Amazon threw Meta's shopping agent out

Wednesday. Halfway.

Amazon has cut Meta's Muse agent off from shopping on Amazon.com. Not throttled, not rate-limited. Muse users hit a popup: "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use, to which our customers have agreed."

Amazon asked Meta to exclude the site voluntarily first, and gave three reasons when that did not happen, to GeekWire. Meta never told Amazon the agent would shop there. The agent does not identify itself when it browses. And it appears to capture and store customer credentials.

Read those three again, because they are not really complaints. Simon Taylor, who writes Fintech Brainfood, spotted it first: "Amazon's three complaints read like the terms of that deal: announce the agent, identify it, keep logins out of it." Those are opening terms, written as grievances.

The money underneath is not subtle. Amazon made over $68bn from advertising last year, and an agent that shops for you walks past every sponsored placement on the way to the button.

Muse launched on 8 September and was the number one free app in the US App Store within a week. Roughly the speed at which a theoretical problem becomes a quarterly one.

Amazon has been here before. It sued Perplexity over Comet, won an injunction in March, then lost it on 4 August when the Ninth Circuit decided the user rather than the AI company was doing the accessing. Blocking is what is left when the courts will not help.

Taylor's closing point is the one I would put on the whiteboard. "The sad thing is that Amazon can afford to turn away Muse's orders. Most merchants can't. The power dynamic in commerce is kinda messed up."

That is your position here. You are not Amazon. When an agent arrives logged in as a customer, you are not negotiating terms, you are deciding whether to take the order without knowing it is an agent.

So work backwards from Amazon's list. Announce, identify, keep logins out. Which of the three could you enforce today?

Today’s sponsor:

Ben Thompson of Stratechery, quoted by TBPN, on X:

He goes on to the thing the agentic decks skip, which is that people enjoy shopping. There is a mountain of detail on a product page because customers asked for it, down to what size the model is wearing. Taking the browsing away is not always the gift it looks like.

15%. The share of their own purchases consumers now expect AI agents to be making within five years, up from 9% a year ago, per Global Payments' agentic commerce research. Forty-five percent have used an AI shopping agent or would consider it.

It's a payments company measuring appetite for payments, and the fieldwork ran in May, so treat it as spring sentiment. The brake stands up anyway: 50% worry about payment security, and a third want to approve every transaction before it goes.

Methodology holds up better than most: 8,027 consumers across seven markets, run by The Lantern, with a matched wave a year earlier, which is what makes the 9% → 15% move real rather than a one-off reading. The best line I left out, if you want a swap: comfort with letting an agent spend up to $50 on cinema tickets went from 32% to 82% in a year.

Our Competitive Pricing Research Framework, run the way a shopping agent would run it:

Agents compare on the fields they can read, so your price, tiers and packaging get flattened into a table whether or not you would have picked that fight. Build the matrix from public pages only, the way a bot sees you, then look at what is missing from your row.

Agent panic is running hot, and most of the heat comes from people selling something. Amazon is not panicking. Amazon is writing terms.

The useful response is rarely an emergency. It is one page saying what you allow on your own property, written before somebody needs it on a Friday.

We provide consulting services for CMOs on AI strategy for marketing teams through our Teamless offering.

Three ways to measure the thing you are now being judged on:

If the marketplaces are going to fight over the middle, spend the week on the parts nobody can block:

Kit — best for teams who want the list itself to be the asset, free to about $50 a month. Try Kit here →

Tailwind — best for ecommerce brands on Pinterest and Instagram, and useless outside those two. Try Tailwind here →

Hover — best for putting the domain and the email on something you own, for the price of a sandwich a year. Try Hover here →

→ We have 300+ curated, filtered, and categorized AI tools in our directory. Browse to find the best ones that help your marketing team.

The practical version of owning your own ground: The AI Content Library. The Lego Brick Method builds content as reusable pieces rather than campaigns, so the same material reassembles for a channel you had not heard of when you made it.

Some of the channels reading you next year will not be people.

→ We publish workshops, courses, guides, manuals, and more on AI marketing for free. Browse our learning resources for free.

Stanford altered a promotional photo of three of its own students with AI, replacing a male student with a Black woman and making another appear thinner, as Straight Arrow News reported.

The university confirmed it broke a rule Stanford itself wrote. "Both the alteration and lack of disclosure in this case violate that policy," said Charlene Gage, its director of university public relations.

Billy Ramirez, class of 2027, remembered posing for the original. He thought it was "hilarious" at first, then said he felt "silenced and erased."

Nobody sat down to do something cynical. Somebody needed a more representative photo by Thursday and the tool was right there. That sentence explains your version too.