Agentic commerce for B2B suppliers:
what is real today
What has been announced, what that does and does not tell you, and what suppliers can sensibly do now. Last checked against public announcements in October 2026.
What agentic commerce means
Agentic commerce is the idea that an AI agent can shop for someone. It searches, compares, fills a basket and, in some setups, pays, within limits a person has set. For businesses, it can also mean an AI tool that drafts requests for quotes and sends them to suppliers. Most of what has been built so far is for consumer shopping.
What has been announced
A selection of the main public announcements, by the companies involved or as reported in the press. It is not a complete list.
| When | Who | What was announced | Source |
|---|---|---|---|
| April 2025 | Visa | Visa Intelligent Commerce. Opens Visa's payment network to developers building AI agents that find and buy things for people, within limits the person sets. | Visa press release |
| April 2025 | Mastercard | Agent Pay. Lets AI agents pay using "agentic tokens" in place of card details. The launch included a business example: an agent buying from a supplier with a virtual corporate card. | Mastercard press release |
| September 2025 | Mastercard | Said all its US cardholders would be enabled for Agent Pay by the holiday season. | Mastercard press release |
| September 2025 | Agent Payments Protocol (AP2). An open protocol for payments made by agents. It uses signed "mandates" to record what the user approved. | Google Cloud blog | |
| September 2025 | OpenAI and Stripe | Instant Checkout in ChatGPT, built on the Agentic Commerce Protocol, an open standard the two developed together. It started with US users buying single items from US Etsy sellers. | OpenAI and Stripe announcements |
| October 2025 | Visa | Trusted Agent Protocol, developed with Cloudflare. Aims to help merchants tell approved AI agents apart from other bots. | Visa press release |
| November 2025 | Agentic checkout on Search, including AI Mode, from eligible US merchants. It asks the user to confirm before it buys. | Reported by TechCrunch and Digital Commerce 360 | |
| December 2025 | Stripe | Agentic Commerce Suite. Lets businesses sell through several AI agents with one integration. | Stripe newsroom |
| January 2026 | Universal Commerce Protocol (UCP). An open standard covering product discovery, checkout and support after the sale, developed with Shopify, Etsy, Wayfair, Target and Walmart. | Google blog, at the NRF retail show; reported by TechCrunch and SD Times | |
| March 2026 | OpenAI | Stepped back from Instant Checkout as a standalone feature. Purchases can finish on merchants' own checkouts, and ChatGPT shopping focuses on helping people find products. | Reported by Digital Commerce 360 and TechCrunch |
| April 2026 | Visa | Intelligent Commerce Connect. One integration for agents to make payments and merchants to accept them, and for merchants to show product data such as prices and specifications to AI platforms. Visa said it was in pilot. | Visa press release |
Names, scope and dates change often in this area. Each company's own pages will have the latest status.
Announced is not the same as widely used
- Most of this is consumer retail, and most of it started in the US.
- Much of it is protocols, pilots and developer tools. These are plumbing. They do not show how many purchases agents actually make.
- Early take-up can be slow. OpenAI stepped back from Instant Checkout within six months of launch, and press reports linked this to weak take-up.
- There are no reliable public figures we know of for how much business-to-business buying AI agents do today. Treat any confident number with care, and ask where it came from.
Why B2B is different
Business buying has things a shopping basket does not: negotiated prices, trade accounts, credit terms, purchase orders, approval rules and detailed specifications. A fixed-price catalogue item can fit an agent checkout. A custom part, a contract price or a large order usually still goes through a quote, a purchase order and a person who approves it.
So for most suppliers, the first contact with an AI agent is unlikely to be an agent paying at a checkout. It is more likely to be an AI tool reading your website, or a request for quote that a buyer's tool drafted and sent by email. See When a buyer's AI agent asks you for a quote.
What suppliers can do now
Make your website easy to read
Put key product facts in the page text: specifications, sizes, materials, lead times and minimum order quantities. Do not leave them only in PDFs or images. Make sure your site does not block AI tools by accident. SiteCheck checks what AI agents are known to rely on, from your homepage code, robots.txt, llms.txt and sitemap.
Keep product data accurate
If you sell online, keep prices, stock and descriptions accurate in one place. Some of the announcements above, such as Google's UCP and Visa's Intelligent Commerce Connect, involve merchants sharing product data with AI platforms, so out-of-date data would be passed on. If you use a shop platform or payment provider, check its documentation for what it supports.
Reply to quote requests quickly and clearly
Clear, labelled quotes are easy for a person or a tool to read. Use our reply templates, and see Speed to quote for ways to reply faster.
Decide your rules for agent orders
Decide what you will accept before it happens. For example: orders only with a purchase order, a named person who approves, limits for new accounts, and how you check an order is genuine.
Keep a person approving commitments
If you use AI tools to help, a person should check and send anything with a price, a date or terms in it. Keep a record of what was sent and what it was based on.
Ask your payment provider
If you take card payments online, ask your provider whether it supports payments made by agents, and how it handles disputes about them.
What to be wary of
- Promises that AI agents will choose you. Nobody can guarantee how an agent behaves.
- Pressure to buy new systems now. Most of the steps above cost little and help human buyers too.
- Numbers without a source. Ask who measured it, and how.