The Gatekeeper Fight Begins: Amazon Blocks Meta's Muse, and the Agent Internet Splits in Two

The fastest-growing consumer AI app in America just hit its first wall — and the wall is Amazon.

On September 28, CNN reported that Amazon has blocked Meta’s Muse AI agent from accessing its store, making the e-commerce giant the highest-profile holdout yet in what is shaping up to be the defining fight of the agent era: who decides whether an AI agent may act on a website that never agreed to host it?

The stakes are not abstract. Muse hit the top of the U.S. App Store free chart ten days after its September 8 debut, and Sensor Tower’s most recent estimates put it past 3.4 million downloads — with rival trackers Apptopia and Appfigures pegging it between 2.3 and 4.3 million. Meta’s head of agents wants Muse to become the gateway to users’ digital lives. Amazon’s answer, effectively, is: not through our door.

What Amazon actually said

Amazon’s objection is notably specific. The company said it was given no notice and no choice about whether its store should be available through Muse, and it raised concerns about how Amazon account data and credentials would be handled by a third-party agent logging in on a customer’s behalf. It also pointed out that agents don’t offer Amazon’s personalized recommendations — the cross-sell engine that turns a single product search into a basket.

“We think it’s fairly straightforward that third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions about whether or not to participate,” Amazon said in a statement to CNN.

That framing matters. Amazon is not claiming agents are useless — it is asserting a property right: platforms get to opt in, and an agent that logs into a store without permission is operating outside the rules. Resy, the restaurant reservation platform, drew the same line, telling CNN it “does not currently permit unapproved third-party bots or agents.” Earlier this month Resy locked a user’s account after a different assistant, Instinct, tried to book hundreds of tables per hour — a preview of what abuse looks like when agents are given credentials and no governance.

Why Muse forced the question

Agents go beyond the chatbot pattern most consumers know. Instead of answering questions or drafting text, they take multi-step actions: comparing prices, filling forms, booking reservations, completing purchases — if the companies on the other end allow it. Muse launched with partnerships that did exactly that: Walmart, GameStop, Sephora, OpenTable, Expedia and Shopify are integrated into the app, and CNN’s own test succeeded in booking a restaurant reservation through Muse after providing login details like a phone number and confirmation code.

Meta has also weaponized distribution. According to TechCrunch’s reporting on Sensor Tower data, house ads for Muse appeared across Facebook, Instagram and WhatsApp the day after launch; within ten days Muse was receiving the majority of Meta’s house ad inventory, ahead of promotions for Facebook itself. Sensor Tower found Muse averaged 55% day-over-day download growth in its first two weeks — more than double ChatGPT’s 24% in its first ten days, while Claude and Grok download growth actually declined during their launches. The app held the No. 1 free spot on the U.S. App Store from September 18 and topped Google Play from September 19.

The market noticed. Shares of Airbnb, Expedia and Tripadvisor fell around 5% last week after Meta’s Connect announcements of new Muse capabilities — video chat with the Muse avatar, computer use on the Mac, a dedicated email address, more connectors, and smart glasses integration in the coming months. On September 25 Meta opened an early-access program for those features, with a twist: you request it by messaging Muse itself.

The economics of refusal

Why would a retailer turn away a channel with millions of engaged users? Because the agent interface strips away everything platforms monetize beyond the transaction itself.

When a human browses Amazon, they see recommendations, promotions and competing products; when an agent browses on their behalf, it sees only what the task requires. Bloomberg Intelligence senior analyst Mandeep Singh laid it out for CNN: you browse for one thing “but then you also see other things that you may end up buying.” An agent collapses that surface. Worse for platforms, Meta CEO Mark Zuckerberg said at Connect that Meta eventually plans to charge a commission on transactions made through Muse — meaning partners could lose both their upsell surface and a cut of the sale to the agent layer.

There is also a harder-edged concern CNN’s reporting only gestures at: agents are strong at comparison tasks like evaluating insurance policies or mobile carrier plans. A layer that automatically switches customers to the best deal converts loyalty programs and dark patterns from growth engines into dead weight.

The upside case is equally concrete. Expedia chose to participate because it wants “to show up wherever travelers are searching,” VP of enterprise alliances and AI Kathryn Heinz told CNN — noting that about two-thirds of Expedia traffic still comes through its own platforms, so the agent channel is incremental reach, not replacement. IDC analyst Francisco Jeronimo argues Meta has simply out-marketed rivals on agent shopping: “Others have not been so vocal about using the agents to be able to shop as Meta has been with Muse.”

Two internets, one question

Strip away the brand names and the conflict is structural. The agent internet is splitting into an opt-in tier — Walmart, Expedia, OpenTable, Shopify merchants, and by extension OpenAI’s and Google’s partner ecosystems — and a walled tier represented by Amazon and Resy, where automated access without agreement is treated as abuse. Every aggregator, booking platform and marketplace now faces the same binary CNN describes: partner with the agent layer and risk becoming a fulfillment backend, or refuse and risk invisibility in the interface where a growing share of demand originates.

Jeronimo’s prediction is the maximal version: “the AI platforms to become the layer and the main interface for most of us when we are shopping.” If he is even half right, Amazon’s block is not the end of the fight — it is the opening move in a negotiation about tolls, data and discovery that will define consumer AI economics for years. The early evidence says the negotiation has already started: block first, then see who blinks. The data point to watch is whether Amazon and Muse eventually sign a commercial agreement — because whichever side concedes first sets the template every other platform will demand.