On September 2, 2026, Anthropic released a blueprint for building "commerce agents" on Claude: AI assistants that search a catalog, build a cart, and hand a shopper to checkout inside one conversation. That's the launch, and it's the news here.
Shopify's part is narrower but more interesting than it first looks. Shopify published open-source example code that runs Anthropic's blueprint against a real Shopify store. The repository holds two of them: a shopping agent for the storefront side, and a merchant agent that reads and updates the store through the Admin API.
Neither is an app you install. Both are code for engineering teams to build on. There is still one thing a non-technical merchant can switch on today, and it comes straight from Shopify's own setup instructions.
This covers what shipped as of September 4, 2026. Both repositories are active, and the merchant-facing tooling around them is moving quickly, so treat the specifics below as a snapshot rather than a settled state.
1. What Anthropic actually announced
A commerce agent searches a store's catalog, compares products, builds a cart, and hands the shopper to checkout, all in a single conversation. No clicking through category pages. No hunting through an FAQ for the return policy. The shopper asks, and the agent does the legwork.
The blueprint splits this into two agents.
The shopping agent lives inside a store's app or website. It searches the catalog and assembles multi-item requests ("a tent, sleeping bag, and stove for a weekend with two kids"), remembers what a customer likes, shows products and comparisons inside the chat rather than as plain text, builds the cart, and answers order and return questions in the same conversation.
(Source: Claude.com) The example for a shopping agent running in the store: ACME
The merchant agent works the business side. Someone on staff can ask what to discount to clear last season's inventory and get an answer based on the store's own sales history. It also tracks inventory, flags problems like an item about to sell out mid-promotion, and drafts campaigns.
Both ship with guardrails. The shopping agent is constrained to real catalog data and real prices, and it's built to avoid manipulative upsell patterns. On the merchant side, when the agent proposes a change, a person approves it before anything goes live.
The blueprint covers retail, travel, telecom, and ticketing, and it deploys wherever you already build with Claude: the Claude API, Amazon Bedrock, Microsoft Foundry, or Google Cloud Vertex AI.
2. The numbers Anthropic is pointing to
Anthropic didn't launch this as a thought experiment. Retailers already running shopping agents on Claude have seen carts up to 35% larger and shoppers 60% more likely to complete a purchase, according to Anthropic.
Those are Anthropic's figures for their own enterprise customers, so read them as direction rather than a promise for a 40-product Shopify store. Still, the named list is worth noting: Priceline, Wix, Square, Klaviyo, Intuit, and Zomato all appear as builders, with Visa, Mastercard, and Accenture working as partners. These are businesses running this in production, not pilots. What they report moving is AOV and conversion rate, not support ticket deflection.
3. What Shopify built: two agents, not one
Shopify hasn't shipped a new app or a toggle in your admin, and the difference matters if you're deciding how much attention to give this. It published open-source example code showing how a store connects to Anthropic's blueprint. The repository holds two separate examples.
The storefront example is a shopping agent pointed at a real store. It uses Shopify's UCP endpoints to read the live catalog, so it searches actual products at actual prices, builds a real cart, answers from the store's own policies and FAQs, and hands the customer to that store's hosted checkout page. The web app even brands itself from the store: name, logo, colors, best sellers. Optional Sign in with Shop lets a returning buyer get personalized results instead of starting cold.
Two limits are worth understanding. It never places an order or takes payment; checkout, shipping, and payment all stay on Shopify's own pages. And order tracking is narrower than it sounds: the agent can only see orders placed through that agent, only when the buyer asks, and only if the credentials carry the right scope. Without it, the agent tells the customer order tracking isn't available and points to the confirmation email.
The merchant example is the other side of the same store, running over the Shopify Admin GraphQL API. It reads products, orders, and inventory, proposes changes, and writes back only what the operator approves. The approval gate is real rather than cosmetic: the agent's staging tools send no mutation at all, and a change only applies when someone explicitly approves it. This example ships no interface, though. The operator-facing app is yours to build.
Shopify's VP of Product, Vanessa Lee, framed the intent simply: We want our merchants to be everywhere customers are shopping, and increasingly that means a conversation with an agent.
Learn more: Agentic checkout: how Shopify merchants stay in control
4. What you can actually turn on today
If you're picturing an app you install this afternoon, the blueprint isn't it. Anthropic's blueprint and Shopify's examples are both for engineering teams to build and customize.
Shopify's README opens with a note that's easy to skip past. If you don't have a Claude account and want an agent answering customer questions on your storefront right now, Shopify points you to Shopify Inbox. That's a real install, available today, no developer required.
You can use Shopify Inbox for a simple setup
Inbox is not the same thing as the blueprint. It answers questions on your store; it doesn't do multi-item cart building against your live catalog with the guardrails Anthropic shipped. Think of it as the entry-level version of the same idea, and the honest answer to "can I do anything today?"
For everything above that tier, three things are true at once:
- This is moving quickly, and Anthropic timed the release for holiday planning.
- Agencies and app developers serving Shopify stores are the ones most likely to turn this into merchant-facing tools next.
- The stores that benefit most when those tools arrive will be the ones whose catalog, policies, cart, and checkout are already clean.
That third point is the one you can act on regardless.
5. How this fits with what Shopify already shipped this year
These are two different things, and they get conflated constantly.
One is selling through someone else's AI. That's Shopify Catalog and UCP putting your products inside ChatGPT, Copilot, and Gemini, where the shopper never touches your storefront. That layer is largely on by default, and your main job there is product data quality.
The other is running your own agent on your own storefront, with your branding, your policies, and your checkout. That's what this week's launch is about. You control the experience instead of renting a slot in someone else's.
Most merchants will end up doing both. They're not competing options.
Learn more: Shopify Spring 2026 Edition: the AI updates that matter most to merchants
6. How to get your store ready before you touch any of this
An AI shopping agent is only as good as the data it can read. Before any agent (Shopify's, a third party's, or your own) can represent your store well, a few things need to already be in order.
- Accurate, well-organized product data. Agents search your catalog the way a person would, just faster. Titles, variants, and options need to actually describe what's for sale. Thin or inconsistent product data is the single most common reason an agent misrepresents a store.
- Written policies an agent can find. Return policy, shipping timelines, FAQs. If an agent is going to answer "what's your return policy?" mid-purchase, that answer needs to live somewhere structured, not buried in a PDF or a support inbox.
- A clear, consistent cart and checkout. Every one of these agents ends the same way: it hands the shopper to your checkout. If your checkout is cluttered or off-brand, that handoff is exactly where a customer notices the seam.
- Cart and upsell logic that doesn't depend on a human spotting the opportunity. Agents avoid pushy patterns by design, but they still surface add-ons and bundles when the logic supports it. Stores with a clean upsell setup translate into this channel more naturally.
None of that requires betting on AI agents. It's ordinary store hygiene that happens to be exactly what makes a store agent-ready.
Qikify's solutions cover a fair amount of that groundwork. Sticky Cart keeps your cart and upsell logic clean and visible. Contact Form Builder gives your policies and customer questions a structured home instead of an email thread. And for Shopify Plus merchants thinking about a branded, agent-ready checkout, Checkout Customizer puts that experience under your control rather than Shopify's default.
Ready to get your foundation in shape before agentic shopping goes mainstream? Try Qikify's app suite free on your Shopify store today.
FAQ
1. What's the difference between a commerce agent and a regular AI chatbot?
A chatbot mostly answers questions. A commerce agent searches your live catalog, builds a real cart, and hands the shopper to checkout, all in the same conversation.
2. Can I install this on my Shopify store today?
Not as a one-click app. The blueprint and Shopify's examples are built for engineering teams. If you want an agent answering customer questions on your store right now, Shopify points merchants to Shopify Inbox as the no-code option.
3. Does the AI agent process payment?
No. In Shopify's example, the agent builds the cart and hands the shopper to the store's own hosted checkout. Payment and order completion happen there, on Shopify's pages.
4. Can the agent tell my customers where their order is?
Only in a limited way. In Shopify's example, the agent sees orders placed through that agent only when a buyer asks, and only with the right API scope. Without it, the agent says tracking is unavailable and points to the confirmation email.
5. Will an AI agent push extra products on my customers without my say-so?
The blueprint includes guardrails against manipulative upsell patterns. On the merchant side, any pricing or promotion change the agent proposes needs a person's approval before it goes live.
6. How is this different from selling through ChatGPT or Copilot?
Those channels put your products inside someone else's assistant through Shopify Catalog and UCP. This launch is about running your own agent, on your own storefront, with your branding and your checkout. Most merchants will do both.
7. What should I fix first?
Product data. Clean titles, complete variants, accurate stock. Then written policies and FAQs, then your cart and checkout consistency.
