Amazon Blocks Meta's Muse: What the Agentic Commerce Standoff Means for Your Business

When Meta released Muse on 8 September, the pitch was simple: a personal AI agent that shops, books and negotiates for you, connected to your email, your calendar and your payment card. Within a week it was the number one free app on Apple's US App Store, and the launch pushed agentic commerce from conference slides into the daily routine of millions of people.

It took Amazon less than two weeks to respond. On the night of Sunday 20 September the company quietly cut Muse off from Amazon.com. Anyone asking the agent to buy something from Amazon now gets a blunt popup instead of a parcel: continued access by an unauthorized AI agent violates Amazon's Conditions of Use.

This is not a small technical spat between two giants. It is the clearest signal yet of how the next few years of e-commerce will be fought, and it has practical consequences for any business that sells online or builds digital products. Let's sink in!

A Block That Landed Overnight

According to reporting from GeekWire and Axios, the block went live on Sunday night with no public warning. Muse works partly through APIs and partly through browser automation, so Amazon appears to be detecting the agent's sessions and refusing them at the door. Users see the warning message, the agent apologises, and the purchase simply does not happen.

The timing stings for Meta. Muse had just topped both Apple's and Google's app store charts, and Meta shares closed up 11 percent on the Monday after launch. A shopping assistant that cannot shop at the world's largest online store is missing a rather large piece of its promise, at least for American and European users who default to Amazon for everyday goods.

Meta had reportedly asked Amazon to simply exclude itself from the Muse shopping experience if it objected. Amazon went further and blocked the agent outright, which tells you how seriously it takes the question of who is allowed to stand between it and its customers.

Why Amazon Pulled the Plug

Amazon's public reasoning comes down to three complaints. First, Meta never told Amazon that Muse would visit the store, and never asked permission. Second, the agent does not identify itself while browsing, so Amazon cannot tell agent sessions from human ones. Third, and most seriously, Amazon alleges that Muse captures and retains customer credentials in ways customers may not fully understand.

An Amazon spokesperson put it this way: third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect the merchant's choice to participate or not. The comparison the company draws is with food delivery and travel apps, which work through formal partnerships rather than by walking in the front door pretending to be the customer.

There is also a quieter reason. Amazon made over 68 billion dollars in advertising revenue last year, nearly all of it from sponsored placements that only work when a human is actually looking at the page. An agent that goes straight to the best-value listing never sees an ad. Every purchase that moves from the Amazon app to a third-party agent chips away at one of the most profitable parts of the business.

Meta's Side of the Argument

Meta has stayed fairly quiet since the block, but its earlier statements set out the defence. The company says Muse has no visibility into people's passwords or payment methods, that credentials are stored securely and separately, and that the user approves sensitive actions before they happen.

The more interesting argument is the one about ownership. Meta's position, shared by most agent builders, is that a user asking software to shop on their behalf is still the user shopping. A federal appeals court leaned the same way in August, when the Ninth Circuit overturned an injunction Amazon had won against Perplexity and ruled that it is users, not AI companies, who access websites under anti-hacking law.

Caught in the middle are the people who actually bought into the product. Early Muse users connected their accounts, trusted the agent with real money, and now find that a chunk of their shopping simply fails. Whatever the legal outcome, that experience is a warning for anyone building products on top of platforms they do not control.

The Agentic Commerce Fight Is Bigger Than Muse

Amazon has spent the past year blocking shopping agents from Perplexity, Google and OpenAI, so the Muse ban is policy, not panic. The Perplexity case went to court, produced an early win for Amazon in March, and then flipped on appeal in August. The legal ground under agentic commerce is still moving, and both sides know it.

What makes this round different is scale. Perplexity's Comet was a niche product for early adopters. Muse arrived with Meta's distribution machine behind it and reached the top of the app charts in days. When an agent has tens of millions of users, blocking it is no longer a technical footnote; it is a strategic statement that retailers, regulators and investors all notice.

The relationship is complicated by the fact that Amazon and Meta remain deep partners elsewhere. Amazon listings have been purchasable inside Facebook and Instagram since 2023, and Meta signed a multibillion-dollar deal in April to run AI workloads on Amazon's custom chips. Both companies are effectively fighting over the customer relationship while doing business with each other in the background.

The Retailers Betting the Other Way

Not everyone is building walls. Walmart has partnered with both Google and OpenAI so their agents can shop its catalogue, while it develops an in-house assistant called Sparky. Target says traffic arriving from external AI platforms is up three and a half times year over year, and it treats those visits as a channel to optimise rather than a threat to block.

The numbers explain the split. Only about 16 percent of shoppers currently trust an AI assistant to make purchases on its own, so agent-driven sales are still small. But McKinsey projects that AI-mediated search and purchase automation could reach a trillion dollars by 2030. Retailers without a dominant marketplace of their own have little to lose and a lot of discovery traffic to gain by welcoming agents early.

Amazon can afford to say no because customers come to it directly. Almost everyone else is in the opposite position, and that is exactly why the agent question deserves a place in ordinary business planning, not just in big-tech news.

What This Means for Businesses Building Digital Products

Start by finding out whether agents already visit you. AI agent traffic shows up in server logs with distinctive user agents and behaviour patterns, and many analytics tools now break it out separately. If a growing share of your visitors are agents and your site quietly fails for them, you are losing sales you never knew you had.

Then make a deliberate choice about access. If you sell through your own site, structured data, clean product feeds, working checkout without dark patterns and clear pricing all make you easier for agents to buy from. If you decide you do not want agents, say so explicitly in your terms and your robots rules rather than leaving it ambiguous, because the Ninth Circuit ruling means quiet technical blocking may not settle the question by itself.

Product teams should also take the credentials lesson seriously. Muse got blocked partly over how it handles logins and payment details. If you are building an agent, an integration or any product that acts on a user's behalf, design for auditability from day one: scoped permissions, visible activity logs, and no silent storage of credentials. It is cheaper than retrofitting trust after a public fight.

Finally, do not bet your roadmap on one platform's goodwill. Meta built a flagship feature on access it did not control, and lost it in a weekend. The same risk applies to a small business whose entire revenue runs through a single marketplace, ad platform or app store. Spreading across channels is boring advice, but this week showed why it keeps being given.

Final Notes

The Amazon and Meta standoff will probably end the way these fights usually do, with a negotiated protocol, a revenue share, or a court ruling that forces one side's hand. Agentic commerce itself is not going back in the box. Too many users clearly want software that handles the boring parts of buying, and too much money is lining up behind it.

For business owners, the sensible response is neither panic nor indifference. Watch how the big retailers position themselves, measure what agents already do on your own site, and make sure the products you build treat user credentials and platform rules with more care than the headlines suggest is common. The companies that come out of this well will be the ones that decided their agent strategy on purpose, before someone else decided it for them.