Agentic commerce means an AI agent shops on a person's behalf. It searches for products, compares them, builds a cart and pays, while the person sets the limits and confirms. Since 2025, dedicated protocols connect agents, merchants and banks: the Agentic Commerce Protocol from OpenAI and Stripe, Google's Universal Commerce Protocol, the Agent Payments Protocol, developed by Google and now with the FIDO Alliance, and the agent programs from Visa and Mastercard. As of 31 October 2026, buying through an agent is mainly available to shoppers in the US. In Europe, the payment rails are running first live transactions, and Stripe offers its agent tools to merchants in Germany too. For small businesses this mostly means clean product data and clear rules about what an agent may do.
This article is about trade and payment through agents. What agentic AI means as a term is covered in Agentic AI Explained: What It Is and Isn't. The information comes from the vendors' pages (as of 31 October 2026), not from a test or case of my own. Where I add my own view, I say so.
Agentic commerce: what is behind it
Stripe's documentation puts it like this: agentic commerce lets AI agents carry out transactions for users; buyers can browse products, get recommendations and complete purchases without leaving the conversation. Technically, it is what makes up any AI agent: a goal, tools and a loop (more in What Is an AI Agent? Explained Simply). What is new is that money changes hands.
That is where today's systems fall short. The Agent Payments Protocol documentation sums it up: today's payment systems assume a human is directly clicking "buy" on a trusted website. When an agent buys, three questions are open: did the person really authorise this purchase? Does the agent's request reflect the person's true intent, without errors or "hallucinations"? And who is accountable if something goes wrong? The protocols try to answer these three questions.
The protocols and initiatives at a glance
One course correction: on 29 September 2025, OpenAI launched "Instant Checkout", buying directly in the chat, starting with US Etsy sellers and single items. On 24 March 2026, OpenAI wrote that the first version lacked the flexibility it wanted; merchants can now use their own checkout. ChatGPT's merchant page says OpenAI is moving away from a standalone Instant Checkout. You search and compare in the chat and buy on the merchant's website or app.
How a purchase by an agent works
The basic pattern is the same everywhere:
- Catalog: the merchant provides products, prices and stock as a structured feed.
- Discovery: the agent finds suitable products and compares them for the customer.
- Checkout: the agent opens a checkout session; the merchant or its payment provider returns totals, taxes and shipping options.
- Confirmation: the customer confirms the purchase. AP2 calls this a mandate, Mastercard an explicit consent confirmed by passkey.
- Payment: the agent does not pass on a card number but a limited token. At Stripe this is a shared payment token: it is valid for one merchant, with a maximum amount and an expiry.
- Order: the merchant receives the order and ships it.
Important for merchants: with ACP, Google's UCP checkout and Shopify, the merchant remains the merchant of record and keeps the customer relationship.
What agentic commerce means for merchants and small businesses
Product data and feeds. An agent only buys what it finds in structured data. ChatGPT, Google and Shopify all rely on product feeds; Google announced new Merchant Center attributes such as answers to common product questions and compatible accessories. Stale data costs sales: according to Stripe, checkouts fail when the price shown to the shopper does not match the price charged, and in most cases updating stock and prices every 15 minutes is enough. OpenAI itself says that changed prices or shipping terms can reach ChatGPT with a delay.
Checkout. There are two paths. Either the agent sends the customer to your existing checkout, as ChatGPT does by default today, according to OpenAI with no fees for purchases on your site. Or the purchase happens in the agent's interface, as with Google's UCP checkout, which needs an integration through your shop system or payment provider.
Fraud and bots. Many shops block automated traffic. Visa writes that bot detection can mistakenly block legitimate agent purchases and built the Trusted Agent Protocol for this: approved agents identify themselves with cryptographic signatures. Mastercard calls this "Know Your Agent". So ask your payment provider whether it recognises such agents before you block or allow bots across the board.
Liability. Who is liable for a wrong or fraudulent purchase by an agent is an open question according to the AP2 documentation: the user, the agent's developer, the merchant, the issuer, the payment provider or the orchestration layer. The protocols are meant to provide evidence through signed mandates and an audit trail. None of the sources gives a settled answer. I am a developer, not a lawyer, and this is not legal advice; when in doubt, have your specific case checked.
My view: for a small business, the first point pays off today. Complete, current product data helps in every search, whether a person or an agent reads it.
Risks: prompt injection, wrong purchases, consent
Prompt injection. An agent reads outside content, and that content can hide instructions. OpenAI gives the example of an agent asked to answer overnight emails that ends up sharing bank statements. As a safeguard, OpenAI says the ChatGPT agent pauses and asks for confirmation before sensitive steps such as completing a purchase. How these attacks work and what helps in a business is covered in Prompt Injection: Protecting AI Agents. What applies in the browser specifically is in Claude in Chrome: How to Use the Extension.
Wrong purchases. An agent can pick the wrong product or work with outdated prices. AP2's mandates address this: when the person is present, they confirm the finished cart; for a delegated purchase, they set rules in advance, such as a price limit and timing. Limited payment tokens cap a mistake at the approved amount.
Consent. All vendors stress that the person stays in control, but they do it differently: confirming each step, a signed mandate, or a passkey at the bank. In my workforce the same rule applies without any protocol: anything that leaves the house or moves money waits for my word. How to define such levels is covered in What Can an AI Lead Role Decide?.
Status in Germany and the EU
As of 31 October 2026, and only what the vendors say themselves:
- ChatGPT: according to OpenAI's merchant page, shopping is currently live for users in the US, with more regions announced. The release notes of 24 March 2026, however, list the new product results for all ChatGPT users.
- Google: according to the Merchant Center help, UCP checkout covers products in the United States, Canada and Australia, for select merchants; the UK is to follow. Germany is not mentioned.
- Stripe: Agentic Commerce Suite and shared payment tokens are available in Germany and other European countries.
- Mastercard: on 13 May 2026, Mastercard reported the first authenticated agentic transaction in Germany, with Deutsche Bank, DZ Bank and N26: an AI agent booked an event on Mastercard's platform Priceless.com. According to Mastercard, as of 2 June 2026 all its issuers in Europe were enabled for Agent Pay at network level.
- Visa: "Agentic Ready" launched on 17 March 2026, first in Europe including the UK; early partners include Commerzbank and DZ Bank. Testing runs with selected merchants.
- Shopify: the Agentic plan for brands without a Shopify store has rolled out globally; the announcement does not say in which countries each AI channel sells.
My view: the payment rails in Europe are being prepared and tested right now, while buying through an agent is so far mainly a US offering for consumers.
Frequently asked questions
What is agentic commerce, in simple terms?
An AI agent does your shopping: searching, comparing, building the cart, paying. You set the limits and confirm.
What is the Agentic Commerce Protocol?
An open standard from OpenAI and Stripe under the Apache 2.0 license for exchanging product data and checkout between agents and merchants. In ChatGPT it currently serves product discovery; the purchase happens on the merchant's site.
Can I already shop through an AI agent in Germany?
According to the vendors, hardly. ChatGPT shows product recommendations to all users, but per OpenAI's merchant page, shopping with merchant feeds is live only in the US so far. Checkout in Google's AI Mode is not available in Germany. First live transactions ran here through Mastercard Agent Pay with three banks.
Do I need to do anything now as a small merchant?
Nothing urgent. Keep your product data complete and current and ask your payment provider which agent features it offers. If you sell through Shopify or Etsy, OpenAI says your catalog is already connected to ChatGPT.
Who is liable if an AI agent buys the wrong thing?
The sources give no uniform answer; the AP2 documentation calls the question open. This is not legal advice.
How to continue
Once you know what an agent does, the next question is what it may do. The basics are in Agentic AI Explained: What It Is and Isn't, everyday examples in AI Agents: 10 Real-World Examples. Everything on the topic is on the AI agents page. How other businesses lead and protect their agents, we discuss in my community.
Your first AI employee up and running within an hour
Join the community$5/month founding price, first 50 members