What is agentic commerce on Shopify?

UCP is the one that matters most for sales today. It is plain HTTP, so an agent does not need to load your theme at all. It reads the same catalog your storefront uses and returns a cart with a link that drops the shopper into your normal checkout.

1. Discovery. Your store publishes a profile at /.well-known/ucp listing what it supports.
2. Catalog. The agent calls search_catalog and get_product to find products, prices, variants and stock.
3. Cart. create_cart returns a cart and a continue_url the shopper can open.
4. Checkout. Checkout tools exist too, but they need an authenticated agent. Payment always stays with the shopper.

In our 30 September 2026 probe of eight Shopify stores, all eight answered every step up to a working checkout link, in about one and a half seconds each. That included stores whose storefronts block automated browsers, because UCP never touches the storefront page.

A UCP product includes a title, description, price range, variants, availability, collections and tags. It has no field for subscriptions, quantity discounts, bundle contents or your free delivery threshold. Those offers reach an agent only if they are written into product titles or descriptions. In our tests, an agent asked for "the cheapest way to buy this every month" found a subscription only on the one store that put "Monthly Subscription" in the product title.

That is the practical merchandising question for agentic commerce: not "are agents here yet?" but "do my offers survive when the shopper is an agent?"

• The agent path works. Discovery, catalog, cart and checkout link, on a schedule, with evidence when a step fails.
• The human path still works. Agents hand shoppers to your normal checkout, so a broken checkout breaks both.
• Your offers are visible. Check that bundles, subscriptions and price breaks are described where an agent can read them.

Revenue Shield runs a daily AI checkout check over UCP alongside its desktop and mobile checkout tests. It is a protocol check: it confirms the technical path works, not how any particular assistant chooses products.