Loblaw is the first Canadian retailer inside Muse. The basket is the new storefront

A shopper walking a grocery store aisle with a basket

The storefront moved again. On November 18, 2026, a Canadian shopper could open an AI agent, ask for a week of meals, and end at a grocery checkout without opening a store app or a store website.

Direct answer: Loblaw made its product catalogue available inside Muse, Meta’s AI agent, on November 18, 2026. It is the first Canadian retailer to do so. Shoppers at participating banners can plan meals, build a basket and check out inside Muse. The catalogue is now the shelf, and the agent does the walking.

That last line is the part worth a morning of thought. For 30 years, retail online meant a search box and a grid of photos. The shopper did the finding. The shopper did the comparing. The shopper did the remembering.

Advertisement

An agent reverses the labour. The shopper states a goal. The agent searches the catalogue, picks products, and builds the basket. Your product page now has 2 readers, and the second one never looks at your photos.

What did Loblaw and Meta actually switch on?

Loblaw joined Muse as a partner connector. A connector is the pipe between the agent and the catalogue. With that pipe open, a shopper can go from store selection to checkout at participating Loblaw banners inside Muse.

The flow Meta describes is plain. Ask Muse to help with meal planning. Muse builds the basket from the catalogue. The shopper checks out in the same conversation.

Muse also remembers. It can hold brand preferences and details such as food allergies, so next week’s planning starts from what this week taught it. That memory is the feature a store app never earned. Most shoppers never set a preference inside a grocery app. They state preferences in plain words when an agent asks.

The safety shape matters to any operator watching this. Muse runs on a secure system that keeps passwords and payment methods out of the agent’s view. The shopper chooses what Muse can reach. Muse checks with the shopper before a sensitive step, such as making a purchase.

What does an agent see when it reads your catalogue?

An agent reads fields, not merchandising. It reads the product name, the size, the price, the stock state and the store. It reads the attributes you filled in and skips the ones you left blank.

A human shopper forgives a thin listing. The photo carries the product. The shelf position does quiet work. A person can guess that the store brand version sits beside the national brand and costs less.

An agent forgives nothing. A missing size field is a product it cannot compare. A vague name is a product it cannot match to a meal plan. A stale price is a basket that fails at checkout, in front of a shopper who blames the store.

This is where the Canadian angle sharpens. Loblaw runs more than 1 banner family at more than 1 price tier. A discount banner and a full service banner sell the same category at very different prices. An agent that plans a meal has to know which shelf it is shopping from, or the basket it builds fits nobody’s budget.

The rule we set out in When an agent buys for you, set the authority before the first order now runs in reverse. That piece told buyers to set limits before an agent spends. This launch asks sellers to set facts before an agent shops. The authority problem and the catalogue problem are the same problem seen from 2 sides of the till.

Which questions did the launch leave open?

Read both releases side by side and 3 gaps stand out. Neither release names the participating banners. A shopper cannot tell from the announcement whether their own store is in.

Neither release states how loyalty points work inside an agent checkout. Points drive banner choice for a large share of Canadian grocery shoppers. An agent basket that earns nothing trains shoppers to check out somewhere else.

Neither release states the delivery and pickup terms. The announcement covers store selection to checkout. It says nothing about how the basket reaches the home, what it costs, or which provinces are live on day 1.

These gaps are normal for a launch week. They are also the whole negotiation for the next retailer in line. The second and third Canadian retailers to sign a connector deal will sign against answers to these 3 questions, because Loblaw’s deal set the shape and left the terms blank.

The approval design is the 1 part with a clear answer today. ChatGPT Wallet lets an agent pay. Approval is the whole contract made the same point on a rival platform. Muse also checks before it buys. 2 competing agents landed on the same rule, and that tells you where the industry floor now sits. An agent that spends without a human yes is a product no retailer will attach its name to.

What should a Canadian operator do with this launch?

Start with your own catalogue, not with a connector contract. Pull your top 200 products by weekly sales. Read each listing the way an agent reads it. Name, size, price, stock, store, and nothing else.

Count the listings with a missing or vague field. That count is your agent readiness score today. A human team can sell around a weak listing for years. An agent simply picks the competitor product with the complete record.

Next, write the preference facts your shoppers repeat. Brand loyalties, allergies, household size, the store they actually use. If your own channels never captured those facts in a field, an agent channel will capture them first, and the memory will live on the platform’s side of the pipe.

The access discipline in Give your AI agent an 8-line job description before access applies to the connector too. Decide what the agent may read, what it may build, and what it may never do without a person. Write that down before a platform rep writes it for you.

Last, pick 1 category and run the agent yourself. Ask Muse to plan 3 dinners and build the basket at a participating banner. Time it. Read the substitutions. Check the total against your own app. You will learn more in 20 minutes of your own basket than in a month of launch coverage.

The adoption pattern in The 5 jobs you keep putting off. Hand 1 to AI this week holds here too. Start with 1 finished workflow. A meal plan that ends in a real checkout is a finished workflow. A slide deck about agentic commerce is not.

Recap: Loblaw put its catalogue inside Muse on November 18, 2026, the first Canadian retailer to do so. The agent plans, picks and builds the basket, and the shopper approves the purchase. The launch leaves the banner list, loyalty treatment and delivery terms unstated. Your catalogue data now does the selling a shelf used to do.

This week, pull your top 200 listings and score each 1 on name, size, price and stock. Fix the weakest 20 before you take any meeting about a connector.

Total
0
Shares
Prev
One prompt finds every AI tool you are wasting money on
A laptop and paper receipts on an office desk

One prompt finds every AI tool you are wasting money on

The renewal email sits in your inbox

Next
Gemini 4 Neon costs $4.30 CAD. Cost per task is the number to test
Programming code glowing on a computer monitor

Gemini 4 Neon costs $4.30 CAD. Cost per task is the number to test

Google priced Gemini 4 Neon at $3 per million input tokens and $15 per million

You May Also Like