Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarMeta Corporate News→ original

Meta taught its Muse agent to fill a grocery cart at Loblaw and check out

Meta has connected its Muse AI assistant to the first Canadian retailer: Loblaw's catalog is now a partner connector. A user asks for a weekly meal plan, and the assistant builds a cart from the chain's stores and completes the order right in the chat. It remembers favorite brands and food allergies, so planning the next week goes faster. For agent developers, this is a template: shopping is built into the conversation, while passwords and cards stay hidden from the model.

AI-processed from Meta Corporate News; edited by Hamidun News
Meta taught its Muse agent to fill a grocery cart at Loblaw and check out
Source: Meta Corporate News. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceMeta Corporate News ↗
Editor-in-Chief
Key metric #1Loblaw is the first Canadian retailer among Muse's partner connectors
Key metric #2Passwords and payment data live in Muse Secure VM, out of the agent's view
Key metric #3Before sending an email or making a purchase, Muse asks the user for confirmation

Grocery shopping is one of the most ordinary scenarios where AI agents get tested in real life. The agent has to decide what to cook, account for allergies and family habits, find products in a store and pay. Meta has moved in this direction: the Canadian retail chain Loblaw has become the first retailer in the country to connect its catalog to the Muse AI assistant as a partner connector.

What has appeared in Muse

A connector is a bridge between the assistant and an external service. Muse can already connect to apps people use every day: Gmail, Google Calendar and Instagram. Through them, it helps sort through email, coordinate schedules and draw on saved content when a routine task needs to be handled.

Now the Loblaw catalog has joined that list. Shoppers in Canada who use the chain's participating stores can ask Muse to help plan a menu, build a cart and place an order. All of this happens inside Muse itself, without switching to the retailer's website. For example, you write that you need dinners for three for the week, without nuts, and the assistant suggests a menu, picks products from the catalog and takes the task through to payment.

Memory is another detail. Muse remembers brand preferences and important information such as food allergies. For weekly shopping, this matters: you don't have to explain every time which yogurt you buy or what you can't eat.

How Meta protects passwords and payments

Any agent that can buy immediately raises the question: what if it makes a mistake or gets tricked? Meta answers with architecture. Muse runs on Muse Secure VM, a dedicated cloud computer. It stores the data and credentials for connected services, while passwords and payment methods stay out of the agent's view. In other words, the language model that makes decisions cannot see your card number or your email password.

In addition, the user decides what their Muse can access. Before sensitive actions, such as sending an email or making a purchase, the assistant requests confirmation. For the commerce scenario, this is an important safeguard: the final "pay" button remains with the human. Meta published details of its security approach in a separate post, but this article names only the measures listed above.

What this means for developers and product managers

First, retail is becoming a channel for agents. If an assistant can work with a catalog, a cart and order checkout, it benefits the retailer to be inside that interface, not only in its own app. Loblaw is an early example of such a partnership in Canada.

Second, the separation of roles works as a template. The model plans and chooses, while an isolated environment stores secrets and performs sensitive operations. Teams building their own purchasing agents should design for this from the start: don't pass payment data to the model, and require explicit confirmation before irreversible actions.

Third, personal memory makes an agent more useful but raises the cost of mistakes. Allergies and preferences are data that determine the outcome of an order, so users need to see and control what has been remembered.

Limitations

The announcement does not make clear how many of the chain's stores already participate, how delivery and payment work, or how accurately Muse matches products to dietary restrictions. Meta did not provide quality metrics or usage statistics. So for now this is more a demonstration of a direction than proven savings in time or money. Still, the combination of a retailer's catalog, remembered preferences and an isolated payment environment already looks like a working model for how agents will handle everyday shopping.

Story

Key Questions Answered

What has appeared in Muse?

A connector is a bridge between the assistant and an external service. Muse can already connect to apps people use every day: Gmail, Google Calendar and Instagram. Through them, it helps sort through email, coordinate schedules and draw on saved content when a routine task needs to be handled. Now the Loblaw catalog has joined that list. Shoppers in Canada who use the chain's participating stores can ask Muse to help plan a menu, build a cart and place an order.

How Meta protects passwords and payments?

Any agent that can buy immediately raises the question: what if it makes a mistake or gets tricked? Meta answers with architecture. Muse runs on Muse Secure VM, a dedicated cloud computer. It stores the data and credentials for connected services, while passwords and payment methods stay out of the agent's view. In other words, the language model that makes decisions cannot see your card number or your email password. In addition, the user decides what their Muse can access.

What this means for developers and product managers?

First, retail is becoming a channel for agents. If an assistant can work with a catalog, a cart and order checkout, it benefits the retailer to be inside that interface, not only in its own app. Loblaw is an early example of such a partnership in Canada. Second, the separation of roles works as a template. The model plans and chooses, while an isolated environment stores secrets and performs sensitive operations.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…