Crossed Wires: Meta vs Amazon — the battle for the future of online commerce

On 20 September, Amazon slammed the door in Meta’s face. This was not one of those predictable internecine tech giant spats. It signalled the beginning of a volcanic disruption in the business model of online consumer commerce.

Here is what happened. Meta launched Muse on 8 September as the world’s first personal AI agent built for the rest of us – no technical experience required. You tell it what you need done, and it does it – books travel, organises your calendar, negotiates on your behalf, and goes shopping. It can open a browser, fill out forms, and complete purchases with your blessing. It runs on a dedicated secure virtual machine, keeps your passwords invisible even to itself, and checks with you before sensitive actions like sending an email or making a purchase. There have been personal agents before Muse, but its major feature is simplicity of use.

Within days of Muse’s release, users trying to shop on Amazon through the agent encountered a pop-up – “Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use, to which our customers have agreed.” The reasons given were the familiar ones – security, privacy, customer credentials, terms of service and the like. All perfectly reasonable, but ultimately deflections. (Amazon’s own “Buy for Me” agent, meanwhile, strolls onto other retailers’ websites to buy things on your behalf, those retailers having been enrolled on an opt-out basis – which is to say, not asked.)

There is a much larger reason why Amazon might not want somebody else’s AI agent wandering freely through its enormous digital shopping mall – agents don’t look at advertisements, sponsored listings, colourful brand logos and “customers also bought” carousels that fill every pixel of the screen. Agents look for price, specs, delivery times and reviews; the rest is largely dross to them.

How much is that dross worth? Amazon’s advertising business booked $68-billion last year, and all of it depends on a human standing in the virtual aisle looking at stuff. An agent that compares 40 options in a second and clicks “buy” strips out every surface that Amazon tries to monetise. Agents threaten a huge whack of the advertiser-supported online economy – in the US, online ad spend hovers around $300-billion to $360-billion a year, while global digital ad spend has climbed toward $970-billion.

It is the business model of the consumer internet, and it is now going to war to protect itself.

The subplots

There are several subplots under the main story. The first is that Amazon does not object to agents at all. It objects to agents it does not own. Amazon wants you to use its own agent, Alexa for Shopping, and only Alexa for Shopping – offered as your exclusive tour guide and shopping assistant, which is the best way for Amazon to maximise its profit. Alexa for Shopping, of course, has Amazon’s interests at heart, not yours. 

But can Amazon refuse entry to other agents?

Amazon’s legal campaign against shopping agents began in November 2025, when it sent AI company Perplexity a cease-and-desist letter over Comet, an AI browser whose built-in assistant shops on users’ behalf. Amazon sued under the Computer Fraud and Abuse Act, the US’s main anti-hacking statute, and in March 2026 won a preliminary injunction barring Comet from the password-protected parts of its site.

In August, the Ninth Circuit threw out that injunction. Comet runs locally and acts only on a user’s instruction, so the court reasoned that it is the user, not Perplexity, who “accesses” Amazon’s servers. Perplexity argued that blaming it was like blaming Apple whenever someone shops in Safari, and the judges agreed that holding the toolmaker liable would stretch the law well beyond its purpose.

That leaves Amazon exposed. Its strongest weapon, the anti-hacking statute, now struggles against any agent a customer chooses to direct. What remains are terms-of-service claims, which are weaker, slower and harder to enforce. Amazon’s best hope lies in a technical detail. 

The Ninth Circuit’s reasoning turned on where the software actually runs – Comet sits on the customer’s own machine, so the instructions reaching Amazon’s servers come from the customer, which makes the customer the one doing the accessing. 

Muse, for now, runs from Meta’s servers, which is the one distinction Amazon has left to argue. But that is an engineering decision, not a legal principle, and engineering decisions can be reversed in a product cycle. Muse, or any other agent, could just as easily run from your laptop, at which point it becomes legally indistinguishable from Comet and very nearly invisible to Amazon. Which leaves the company’s remaining protection resting on its opponents choosing to keep an architecture they have every reason to abandon.

A further subplot – even if Amazon and others welcomed the agent (Meta has deals with other big marketplaces), how would Meta get paid? No one seems to know, and Meta CEO Mark Zuckerberg has hand-waved about a “commission” sometime in the future, but did not specify from whom. The merchant? The shopper? The payment processor? All of them are a challenge, and I suspect that agents will be so ubiquitous within the year that no one will be able to charge anything for them – they will simply be part of your operating system or chatbot or browser, or whatever runs your computer in the future.

Finally, there is this. Every single AI company and operating system developer knows that agents are going to be at the vanguard of user interaction – the old behaviour of “search internet, visit some websites, then browse and buy” will very soon become a quaint memory of a bygone age, rather like the mail-order catalogues of the early 1900s. All the big players are now rushing to own the agent space, pouring great globs of capital into the race. There is unlikely to be only one winner; agents will quickly commoditise, just like other internet tools.

Brand giants panic

And advertisers and brand owners? I suspect that in the strategy sessions of brand giants there is no small amount of panic. The advertising industry can see the agent storm approaching, but so far its answers look distinctly tepid. The creative core of advertising has always appealed to deep human emotion, not much of an advantage when you are trying to convince a piece of cold software to buy your product.

Brands are being told to restructure product data so agents can understand them, while Google and Amazon are experimenting with sponsored recommendations inside AI conversations. A new vocabulary of agent optimisation is emerging, essentially search engine optimisation (SEO) for machines. But none of this quite addresses the underlying problem. For 25 years digital advertising has been built around capturing human attention – the sponsored Google result, the promoted Amazon listing, the Instagram advertisement slipped between photographs of someone’s holiday.

An agent may bypass all of it. Ask it for a washing machine, and it could examine 200 models, compare price, reliability, warranties and reviews, and simply say: “Buy this Bosch.” The other 199 products have disappeared before the human even enters the conversation. Advertisers can try to influence that shortlist by paying agents for attention, but doing so creates an uncomfortable contradiction. If Bosch can pay the agent for preferential treatment, then the agent is no longer unambiguously working for you. If Bosch cannot, much of the machinery of digital advertising becomes irrelevant. And so the industry’s response amounts largely to making products easier for agents to find and hoping there will still be somewhere to put the ads. That is not an obvious answer to the problem.

Amazon’s rude blocking of Muse is a shot over the bows. It signals that the incumbents see the threat and are willing to fight. But it also reveals the limits of their arsenal.

The platforms that survive the agentic transition will not be the ones that ban agents. They will be the ones that build for them – that treat agents as first-class customers with their own needs for data, pricing signals and transactional APIs. Amazon could have opened its systems to Muse and captured a share of agent-mediated commerce. Instead, it shut the door. That choice will look increasingly anachronistic as agents become the primary interface between consumers and the digital economy.

The advertising model built for human eyes has no answer for software that doesn’t look. The platforms that understand this – and build accordingly – will inherit the next era of internet commerce. The ones that don’t will find themselves shouting into a void, selling impressions to no one. DM

Steven Boykey Sidley is a professor of practice (ex-JBS, University of Johannesburg), a partner at Bridge Capital and a columnist-at-large at Daily Maverick, where he writes the weekly Crossed Wires column. His new book, It’s Mine: How the Crypto Industry is Redefining Ownership, is published by Maverick 451 in South Africa and the Legend Times Group in the UK/EU, available now.