Michael JohnsonThe blocking of Meta's Muse agent showed that an algorithm can search for products and act on a person's behalf, but the platform itself still controls access to its store. There is no evidence yet that this has made brands or advertising stop working.
September 29, 2026
On September 20, users of Meta Muse who tried to shop on Amazon saw a message saying that continued access by an unauthorized AI agent violated the platform's terms. GeekWire and TechCrunch reported the block.
This episode does not prove that Amazon is losing control of shoppers. It shows something else: a person can ask an agent to search for and compare products, but that request does not obligate a store to let the agent in. The dispute is about who will mediate between the shopper and the storefront.
An agent instead of a storefront
On September 8, Meta introduced Muse as a personal agent for multistep tasks. According to the company, it operates in a virtual environment with its own browser, can fill out forms, and can request approval before sensitive actions, including purchases. These are Muse's stated capabilities, not an independent assessment of the agent's reliability.
The shopper does not disappear in this arrangement: they set the goal and criteria, and sometimes confirm the order. But part of the journey from request to selection may move from the store's website to the agent's interface. For a platform, this could mean less direct interaction with the customer and less control over which products they see. For now, that is a possible consequence, not a measured effect of Muse.
Amazon blocked access—and explained why
Amazon said that Muse had not told the platform it was acting as an agent, and that third-party apps buying on a customer's behalf should operate transparently and respect a service's decision about whether to participate. The company also cited risks to data and security. Meta, for its part, said Muse could not see passwords or payment details. These explanations come from the parties to the dispute; the available material contains no independent technical audit confirming the specific blocking mechanism or either account.
This is a clash between user authority and platform rules. An agent may act on a shopper's instructions, but that alone does not give it the right to use a store in any way it chooses. Amazon retains the ability to restrict access to its storefront—and thereby influence where an agent can search and place orders.
The block happened on September 20, so this is not news from the past 24 hours. However, on September 29 Meta announced an expansion of Muse for small businesses, with stated capabilities including connections to business services and company accounts. The announcement shows that Meta is developing the agent as a business intermediary too, but it does not indicate widespread use of Muse for shopping or a new turn in the conflict with Amazon. Meta's announcement describes product capabilities, not confirmed commercial results.
Advertising hasn't disappeared—the question of visibility has changed
If an agent compares products on its own, people may browse familiar search results, sponsored listings, and brand storefronts less often. This could complicate the use of advertising formats built into a store's interface. But the available material does not show how Muse affected ad impressions, conversions, sales, or real shoppers' choices. There is no basis for claiming that branding or traditional advertising has already stopped working.
A limited experimental signal comes from a preprint on arXiv, published on September 23. The authors studied the behavior of eight language models in product-selection tasks where obtaining information required additional actions. Under some conditions—with a vague goal and limited access to information—agents overlooked details needed to compare prices. A more specific goal helped them find relevant information. The study does not examine Amazon advertising, brand strength, or real-world shopping at scale; it is a preprint, not proof of how the market will change.
For brands, then, the question is not whether “the algorithm will replace advertising,” but which signals an agent can find and take into account. Price, features, reviews, and warranties may influence a recommendation if the agent obtains that information. But there is no data yet on how much weight these factors carry in real purchases through Muse.
Who will become the shopping interface?
Amazon's block shows not a transfer of power to algorithms, but a fight over how that power is distributed. Users can delegate part of the decision; Meta aims to make Muse an intermediary; Amazon retains the ability to set the rules for access to its platform.
The scale of agent-led shopping, its effect on advertising, and the specific technical method used to detect Muse have not been established. For now, it is more accurate to talk not about the end of branding, but about competition for the shopping interface—and who decides which intermediaries are allowed into the store.