Agentic commerce for your store
An AI agent already searches, compares and, on some sites, buys for your customer, and most of what you hear about it mixes what exists with what has been announced. This path lays out agentic commerce from start to finish. By the end you will know at which stage of a purchase an agent steps in, what each protocol does and who maintains it, what a store like yours can use today and what is only announced, and which data has to be in order before any of those pieces matter.
First what an agent is and what it buys, then how it finds your catalog, then the language that talks to your store, and last the payment, the stage that has opened up the least. In each stretch, the path keeps two things apart: what a store can use today, according to the page of whoever maintains it, and what that same page presents as announced. The path has five stretches.
- Steps 1 to 5 · What it is and who buys. Agentic commerce is buying in which an agent acts for the shopper, and the store is still the seller. It rests on agentic AI, the AI agent that carries out tasks, the line between an autonomous and an assisted agent (who approves each action) and the shopping agent, which works for the customer and knows only what you publish about your store.
- Steps 6 to 11 · How an agent finds you. Discovery is the first stage, and the only one that needs no payment protocol. It starts with AI product discovery and purchase intent, goes on to the machine-readable catalog and ends at the three routes with a name of their own: the ChatGPT product feed, ChatGPT Shopping and Shopify's agentic storefronts. Almost everything a store can use today is here.
- Steps 12 to 17 · The language between agents and stores. Before the buying protocols there are layers of plumbing that are worth not confusing. Tool calling is what lets a model ask for an action; MCP and the MCP server standardize how tools are offered to it; A2A and its Agent Card set how agents introduce themselves and talk to each other. The protocol gateway is the name some vendors give to the intermediary that translates between them.
- Steps 18 to 22 · The buying protocols and delegated payment. This is where most of what is announced sits. ACP is maintained by OpenAI and Stripe under the Apache 2.0 license, and ChatGPT was the first platform to implement it; UCP was introduced by Google on January 11, 2026, with Shopify, Etsy, Wayfair, Target and Walmart, and covers the whole buying journey; AP2 answers a different question, how an agent proves that the person authorized the payment, with the signed mandate as its central piece, and it is one way of solving delegated payments. In all three, the store is still the seller.
- Steps 23 to 26 · Buying. Agentic checkout is the payment stage; Instant Checkout is its version inside ChatGPT; the agentic browser is the other design, in which the agent goes through the store's website; and coupons for agents close with a practical detail that the protocols treat as an extension of payment.
What a store can use today and what is announced
Each stop is judged against the page of whoever maintains the piece, not against a third party's write-up. This is what those pages say, read on October 7, 2026.
- ACP and the ChatGPT feed. Today: the protocol is open and any business can implement the specification; each platform manages how businesses take part, and OpenAI is the first to implement it, with ChatGPT. Also today: the ChatGPT product feed is open to approved partners, who request access through a form.
- UCP. Today: the specification is published and covers catalog, cart, payment, identity and order management; lodging reservations are in draft and food delivery is listed as "coming soon". Announced on January 11, 2026: that UCP would "soon" power checkout on eligible listings in Google's AI Mode and in the Gemini app for eligible U.S. retailers, and that Google would expand the scope "in the coming months" with capabilities such as discovering related products or applying loyalty points.
- Google's Business Agent. Today, for eligible U.S. retailers: a branded agent that the merchant turns on in Merchant Center. Announced in the same text: that the agent could be trained on the retailer's data and process purchases, and dozens of new Merchant Center attributes for conversations, which Google planned to roll out first to a small group of retailers.
- AP2. Today: version 0.2.0, published on April 28, focused on flows without the person present; its initial version supports credit and debit card payments. Announced: e-wallets, instant transfers and digital currencies are on the roadmap, and standardization continues in working groups of the FIDO Alliance.
- MCP and A2A. Today: both have a published specification, the July 28, 2026 one for MCP and 1.0.0 for A2A. They are not buying protocols, so there is nothing to turn on in a store: they are the layer the others travel over. UCP declares itself compatible with A2A, AP2 and MCP, and ACP is published with a traditional API or with MCP.
- Shopify. Today: agentic storefronts are on by default in eligible stores and carry the catalog to ChatGPT, Google AI Mode, Gemini, Microsoft Copilot and Meta; the merchant chooses channel by channel under Sales channels, in the Agentic section. It is the shortest route for a store already on Shopify.
One pattern repeats at every stop: what is available is letting the agent read and recommend your catalog, and what is still announced, or open only to approved partners, is the agent completing the payment. That is why a store's first job is not choosing a protocol: it is making its catalog complete and consistent on every channel that reads it.
Start here
If you are starting from zero, read these three stops first: agentic commerce, ChatGPT Shopping and agentic checkout. Then go back to the full path wherever you left off.
Agentic commerce
The umbrella: the three stages of buying with agents, discover, recommend and pay, and who the seller is in each one.
ChatGPT Shopping
The case closest to a store: how a catalog reaches ChatGPT's shopping answers.
Agentic checkout
The last stage, and the least open: the payment an agent completes, with what is published and what is announced kept apart.
The path, step by step
-
Agentic commerce
Agentic commerce is a way of buying in which an AI agent acts for the buyer: it searches for products, compares them, recommends them and, when the platform and the store allow it, prepares or completes the payment.
-
Agentic AI
Agentic AI is a type of artificial intelligence system that, instead of just answering or generating content, pursues a goal with little human intervention: it plans the steps, uses tools to act in other systems, checks the result and adjusts the plan, coordinating one or more AI agents to do it.
-
AI agent
An AI agent is a software system that uses a language model to decide which steps to take and which tools to use to complete a task on a person's behalf, rather than just answering a question.
-
Autonomous vs assisted agents
Autonomous vs assisted agents is the distinction between two ways an AI agent can work, depending on who approves its actions with consequences: an assisted agent stops so a person confirms each important step, such as paying or sending a message, and an autonomous agent acts alone within limits that person set beforehand.
-
Shopping agent
A shopping agent is an AI agent that acts on behalf of the person buying: it takes an assignment, searches and compares products across several stores, prepares the cart and, if it has permission, completes the payment, normally after the person confirms.
-
AI product discovery
AI product discovery is the stage of commerce with AI assistants and agents in which the system finds which products answer a purchase request, working from the feeds and catalogs it receives from stores and the pages it crawls, and decides which ones to show.
-
Purchase intent
Purchase intent is a person's willingness to buy a product, expressed in what they search for or ask for.
-
Machine-readable catalog
A machine-readable catalog is a product catalog published in structured formats that a program can read without interpreting the page the way a person would: structured data on every product page, feeds for each platform and tools or APIs for AI agents.
-
ChatGPT product feed
The ChatGPT product feed is the file or connection a store uses to send its catalog to OpenAI so its products can appear in ChatGPT's shopping answers.
-
ChatGPT Shopping
ChatGPT Shopping is the set of shopping features in ChatGPT: when a question has purchase intent, ChatGPT shows products with an image, a price and the stores that sell them, compares them and links to the merchant's site to buy.
-
Agentic storefronts
Agentic storefronts are the Shopify feature that takes a store's products to AI channels, such as ChatGPT, Microsoft Copilot, Gemini and Google's AI Mode, through Shopify Catalog and its sales channels, without a separate integration for each one.
-
Tool calling
Tool calling, or function calling, is the ability of a language model to answer, instead of with text, with a structured request to run a function the application has described to it, with the function's name and its arguments.
-
MCP
MCP (Model Context Protocol) is an open protocol that standardizes how an AI application connects to external data sources and tools: an MCP server exposes resources, prompt templates and tools, and the application uses them through JSON-RPC messages.
-
MCP server
An MCP server is the program that offers tools, data and prompt templates to AI applications through the MCP (Model Context Protocol) protocol.
-
A2A
A2A (Agent2Agent) is an open protocol that lets AI agents from different vendors work together: one agent publishes an Agent Card listing what it can do, and another hands it tasks and gets the results back without seeing its tools or internal state.
-
Agent Card
An Agent Card is the JSON document an AI agent uses to introduce itself in the A2A (Agent2Agent) protocol: it declares its name, its version, its skills, the addresses it listens on and the authentication it requires.
-
Protocol gateway
A protocol gateway is an intermediary that sits between AI agents and the services they use and translates between protocols: it turns, for example, the operations of a REST API into MCP tools, or gives a single entry point to MCP, A2A and model traffic.
-
ACP
The ACP (Agentic Commerce Protocol) is an open standard, created by OpenAI and Stripe, that defines how an AI agent, the buyer and the store exchange data to complete a purchase: the product catalog, the checkout session and the delegated payment credential.
-
UCP
UCP (Universal Commerce Protocol) is an open standard, driven by Google and developed with Shopify, Etsy, Wayfair, Target and Walmart, that defines how an AI agent discovers what a store sells, builds the cart, pays and tracks the order.
-
AP2
AP2 (Agent Payments Protocol) is an open protocol, introduced by Google in September 2025, that lets an AI agent pay on a person's behalf and prove to the store and the payment network that the person authorized the purchase, using signed mandates that fix what is bought, for how much and under what conditions.
-
AP2 mandate
An AP2 mandate is a digitally signed record that proves what a person authorized when an AI agent buys on their behalf using the AP2 protocol (Agent Payments Protocol).
-
Delegated payments
Delegated payments are payments that an AI agent starts on behalf of a person with an authorization limited in advance, without receiving their full payment details.
-
Agentic checkout
Agentic checkout is the payment phase of agentic commerce: the moment an AI agent completes the purchase on the shopper's behalf, inside the assistant itself or on the store's site, with that person's confirmation or prior permission.
-
Instant Checkout
Instant Checkout is the ChatGPT feature that lets someone buy a product without leaving the chat: the person confirms the order, shipping and payment, and ChatGPT passes that data to the store with the ACP protocol so the store charges and ships.
-
Agentic browser
An agentic browser is a web browser, or a mode of one, in which an AI agent works through pages on a person's behalf: it opens tabs, clicks, fills in forms and adds products to the cart to finish a task.
-
Coupons for agents
Coupons for agents are the discount codes and promotions an AI agent applies when it buys on behalf of a person.
Make your catalog easy for bots to read
Clione fills every product page with what bots read best: the JSON-LD graph with its structured data, the meta description and the FAQs. You review and press Push; then we check your live store so it all stays published.