Agentic commerce: when the store sells to your AI, not to you

Instant Checkout is the 'buy' button moved inside the conversation with ChatGPT. You ask about shoes, ChatGPT shows you three pairs, you say 'get the second one', and the payment goes through right there, without leaving the window. The first stores were from Etsy (a marketplace for handmade and vintage goods), and now more are joining that sell through Shopify.

The technical trick is called a Shared Payment Token: Stripe issues ChatGPT a one-time, limited token tied to a specific order and a specific store. ChatGPT never sees your card number. It works outside Stripe too, because the Agentic Commerce Protocol is an open standard, not a locked-in system.

Shopify didn't wait for someone else's rules to be imposed on it. It launched the Universal Commerce Protocol, developed together with Google, and more than a million stores worldwide have already opened it up, including well-known brands. The difference is in the scale of ambition: while Stripe and OpenAI solve the payment, Shopify wants the agent to do everything, from searching the catalog to tracking a shipment, directly through Google Search, Gemini, or Microsoft Copilot.

In October 2026 another question got a seat at the table: who's actually talking. Meta and Sierra are developing the Personal Agent Protocol, and on October 9 Sierra published its draft, called Poppy, along with 35 more design partners, among them OpenAI, Visa, Mastercard and PayPal. The idea is for your personal agent to introduce itself instead of walking into the store with your account. The company publishes one file at a standard address on its site, the agent says whose it is, the customer decides what access to give it, and the company decides what the agent can do. In Sierra's example the agent can start returning a jacket, but can't change the payment details. Sierra describes the protocol as a complement to the Universal Commerce Protocol from the previous paragraph. For now it's a draft: the partners take part in the design with their feedback, which doesn't mean the protocol already works at their end.

For a store owner, this shifts the target. Until now the photo, the description, the reviews were written to convince a human to press a button. Now part of that conversation is carried by software that reads structured data, without feeling a mood off a photo. The store has to speak two languages at once - human and machine - and the second one is only just starting to dictate the rules.

Nobody asks the customer whether he wants this. The standards are written among companies with an interest in having payments flow through them - buyers don't sit at that table. The convenience is real. So is the risk of a wrong order made by a hasty agent.

The visual is generated code-art, no third-party images.