Visa, Mastercard and Amex Built Rails for AI Agents. What Changes for You?

Visa, Mastercard and American Express each launched a program for payments made by AI agents between early March and mid April 2026. If you run ecommerce at a DTC brand, the natural question is whether any of that changes what your team has to do. The short answer: the payment side is carried by your processor and your commerce platform. The part that stays with you is whether the assistant recommends your product in the first place.

What each card network launched

Visa's piece is Intelligent Commerce Connect, announced by Visa on April 8, 2026 as part of its Intelligent Commerce portfolio. Visa describes it as a single integration that lets agents pay and merchants accept agentic transactions, in pilot with partners including AWS, Highnote and Mesh, with more partners due through the year.

Mastercard introduced Verifiable Intent on March 5, 2026. PYMNTS describes it as an open-source framework that links a shopper's identity, their instructions and the outcome of a transaction into one tamper-resistant record, so every party sees the same facts if a dispute arises. It sits on top of Mastercard Agent Pay, which registers and authenticates agents before they transact. Mastercard and Visa also partnered with FIS so issuing banks can use the networks' Know Your Agent authentication.

American Express released an agentic commerce developers kit on April 14, 2026. American Banker reports that the kit verifies agents and their intent, that Amex has completed thousands of test agentic payments with AI platform partners, and that Amex will cover errors made by AI agents. The closed loop Amex describes in its newsroom, where it issues the card, runs the network and processes the merchant's payment, is what lets it make that promise.

Where Stripe and the checkout protocols fit

Stripe published the Machine Payments Protocol with Tempo on March 18, 2026, an open standard for agents to pay programmatically that Stripe users can accept through the PaymentIntents API. Stripe also co-authored the Agentic Commerce Protocol with OpenAI, and its Agentic Commerce Suite, announced in December 2025, lists URBN, Etsy, Coach, Kate Spade and Revolve among the brands onboarding.

On the Google side, the Universal Commerce Protocol covers checkout from AI Mode and Gemini, and CNBC reported in March 2026 that Gap would be the first major fashion retailer to launch checkout inside Gemini.

Treat all of this as reference. None of it is a build for an ecommerce team. Your payment processor, your commerce platform, or both will support whichever rails matter, and on Shopify the Agentic storefronts settings already manage the channel side for ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta.

What the rails do not solve

Payment rails move money after a shopper has chosen. The choice happens earlier, when someone asks ChatGPT or Google AI Mode for a waterproof commuter jacket under $200 and the assistant assembles an answer from product pages, Merchant Center feeds and structured data. Google's retail team says plainly that conversational shopping in AI Mode runs on the product data merchants provide, and that a messy or incomplete Merchant Center feed means customers will not find the products.

Referral data shows how uneven the outcomes already are. Similarweb figures reported by Digiday in September 2025 put ChatGPT at one in five of Walmart's referral clicks, while Amazon, which blocks AI crawlers, drew under 3 percent. Same payment rails, different results, and the difference sits in what each retailer lets assistants read.

So the commercial question is not whether your checkout can accept an agent. It is whether your product shows up as the recommendation, at the right price, in stock, matched to the shopper's question. That is product information work, and no network launch does it for you.

What to check in your own store

• Ask the questions your shoppers ask. Pick the products that carry your revenue and ask for them the way a shopper would in ChatGPT and Google AI Mode. Record whether you appear as a product card, a mention, or not at all. The ChatGPT product results guide walks through what those answers are built from.
• Check Merchant Center for those products. Disapprovals, missing GTINs, and price or availability mismatches between feed and page take a product out of consideration before any assistant judges it. Fix them at the source so they stay fixed. That is the product feeds work Paz handles for brands.
• Read the product page the way a question would. Size and fit, materials, compatibility, what is in the box. If the answer lives in a PDF or a support article, it is not on the page, and assistants are unlikely to find it.
• Get a baseline. The free AI readiness report shows what an assistant can read on a product page and where the gaps are.

Frequently asked questions

Which card networks support payments by AI agents?

As of spring 2026, all three major US networks have a program: Visa Intelligent Commerce Connect, Mastercard Verifiable Intent on top of Agent Pay, and the American Express developers kit. Stripe's Machine Payments Protocol and the Agentic Commerce Protocol add rails on the processor side. Each is dated above with its source.

Do shoppers complete purchases inside ChatGPT?

Mostly no, as of October 2026. Shopify's ChatGPT channel documentation describes it as a discovery channel where shoppers buy in your online store checkout, with no direct checkout setting to turn on. OpenAI's commerce documentation still publishes an agentic checkout spec for merchants that integrate directly. Plan for the referral, and measure checkout inside the assistant only if it appears for your products.

Does my team need to implement any of these protocols?

Usually not. Platforms and processors carry the integration, and the merchant stays the merchant of record. The work that stays in-house is product information: a complete, accurate feed that stays eligible, and product pages that answer the questions shoppers ask. That is also the work that pays off in Google Shopping and on your own site whether or not an agent ever pays.

This is the work Paz does for lean ecommerce teams. A Paz expert finds the gaps in your product information across feeds, product pages and AI answers, you approve the fixes, and the platform publishes them and tracks whether your products show up. If you want to know where you stand before the next network announcement, talk to us.