The split between ecommerce platforms and AI agents highlights a growing question: Who controls the path to purchase?

• Last week, Amazon blocked Meta's Muse AI agent from its store. 
• Meanwhile Meta has announced partnerships with Walmart, GameStop, Sephora, Expedia, OpenTable, and Shopify with its new agent Muse. 

The numbers behind the standoff

Muse, a personal AI agent that carries out tasks like searches and shopping on its user’s behalf, has already proven popular. It launched in the US on September 8 and passed 2.5 million downloads by September 21, with an estimated 448,000 daily users by day 10 per Sensor Tower. 

The number that matters for retailers is potential reach. Muse sits inside WhatsApp, which has more than 3 billion monthly users, per Meta CEO Mark Zuckerberg. Muse is restricted to users 18 and older and currently limited to the US and Canada, so actual users will be a fraction of that ceiling.

Amazon's stake is larger. Third-party analysts estimate Its catalog is estimated at 350 million to more than 600 million listings, and roughly 588 million per SmartScout of the higher estimate come from third-party sellers. Amazon doesn't publish a count, so trackers disagree, but the scale is clear. A catalog that size is hard to browse, so shoppers lean on search, and search is where Amazon earns. Advertising brought in $19.8 billion in Q2 2026, up 26% YoY.

Security or control?

Amazon's decision to block Muse in part came from a lack of notice about the agent accessing its marketplace and concerns about users' accounts, according to a statement the retail giant provided to CNN. 

Jessica Wagner, senior vice president of product and digital transformation agency Fueled, thinks that matters: an unidentified agent inside customer accounts is "a major problem for any retailer, big or small."

Others see a second motive. Rutgers Business School professor Chris Ribeiro, who specializes in marketing, says the security argument is cover, and that "it is really about encroachment." In his view, Amazon doesn't want bots crawling its site without taking a cut. 

Amazon isn't anti-agent; it runs its own called "Buy for Me". Amazon being a player in the shopping agent market means it will most likely prefer its own solution over third-party products.

What agents do to retail media

Dima Okuniev of paid social and marketing firm iCanvas zeroed in on the risk. Amazon isn't afraid of losing a toaster sale, it's afraid of "losing the few minutes before someone buys the toaster," Okuniev said. That's where sponsored listings do their work. An agent could send the shopper straight to one product, and the sale would survive while the ad exposure opportunity evaporates.

Retail media is priced on attention. Advertisers pay for a person to see a sponsored listing on a results page, and that pricing assumes a human is doing the scanning. Wagner points out that an agent doesn't respond to placement that way. It reads the data it's given, compares options, and picks. If agents like Muse or ChatGPT do more of the browsing, the pool of human impressions those ads depend on shrinks. 

We forecast US retail media spend at roughly $69 billion to $71 billion in 2026, about 18% growth, with Amazon and Walmart capturing about 89% of the net-new dollars. Sponsored Products remained the largest contributor to Amazon's advertising growth in Q2, when ad revenue rose 26% YoY to $19.8 billion. The more the market concentrates in a few networks, the more exposed advertisers are if agents change how shoppers reach those networks.

Control vs. access

Amazon can afford to gate access while most retailers can't. Its shoppers arrive on their own, and its ad business, which brought in $19.8 billion in Q2 alone, runs on the traffic it already controls. Most retailers depend on being discovered elsewhere.

As AI assistants become a front door to shopping, a retailer that can't be read or purchased from inside them risks going unseen. The shift toward AI-led discovery will continue, whether or not Muse succeeds. 

What retail marketers should do

Data is where the experts converge. Wagner says to treat product data as infrastructure, as you once treated SEO: Price, availability, specs, and return policy must be structured and machine-readable, not buried in a PDF. Brands with sloppy product data can lose to a smaller competitor with clean data.

Three moves follow:

1. Audit your feeds and APIs for completeness, live pricing, and inventory accuracy.
2. Budget for eligibility, not just placement, and watch for sponsored inventory inside agent interfaces.
3. Choose your access stance deliberately. Weigh the reach of open agent integrations against the customer relationship you'd cede.

Keep funding retail media, but don't assume human impressions will remain the only currency.