Amazon blocked Muse for violating its terms of service. Which is technically true and strategically hilarious, given what Amazon's own agents are doing on everyone else's websites.

When Amazon blocked Meta's new Muse assistant from its marketplace this month, the official language was about as exciting as you'd expect from a terms-of-service notice. Continued access by an unauthorised AI agent, it read, violates the conditions of use. Very legalistic. Very polite. Very "we're not angry, we're disappointed."

But strip away the corporate prose and what you're looking at is Amazon essentially padlocking its front door because Meta sent a very well-dressed robot inside to do all the shopping.

Here's why this matters more than the usual platform slap-fight: Muse isn't some niche tool for people who enjoy configuring browser extensions on a Sunday afternoon. Within a week of its US launch, it reportedly sat at the top of Apple's free App Store charts. Ahead of ChatGPT. It doesn't just answer questions about products — it books, buys, and fills in forms on a user's behalf. That combination of mass distribution and transactional capability is what separates this from every previous agent skirmish, and it's why the block should be read as a strategic declaration rather than a customer service memo.

Why Muse Specifically Made Amazon Nervous (And Perplexity Didn't, Really)

Amazon has been fending off rival agents for over a year now. The earlier confrontation was with Perplexity's Comet browser, which was concerning but manageable — a product that a relatively small, self-selecting group of technically curious people chose to install. An agent is only as disruptive as the number of people who actually let it shop for them, and a browser aimed at early adopters has a natural ceiling on that number. (Most people are still figuring out how to clear their cache. We're fine.)

Muse doesn't have the same ceiling. It arrives with the distribution power of one of the largest consumer platforms on Earth. But here's the part that should genuinely worry anyone selling on Amazon: behind Muse sits Meta's behavioural data. What people look at, what they react to, what they linger on, what they want before they know they want it.

Amazon's most durable competitive advantage has never really been price or logistics alone. It's the recommendation engine — the system that decides which products a shopper sees first and in what order. That system is the discovery layer, the top of the funnel where demand is shaped before a purchase is ever made. Whoever controls discovery controls the most valuable real estate in commerce.

An agent built on Meta's data could, in principle, reconstruct that entire discovery layer outside Amazon's walls. If a shopper starts inside a Meta assistant that already knows their preferences, the decision about what to buy is effectively made before Amazon is even involved. Amazon still ships the box, sure — but it gets pushed down the stack toward logistics while someone else owns demand.

Which is, to put it mildly, not where Amazon wants to be.

Follow the $68 Billion (It's Always About the Ads)

The part of this story that matters most to anyone selling products online isn't discovery in the abstract. It's money. Amazon's advertising business generated more than sixty-eight billion dollars last year. That business rests on a simple physical fact: people browse Amazon's pages, and while they browse, they see sponsored placements, recommendations, and all the other real estate that brands pay handsomely to occupy.

Remove the browsing and you remove the surface the advertising is sold against.

An agent that moves straight from a spoken request to a completed checkout collapses that surface entirely. There are no search results to sponsor, no product grid to rank within, no page for an ad to sit on. This is the quiet threat inside agentic shopping, and it's the reason the Muse block is about considerably more than security or digital etiquette.

Meta makes this threat particularly acute because — of all the companies currently building shopping agents — it is the one that also runs a massive performance-advertising business of its own. If product discovery migrates into a Meta assistant, the advertising budgets currently chasing attention on Amazon don't just evaporate. They have a very obvious new home. For brands, the practical consequence is concrete: a channel strategy built on the assumption that a human will see the listing, read the copy, and notice the sponsored placement is now exposed both to a shift in where that attention goes and to who collects the fee for it.

The "Do as I Say, Not as I Agent" Problem

Here's where things get genuinely entertaining. Amazon is not, let's be very clear, opposed to agentic shopping. Amazon is building agentic shopping with visible enthusiasm. It just wants to be the one doing it.

Earlier this year, Amazon launched its own shopping agent — Alexa for Shopping — combining its Rufus recommendation system with Alexa+, a generative upgrade of its voice assistant. That agent can compare products, show price histories, find deals, and add items to carts. A companion feature called Buy for Me shops other retailers' websites on a customer's behalf, and Amazon is careful to note that it identifies itself to those retailers and lets them opt out. (Very considerate.) They've since added Update Me When, which watches not just Amazon's own catalogue but the wider web, alerting shoppers about availability. On its own promotional calendar this summer, agent-led shopping reportedly outpaced conventional search. And this month, Amazon agreed to let its advertisers appear inside ChatGPT.

So to summarise: on the same day Amazon told Meta that a third party sending an agent onto someone else's store should operate openly and respect the host's decision about whether to participate, Amazon's own agents were cheerfully ranging across everyone else's turf. The company's defence — that its outbound agents announce themselves and honour opt-outs — is a real distinction. It's also a suspiciously convenient one.

The newest feature is even more revealing. Amazon now offers fraud verification: its assistant checks a suspicious email or text against Amazon's own records and tells the shopper whether it's genuine. Useful? Absolutely. But consider what it does to Amazon's position in this fight. The company is making its own agent the authority that certifies what can be trusted, at the exact moment it argues that nobody else's agent can be trusted. Trust is being claimed as territory.

How Long Can Contract Law Hold the Line?

If blocking is a delaying action — and it is — the interesting question is how long the delay lasts. The legal manoeuvring is more instructive than the press releases.

Amazon won an early court order against Perplexity in the spring. Then lost it on appeal in the summer, when the court held that under anti-hacking law, it's the user — not the AI company — who accesses Amazon's systems when an agent acts on the user's instruction. A request to rehear was refused. Having lost the argument that agent access equals unauthorised intrusion, Amazon has fallen back on contract. The Muse notice cites the conditions of use. The amended case against Perplexity leans on a clause Amazon added requiring agents to identify themselves in the short technical string every browser sends.

The problem with a contract-based strategy is visible in the same filings. Amazon alleges that Perplexity has changed its identifying string each time Amazon learned to detect it. Terms of service bind the parties who agree to them, but an agent maker that instructs its software to present itself as an ordinary human browser is, by definition, a moving target. A barrier that has to be rebuilt every time the other side changes its disguise is a barrier with a significant maintenance problem.

You Can Block a Company. You Can't Block a Category.

The deeper reason the gate may not hold is that Amazon isn't facing one challenger. It's facing a direction of travel.

Meta is building toward the discovery surface. So is Perplexity. So is Google, which has been folding shopping into its own AI products. Each can be blocked, sued, or negotiated with individually, and Amazon has shown it's willing to do all three. But a single company can be walled out in a way that a category cannot, and the pressure on Amazon's discovery-and-advertising position is arriving from more than one direction at once. TikTok Shop, for what it's worth, has reportedly been compounding at somewhere near eighty-five per cent year on year — a pace incumbents almost never sustain at that scale. I wonder when they will roll out their AI shopping agent?

Whatever one makes of any single projection, the point holds: the place where products are discovered is being contested faster than any one lawsuit can settle.

The Bottom Line

Amazon blocked Muse not because it has a principled objection to shopping agents — its own agents make that argument rather difficult to sustain. It blocked Muse because Meta, specifically, has the distribution, the data, and the advertising business to genuinely threaten Amazon's control of the discovery layer. And when discovery moves, advertising revenue follows.

For sellers, the practical takeaway is uncomfortable but clear: the assumption that shoppers will continue browsing Amazon's pages — seeing your listings, your A+ content, your sponsored placements — is the assumption that's now under pressure. Not from one direction, but from several simultaneously.

Whether Amazon's contract-law strategy buys it enough time to make its own agents the trusted default is an open question. What's not open is that the company regards agentic shopping as the entire game, and intends to be the one holding the cards when it arrives. Every move to keep other agents out is also a move to buy time for its own. The question for the rest of us is what we're doing with that same window.

🔎 Camp - An app to extract information, people, location, and music from screenshots.

🗣️ ChatGPT Sugar - Enrich your daily interactions with ChatGPT

🤖 Top GPTs - Access the best OpenAI GPTs without a Plus subscription

📊 Syncly - Surface real customer pains with AI feedback analysis

😍It was great, all I needed

🙂Good, not great

🙄It sucked