Agentic Commerce Explained: How It Works for Online Stores

Agentic commerce is a way of shopping in which an AI agent does part or all of the work for the buyer: it discovers products, compares them, and in some setups completes the purchase. The shopper states a goal ("a waterproof jacket under 150 dollars, delivered by Friday") and the agent handles the search and the steps. The store is still the seller. This guide explains the definition, the mechanics, real examples and the limits, as of October 2026. This area changes quickly, so check vendor documentation before acting on any detail here.

What is the agentic commerce definition?

A working definition: agentic commerce is shopping where an AI agent acts on behalf of a user to discover, evaluate and sometimes buy products from merchants, while the merchant remains the seller.

Two parts matter:

• Agent. Software that takes a goal and carries out several steps, rather than only answering a question.
• Merchant of record. The business legally selling the product and handling payment, tax, fulfilment and returns. Google states that merchants remain the merchant of record under its Universal Commerce Protocol (UCP). The Agentic Commerce Protocol (ACP) is described the same way.

Agentic commerce is not the same as a chatbot on your website. A site chatbot helps visitors who are already in your store. An AI shopping agent works outside your store, in an assistant the shopper already uses, and your products have to be readable to it.

How does agentic commerce work?

It helps to think in three layers.

1. Product data and discovery. The agent needs structured, accurate information: titles, attributes, prices, stock, shipping and return terms. This comes from product feeds, catalogs and tools such as Google Merchant Center. Poor data means the agent cannot match your product to a request.
2. Checkout protocol. A protocol is a shared set of rules so that any agent can talk to any store. Two main ones exist: ACP (from OpenAI and Stripe) and UCP (from Google with Shopify). They define how an agent reads a cart, offers shipping options and submits an order.
3. Payment credential and trust layer. The agent should not see raw card numbers. Instead it uses tokens limited to a specific amount and merchant, with the user's permission. OpenAI describes this as encrypted tokens authorized for specific amounts and merchants, with the user confirming each step and only minimum data shared with the merchant. Fraud signals and card-network programs sit in this layer too.

A simplified flow looks like this:

1. The shopper asks an assistant for a product.
2. The agent searches catalogs and feeds, and shortlists items.
3. The shopper picks one and approves the purchase.
4. The agent calls the store's checkout through a protocol, passing a limited payment token.
5. The store processes the order as usual and ships it.

What does an AI shopping agent actually do?

An AI shopping agent can:

• Interpret a vague request and turn it into product criteria.
• Compare items across merchants using catalog data.
• Remember preferences within the product's limits.
• Start a checkout, with the user's approval.

It cannot reliably:

• Fix missing or wrong product data. It reads what you publish.
• Guarantee that your product will be shown. Ranking is controlled by the platform.
• Replace your customer service, returns policy or fraud checks.

Most current implementations keep a human confirmation step before payment.

What are real agentic commerce examples?

ChatGPT: Instant Checkout was retired

OpenAI launched Instant Checkout on 29 September 2025, first with Etsy and later a small number of Shopify brands, built on ACP. According to reporting, OpenAI retired it in March 2026 and now focuses on product discovery, letting merchants use their own checkout. OpenAI's reported statement was that the first version did not offer the flexibility it wanted. Secondary sources say only about a dozen Shopify merchants ever went live. Many older articles still describe Instant Checkout as live, so treat them with caution and verify the current state in OpenAI's own documentation.

Shopify Agentic Storefronts

Shopify says that starting the week of 24 March 2026, merchants could sell to users of ChatGPT, Microsoft Copilot, AI Mode in Google Search and the Gemini app, managed from Shopify Admin. According to Shopify, it is activated automatically for eligible merchants, products syndicate through Shopify Catalog, and there are no apps or additional fees. Controls sit under Settings, Sales Channels, Agentic Storefronts. Shopify also states that purchasing across these channels is currently limited to US buyers.

Shopify also describes an Agentic plan for sellers who are not on Shopify. It lets them add products to Shopify Catalog and sell through agentic storefronts without a Shopify online store or migration. Sources disagree on the price, so check Shopify's pricing page.

Google UCP in AI Mode and Gemini

UCP was announced on 11 January 2026 and co-developed by Shopify, Etsy, Wayfair, Target and Walmart, with endorsements from payment and retail names including Adyen, American Express, Best Buy, Mastercard, Stripe and Visa. Merchants publish a machine-readable profile of what they support (checkout, orders, catalog) and which payment handlers they accept, such as Google Pay or Shop Pay. Agents read that profile and negotiate what to use. Google's merchant documentation lists AI Mode in Search and the Gemini web experience as the places UCP checkout runs, with the Gemini app described as coming soon. That page may be out of date, so recheck it. Merchants use their existing Merchant Center feeds.

How do the protocols compare?

Name	From	What it does
MCP (Model Context Protocol)	Anthropic, open standard	Connects an assistant to tools and data, such as reading your store or reports
ACP (Agentic Commerce Protocol)	OpenAI and Stripe	Agent-initiated checkout and payment delegation
UCP (Universal Commerce Protocol)	Google with Shopify	Discovery and checkout between agents and merchants
AP2 (Agent Payments Protocol)	Google	Secure payments by agents; UCP is described as compatible with it
A2A (Agent2Agent)	Google	Lets agents communicate with each other

MCP is the one most store teams will touch directly, because it lets an assistant such as Claude read your catalog or reports. It is not a checkout standard. For the basics, see "What Is MCP? A Plain-English Guide for E-commerce Merchants". Reports suggest ACP's latest specification also covers integration with MCP, but verify that in the ACP repository before relying on it.

What should a store do about agentic commerce now?

1. Clean your product data. Accurate titles, attributes, GTINs (product barcodes), prices and stock matter more than any new tool.
2. Check your feeds. Confirm Merchant Center is error free and shipping and return policies are complete.
3. If you use Shopify, review Agentic Storefronts. Look at the settings, see which products are syndicated, and exclude any you do not want sold through AI channels.
4. Confirm your payment provider's position. ACP supports processors other than Stripe through a delegated payments specification, and UCP lets each provider publish its own payment handler. Ask your provider what it supports.
5. Track the orders. Check how AI-originated orders appear in your analytics and order tags, and compare margins, not only revenue.
6. Start small. Treat it as an extra sales channel to test, not a replacement for your own site.

Developers working with Shopify's MCP tools should also read "Shopify MCP Server: What Dev MCP and Storefront MCP Do". One secondary source claims older Storefront MCP catalog and cart tools are being replaced by UCP on a set timeline. That is unverified, so confirm in Shopify's developer documentation on shopify.dev.

What are the risks and limits?

• Volatile products. Instant Checkout shows that features can launch and disappear within months.
• Thin evidence of volume. Reported adoption has been small. Do not plan revenue around it without your own data.
• Less control over presentation. The agent decides how your product is described and ranked.
• Regional limits. Shopify's current availability is for US buyers.
• Margin and fees. Fee terms have changed between platforms. A 4% merchant fee was reported for Instant Checkout, but that product is retired, so confirm current terms.
• Disputes and returns. You remain merchant of record, so chargebacks and returns remain your responsibility.
• Data sharing. Understand what customer data reaches you and what stays with the platform.

What happens next?

Expect more protocol convergence and more assistants supporting checkout, but the details are unsettled. The practical position is to keep data accurate, use the channel controls your platform provides, and measure results before investing heavily.