls, data and each other's services — is being taken seriously alongside consumer shopping. The agentic commerce guide covers the discovery and checkout layers that sit above these rails.

What Agentic Payments Change for Merchants

A new order source with its own rules

Agent-originated orders arrive through protocols, not browsers. The merchant needs to accept the credential type, verify the agent, expose real-time price and inventory the agent can read, and support a programmatic checkout. Merchants on Shopify, Etsy or Salesforce Commerce Cloud were largely enrolled in ChatGPT Instant Checkout by their platform; others have to integrate deliberately. Structured product data is the precondition on the discovery side; protocol support is the precondition on the payment side.

Fraud and abuse look different

The old bot problem was scrapers and card testers. The new one is distinguishing an authorised agent from an unauthorised one that looks identical at the network layer. Agent identity signals — Trusted Agent Protocol, registered agents under Agent Pay — are what let a merchant admit the first and block the second. Blocking all agents, which many sites did in 2025, now means blocking paying customers.

Attribution has to be rebuilt

The click that used to mark the start of a purchase does not exist when an agent buys. The agent's recommendation happened in a conversation; the transaction arrived through a protocol. Order-source fields, protocol metadata and referral tagging are the only way to know what agentic channels are worth, and they have to be instrumented before volume arrives. Tracking revenue impact from AI chat and measuring AI-attributed revenue cover the method.

What Agentic Payments Change for Brands

The payment is the last step of a decision the agent made earlier. When a person tells ChatGPT to find and buy a laptop under $1,200, the agent shortlists two or three options from what the answer engines know and recommend, and the payment goes to one of them. A brand that is not in the shortlist never sees the transaction. That makes AI visibility — how often the agent recommends you — a distribution metric, not a marketing one, and it is why MultiplierAI's Recon agent measures share of recommendation across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews as the leading indicator of agent-originated revenue. Paid placement — ChatGPT ads, sponsored cards beneath a conversation — can cover a gap; it does not replace being recommended.

Agentic Payments in B2B

Consumer shopping is the visible case; procurement is where the dollar volume will sit. A procurement agent that requests quotes, compares against a rate card and raises a purchase order needs the same scoped credential and consent model, with approval thresholds instead of spend caps. Virtual cards with per-agent limits — Stripe Issuing and the networks' commercial products — are the early mechanism, and the machine-to-machine case (agents paying for data, compute and other agents' services) is where x402-style protocols are aimed. The enterprise AI agent governance rules apply directly: decision rights, spend limits, approval gates, full logs, a kill switch.

What to Do Now

1. Find out whether you can already accept agent payments. If you sell through a major commerce platform, you may be enrolled in ACP or a network program without having done anything. Confirm it and read the order data.
2. Stop blocking agents indiscriminately. Implement agent identity verification so authorised agents can transact and abuse is still stopped.
3. Expose machine-readable price, inventory and checkout. Agents cannot buy what they cannot read.
4. Tag agent-originated orders now, so the first meaningful quarter of agentic revenue is measurable and attributable.
5. Measure share of recommendation. The payment follows the shortlist. Know whether you are on it — an AI visibility audit is the starting point — and work on the third-party corroboration that changes the answer.

Frequently Asked Questions

What are agentic payments?

Payments initiated and completed by an AI agent on behalf of a person or organisation, using a credential scoped to that agent — merchant, spend limit, consent policy — so the merchant can verify authorisation and the payer keeps control. They are the payment layer of agentic commerce.

How is an agentic payment different from a normal card payment?

The agent presents a token bound to it and to a scope, not the person's card. The network checks the scope and the consent policy before settling, and the receipt names the agent. A normal card payment has no way to express "this software may spend up to $200 at these merchants."

What is Mastercard Agent Pay?

Mastercard's framework, announced in April 2025, for letting verified AI agents transact using Agentic Tokens — tokenised card credentials bound to a specific agent, merchant scope and consent policy — with agent registration and merchant verification built in.

What is Visa Intelligent Commerce?

Visa's platform for agent payments, announced the same week: agent-ready credentials, spending controls, authentication and fraud protection built into automated buying, extended in October 2025 by the Trusted Agent Protocol for verifying agent traffic and integrated with OpenAI's agent system in 2026.

Are agentic payments safe?

Safer in principle than an agent using a saved card, because the credential is scoped and revocable and the agent is identified. The risks that remain are the usual ones — misconfigured limits, compromised agents, merchants that cannot tell authorised agents from abuse — which is why identity verification, scoped credentials and consent controls are the core of every rail.

References

1. https://www.mastercard.com/us/en/news-and-trends/press/2025/april/mastercard-unveils-agent-pay-pioneering-agentic-payments-technology-to-power-commerce-in-the-age-of-ai.html
2. https://usa.visa.com/about-visa/newsroom/press-releases.releaseId.21365.html
3. https://stripe.com/guides/agentic-commerce
4. https://www.agenticcommerce.dev/
5. https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol
6. https://www.x402.org/
7. https://www.forbes.com/sites/boazsobrado/2026/03/19/stripe-visa-and-mastercard-race-to-build-ai-agent-payment-rails/
8. https://www.techtimes.com/articles/320813/20260717/visa-mastercard-stripe-back-open-standard-letting-ai-agents-pay-autonomously.htm