Let the agents in, but watch who sends them

Walmart said on 21 October that it will let shoppers send their own AI agents to buy for them. The feature is called Agent Profiles. You connect the assistant you already use, and it can search and buy across Walmart and Sam’s Club, either asking you to approve each purchase or spending on its own up to a cap you set. The agent never sees your card details. For developers there is a gateway, still in testing, that lets approved agents talk to Walmart’s stock directly. Axios has the detail.

Amazon is doing the opposite. It blocked Meta’s Muse agent in September, and it has been in court with Perplexity since last November over the shopping agent in Perplexity’s Comet browser. Amazon’s stated reasons are about security: agents logging into customers’ Prime accounts, bots dressed up as ordinary browsers, and customers not being told that a machine was acting for them.

Why does Amazon say no?

Some of that is fair. But Amazon also has Rufus, its own shopping assistant, and every purchase that goes through Rufus stays inside Amazon. My read is that this is less about security than about precedence. If customers get used to sending someone else’s agent to Amazon, Amazon becomes a warehouse that other people’s AI shops from, and it loses the part it cares about most, which is being the place where the decision gets made.

There is a second reason worth saying out loud. A large part of Amazon’s business is advertising, the sponsored listings a person scrolls past on the way to the thing they wanted. An agent doesn’t scroll, and it isn’t swayed by a sponsored slot. A shopper who never looks at the page is a shopper you can’t sell ad space against.

Should a premium brand let agents shop its site?

For a premium fashion or sports brand, I think the answer is yes. It helps sales, and materially. An agent that can find the right size of a jacket in stock and buy it while the customer is in a meeting is a sale that would otherwise have died in an abandoned basket. Walmart’s controls, approval by default and a spending cap the customer sets, are a sensible template for anyone building this.

None of the brands I’ve worked with has raised it yet. That will change quickly, because customers will get there first, as I wrote when the checkout started leaving the website.

Who is the agent working for?

The worry I’d have isn’t the customer’s own agent. It’s the agents sent by somebody else.

Fashion already has aggregators. Lyst built a business on pulling thousands of retailers’ stock into one search. Now imagine that job done by an agent, on every visit, for every shopper, reading your site to tell the customer whether a stockist or a marketplace has the same item for less. That agent isn’t your customer and it isn’t neutral. It works for whoever built it, and its job is to find a better deal than yours.

Premium brands put a lot of effort into keeping prices consistent across stockists, because a discount in one place teaches the whole market to wait for the sale. Agents that read every site at once make any gap visible in seconds. A stockist quietly running an unauthorised discount becomes the default answer for every shopper whose assistant was asked to find it cheaper.

So open the door, but know who’s knocking. Agree which agents you welcome, watch which ones are reading your prices, and tighten your stockist agreements before the bots find the gaps for you. Otherwise your full-price customer gets told, politely and instantly, where to pay less.