AI shopping agents: your shop has to be readable too

What Walmart has announced

On 21 October, at the Masters of Scale Summit in San Francisco, Walmart's chief executive, John Furner, unveiled Agent Gateway, according to TechCrunch. It is an MCP server: MCP is an open protocol, a kind of common plug that lets AI assistants connect to other systems. Through it, approved agents can search Walmart's and Sam's Club's online inventory and buy on behalf of their user.

• Agent profiles: each customer authorises each agent with its own profile, and its purchases show up separately. They arrive next month, according to TechCrunch.
• Spending control: the customer chooses between approving every purchase or letting the agent buy within a limit they set themselves.
• No extra fee: according to Storyboard18, anyone with a Walmart account can use it, with no need for a Walmart+ subscription.
• Rules for agents: they cannot go beyond what the customer asks, cannot access payment details or information they do not need for the task, and any agent that breaks the rules risks losing access, according to Storyboard18.
• Same prices: agents see the same prices as a human shopper, according to TradingView.

If an agent does not do what was expected, the usual returns policy applies, TechCrunch reports. The same outlet notes that Amazon has started blocking AI agents that buy for shoppers, and that Walmart has taken the opposite approach.

The roaster selling coffee to someone who never sees the photos

Imagine Nuria roasts speciality coffee in a small workshop with her business partner and sells it through their online shop. They take care over every detail: beautiful photos of the bags, copy about the farm each bean comes from, a friendly tone. Customers who find them keep coming back.

But the site has grown piece by piece. The roast level of each coffee is printed on the bag in the photo, not written in the text. When a batch sells out, they post about it on Instagram. The delivery cost to the Canary Islands only appears at the very end of checkout, and the returns policy sits in a paragraph nobody has reread since it was written. A person copes. An assistant asked to “buy a medium roast for espresso that arrives before Friday” may well not.

Why are so many online shops hard to read?

The scene is made up, but it looks like many shops we see. An online shop is designed for the eye: the home page, the photos, the buy button. The less visible part, the data behind each product, gets filled in quickly or left for later.

For years that did no harm, because the buyer was a person who looked at the photo, skimmed the page and sent a WhatsApp message if in doubt. An AI agent works differently: it reads what the site states explicitly and compares. Anything that only lives in an image or an Instagram story simply does not exist for it.

What it costs when an agent cannot understand you

The risk is not that agents will turn up at every shop tomorrow. It is that, when they do, they will pick whoever is easiest to understand. If your product page does not give the weight, the grind or the delivery time, an assistant comparing five shops may choose one that does, even if your coffee is better.

There is a quieter cost too: messy data confuses people as well. A customer who cannot find the delivery price abandons the basket just as an agent would. Tidying your shop's information for AI shopping agents tends to improve things, along the way, for anyone buying with their own hands.

What can a small shop do today?

You do not need to connect your shop to any protocol to get started. These steps work for people and AI assistants alike:

• Put into words everything that is currently only in a photo: size, weight, variety, what each product is for.
• Keep stock up to date on the site itself, not on social media.
• Show the delivery price and delivery time before the end of checkout, by area if it varies.
• Write the returns policy in short sentences in one fixed place, linked from every product.
• Use the same names across the site: if a coffee is “medium roast” on one page, it should not be “light roast” on another.
• Run the test: ask an AI assistant to find a product like yours and see what it makes of your product page.

In Nuria's workshop, the first step would be to write out the roast and use of their six best-selling coffees and add delivery details to each page. It is the kind of small change both types of visitor notice.

How we help at digizone

In our UX/UI design work we design every screen for the person who will use it, and today that includes whoever reads it through an assistant. We go through how your shop's information is organised with you, what each product page is missing and how it is presented, so it makes sense first time, whether a person looks at it or an agent reads it.

For Go Sushing we built the app and website customers order from, a dashboard for the team to manage content, orders and the chain's outlets, another for each restaurant to handle its orders with integration into its till system, and an app for delivery riders. Each role sees exactly the information it needs to work.

Start with your six best sellers

There is no need to rebuild the whole shop or wait for agents to reach your sector. The first version can be as simple as getting the pages for your best-selling products spotless, with every detail written down, and checking what an assistant makes of them. If it works, you repeat it across the rest of the catalogue.

If you would like to see how it would work for you, our free workshop, in Spanish, walks through it step by step.