What Amazon’s fight with Meta’s Muse tells investors about who will own retail’s checkout
On Tuesday, September 8th Meta (META) launched Muse, a personal AI agent that can browse, compare and buy products for its users, and by Sunday night of that week Amazon (AMZN) had started blocking it from its retail site after Meta declined a request to remove the bot. Shoppers who now ask Muse to buy something on Amazon get a pop-up, warning them that an unauthorized AI agent is violating the site’s conditions of use.
However on Monday, Shopify (SHOP) went the other way, announcing a partnership with Meta that lets the AI assistant check out on every Shopify store using Shop Pay. The market reacted quickly, with Meta shares closing up 11.43% on Monday, and Muse climbing on the top of the free app charts. Shopify’s day, meanwhile, finished up 7.3%.
Within just 48 hours, two of the most important companies in the U.S. e-commerce landscape made opposite bets on the same question. When an AI does the shopping, who owns the checkout?
For investors, the answer will shape where retail margins, advertising dollars and customer data end up over the next decade. Here are three things for investors to keep in mind.
Why Amazon is drawing a line
Amazon's stated objections are about security and consent, saying Meta never told it Muse would be shopping on its store and that the agent doesn’t identify itself while browsing, appearing to capture and store customer credentials.
Meta disputes that last point, however, saying Muse has no visibility into passwords or payment methods because they sit in secure storage. An Amazon spokesperson told Forbes that third-party shopping agents "should operate openly and respect service provider decisions."
There is a business case underneath the security case, too. Amazon generated more than $68 billion in advertising revenue last year via a business depending on, put simply, people scrolling through pages and seeing sponsored products. An agent that goes straight to the item, however, skips the sponsored listings too.
Amazon is far from being the only retailer in this crux. After years of thin margins, many chains have turned their websites and apps into ad businesses. Leading Salesforce-native advertising management firm ADvendio, whose ad management platform is used by Albertsons, Ulta and Ahold Delhaize, launched a suite of AI agents in March this year to take some of the manual work out of running those networks. “Retailers need profitable growth to stay competitive,” said Julian Ahrends, the company's CTO.
So, retailers are already using AI agents to sell more ads, while shoppers are only beginning to use AI agents that never see them.
Muse is also not the first agent Amazon has shut out, with the company allegedly moving against shopping agents from Google and OpenAI. In the case of Perplexity, Amazon went as far as suing the firm over its Comet browser.
Yet, the stance has an awkward wrinkle. Amazon’s own Buy for Me feature completes purchases on outside retailers’ websites that never invited it in; retailers could only opt out after the fact, which makes it hard to argue that agents must ask permission when your own agent didn’t.
Courts have moved the goalposts
Amazon has tried to settle this in court before, and so far it has lost ground. When it sued Perplexity in November 2025 under the federal Computer Fraud and Abuse Act, it won a preliminary injunction on March 9, but then, on August 4, the Ninth Circuit vacated that injunction. Judge Milan D. Smith Jr. wrote that when a user directs an agent to act on Amazon.com, it is the user, not the AI company, who accesses Amazon's computers.
The court described Perplexity's assistant as "a tool, not a person,” though law firms have been quick to note that this was a preliminary ruling, not a final decision on the merits.
Still, it was the first real federal appellate guidance on AI agents and anti-hacking law; in the end, it leaned toward the agents.
That’s why the pop-up points to Amazon’s Conditions of Use and not to any federal statute: without the anti-hacking law behind it, Amazon can't win this once in court and move on. Rather, it has to keep spotting and shutting out agents, while explaining the error messages to its own customers.
Amazon has the money and engineers for that. Most retailers clearly don’t.
The ruling also leaves a harder question open. Who answers for a purchase when an agent gets it wrong? Traditional courts are poorly suited to the problem; civil disputes in the U.S. take an average of 344 days to resolve, far longer than the seconds an agent needs to buy something. And the gap is drawing new infrastructure.
In July, the GenLayer Foundation and 26 other companies, including OKX and MetaMask, launched Internet Court, an open standard for settling disputes between AI agents. GenLayer CEO David Riudor warned that the industry is “not prepared for the potential fallout.”
Three business models, three bets. Or is it?
Amazon isn't the only major retailer deciding what to do about AI agents, and its peers even beyond Shopify have reached very different conclusions. Those choices make more sense when looking at how each company makes its money.
Amazon's walled garden protects its ad business and leaves room for its own agents, including Alexa for Shopping, which launched in May; Shopify earns much of its revenue from payments, so it cares less about whether shoppers ever land on a merchant's website, as long as the transaction runs through its rails.
Shopify president Harley Finkelstein even told investors earlier this year that AI models do not bypass Shopify's checkout. Shop Pay has already processed more than $400 billion in lifetime accelerated gross merchandise volume, and agents are one more way to add to that total.
According to data from the company, Shopify paid out more than $1.3 billion to the developers who build across its ecosystem. For example, StarApps is the company behind some of the most widely installed product merchandizing apps on the Shopify App Store.
The banks behind those transactions are paying attention, too. Financial institutions including Bank of America, Capital One and NatWest have already been looking into agentic AI.
In fact, many are indeed already putting the technology to work. GFT Technologies Canada, led by CEO André Gagné, is one company playing an important role here.
Payment processors that serve smaller merchants are bracing for the same shift. Dimitri Akhrin, president of BAMS, a merchant services provider for small and mid-sized retailers, said last week that agents completing purchases on shoppers' behalf will force payments companies to rethink fraud prevention, consent and authorization from the ground up.
And then, there’s Walmart sitting somewhere in between. Sure, it is partnering with Google (GOOG) and OpenAI to make its products discoverable inside their agents, all the while it builds its own, Sparky. The retail giant also backs the Universal Commerce Protocol, an open standard for agent checkout, alongside Target (TGT).
Even open doors come with conditions, however. When CNN tested Muse this week, the agent could browse Target's site, but the reporter still had to enter personal details manually because Target blocked autonomous clicks at checkout, for instance.
Those the dispute leaves out
But let’s be clear: this three-pronged framing leaves out most of the market. Mid-sized retailers have neither Amazon's destination traffic to justify a wall nor Shopify's payments network to monetize an open door. For them, the urgent question is whether their systems can handle agents at all.
The pressure is already showing up at industry events. Nisum, a global technology consulting firm whose retail practice includes building agent-ready commerce, is presenting at RetailClub AI Festival in California this week, with a session by its Board Member Sajid Mohamedy on why some retail AI pilots stall before they ever reach P&L.
Martin Lewit, the firm’s SVP of Corporate Development, was also earlier interviewed on AI in Forbes.
The traffic is already arriving. Adobe found that visits from AI sources to U.S. retail sites grew 393% year-over-year in the first quarter and, by March, those visitors converted 42% better than non-AI traffic, compared with 38% worse a year earlier.
Yet individual product pages scored just 66% on Adobe's machine-readability benchmark, which means about a third of the content agents rely on is invisible to them. McKinsey estimates agentic commerce could orchestrate $900 billion to $1 trillion in U.S. retail revenue by 2030. We're also increasing seeing innovation in the retail space, with companies such as Effie AI introducing the first agentic retail execution platform.
Further, in the financial sector Y Combinator-backed company Ontop, led by CEO Julian Torres Gomez, is building the financial infrastructure for the global workforce, and is certainly one company to watch.
Time to decide, but not much of it
Consumers are not fully sold yet, which gives retailers some breathing room. Only 16% of shoppers say they are comfortable letting an AI assistant both find and buy products for them, according to Coveo's 2026 Commerce Relevance Report.
In sum, most agent activity today is still research, not true purchasing.
The respite is shrinking, though. Muse went from launch to the top of the app charts in less than two weeks, and conversion from AI referrals has flipped from a liability to an advantage in a single year.
The test investors should watch over the coming months is simple: if Muse users abandon the agent to shop Amazon directly, the wall works. If they keep using Muse and buy elsewhere, Amazon will have handed traffic to its competitors.
For the thousands of retailers without Amazon’s scale, walling off the store isn’t a realistic option. They can partner, they can get their data and systems ready for agents, or they can get skipped.
The firms that make that call deliberately, and not by default, are those most likely to hold onto the customer relationship when a machine starts doing the buying.