Whose Agent Is It? The Race to Own Your Customer

Part 3 of 3: Disintermediation and the fight for the front door

So far we have seen the middleman survive banks, catalogues and the web, and we have seen Amazon become the biggest one of all. Now look at the last row of that table, because something new is trying to stand between Amazon and its own customer.

What an Agent Changes

Every previous interface was oversold, but an agent really is different in kind.

A search engine answers a question. A marketplace organises a catalogue. A recommendation feed proposes things. All three wait for you to show up. A personal agent holds context over time and acts toward a goal you gave it. It can remember your preferences, reach across services, take a plain-English goal instead of a keyword, plan several steps, compare across merchants and marketplaces, keep working while you sleep, and, within limits you set, spend your money.

Three products show how fast the contest has started. Meta launched Muse in the US on 8 September 2026 as a free personal agent for multi-step tasks across email, calendars, payments, dining and shopping. Instinct, from the San Francisco startup Spear Street Technology, is an invite-only agent you reach by text message or phone call, with no app at all, which runs errands such as bookings, rides and subscription cancellations. And xAI's Grok Bot (launched in August) gives you a team of always-on bots, each with its own cloud computer, which hand work to each other. Whether any of them turns out to be the real thing is not the point. They are early moves in a contest over who gets to hold the buyer's delegated intent, and that is the contest that matters.

And the contest is already physical. On 20 September, twelve days after Muse launched, Amazon blocked it from shopping on Amazon.com. A company does not bother to lock the door against something it thinks is harmless.

Who Wins the Race to the Front Door

If you made me bet, I'd back them in the order Muse, then Grok Bot, then Instinct, though for different reasons and not really for the same customers. Muse has the oldest advantage in the book, which is distribution. It lands free in front of a user base that already lives inside Facebook, Instagram and WhatsApp, and it was reportedly sitting at the top of Apple's US free-app chart within a fortnight. Nothing else on this list starts with billions of existing relationships, and the history above says the winner of an interface war is usually whoever is already in the customer's hand. Its weakness is trust and incentive. Meta earns its living from advertising, so the question of whose agent this is hangs over it from the first day, and Amazon's block shows how quickly a platform can shut a door in its face. Instinct has the best answer to friction. There is no app, no dashboard and nothing to learn; you text it or call it like a person, and for the ordinary household errand that is very hard to beat (I believe friction is the single most underrated variable in consumer technology). But it is invite-only, it asks you to hand a startup your email, messages, location and payment details, and it is up against Apple and Google, who can turn the same behaviour into a built-in feature of the phone. A text-message interface is easy to copy. An operating system is not. Grok Bot is playing a different game. Each bot gets its own cloud computer, you build a team of narrow specialists that pass work to each other, and you teach it your routines, which means more setup and a steeper climb for the casual buyer, and a much better fit for a small business owner with an inbox, a CRM and a pile of invoices. It is unlikely to win the kettle. It could win the back office, and in the long run that may be the more defensible position, because a business that has trained a team of bots on its own processes has switching costs that a consumer texting for a taxi will never have.

Why an Agent Could Dethrone the Storefront

Amazon holds four valuable positions in the buying process, and an agent can in principle take every one of them.

Discovery. "What should I buy?" used to start with Amazon's search bar. It could start with the agent.

Comparison. "Which option actually fits my constraints?" Amazon's ranking is built to maximise Amazon's revenue. An agent that works for the buyer can compare on the buyer's terms.

Loyalty. "Where do I habitually start?" Habit is the real moat. The agent breaks the habit by becoming the habit.

Interface. "Which service represents me in this transaction?" If the answer is the agent, then Amazon is a back-end.

The agent's advantage is not that it searches better. It is that it knows things. It knows what you bought and regretted, what you can afford this month, what you'd pay extra to get by Friday, whose birthday is coming, and which brands you refuse to touch on principle. No storefront will ever know you like that, because the storefront is built to serve every customer at once, and the agent is built to serve one.

So Amazon could remain a supplier, a fulfilment network, a marketplace and a merchant of record, and still lose the thing that made it Amazon, which was the place where the customer decided.

The Case Against Me

This is a pattern I see, but I might not be right. I have been wrong before - I think back in ’87? Anyway, I digress...

First, agents need clean data. Pricing, stock, delivery windows and return terms have to be reliable and machine-readable, and a good deal of retail data is neither. An agent that buys the wrong kettle once will not be forgiven.

Second, the incumbents can simply shut the door, and Amazon already has. The pop-up Muse users now see says that continued access by an unauthorised AI agent violates Amazon's Conditions of Use. Amazon says Meta never told it Muse was coming, that the agent doesn't identify itself when it browses, and that it appears to capture customer credentials. Meta's launch post says Muse has no visibility into passwords or payment methods. Amazon has also gone after agents from Perplexity, Google and OpenAI, while running its own shopping agent, Alexa for Shopping, behind the wall. The legal ground is less comfortable than it sounds, though. In August the Ninth Circuit, in Amazon's case against Perplexity, ruled that the user and not the AI company is the party accessing the computer under federal anti-hacking law. That does not settle whether Amazon can enforce its terms against agents, which is exactly the fight to watch.

Third, payments, fraud, permissions and liability are unresolved. When an agent buys the wrong thing on your behalf, who pays? Nobody has settled that, and the law has barely started.

Fourth, for the most part consumer buying is not logical. People choose on taste, mood, status and ambiguity. Half of what we buy is a statement about who we are, and an agent optimising for "best kettle" misses that.

Fifth, and the one I take most seriously: the agent provider may simply become the next advertising-funded gatekeeper. If the agent is free, someone else is paying for it, and that someone will want to be recommended.

Sixth, Amazon can build its own agent, buy one, or offer preferential access to partners. It owns the data, the logistics and the checkout. A company with that many cards does not need to win every hand.

All of this is true, and I can't dismiss any of it. The most it does is delay the outcome and change who profits. What it cannot do is cancel the underlying mechanism, which is that whoever stands closest to the buyer's intent collects the margin. Amazon built an empire on exactly that insight, and it would be strange if the lesson only ran in one direction.

Whose Agent Is It?

An agent is not on your side just because it speaks in the first person. The grammar of "I found you the best option" is a costume. Underneath it there is a business, and the business has an incentive.

So ask the questions you would ask of any human who handles your money. Does it maximise your welfare, or merchant commissions, or advertising revenue, or the provider's ecosystem? Can you inspect why it recommended a product? Can you set rules (no sponsored results, local shops first, repair before replace, no suppliers with a dodgy record)? Who owns the record of your preferences? And can you take your agent, with its memory and its habits, to another provider, or are you locked in?

That last question is the one that decides everything. The preference graph (the accumulated model of what you want and will pay for) is the most valuable commercial asset in this entire story, and at the moment it belongs to whoever hosts it. If you cannot move it, then the agent is not your representative. It is your landlord's representative, wearing your face.

As an aside, and backed by nothing more than a lifetime of watching how commercial incentives behave: I'd even argue that the agent market will be won by whoever makes switching easy, and lost by whoever tries to make it hard. Every dominant platform in history has done the opposite, and every one of them eventually attracted a regulator or a rival with a cheaper door. I wouldn't put my own money on any firm that thinks the lock-in will hold this time.

This is the same old libertarian question wearing fancy dress. A free person is one who can walk away from a bad arrangement. Property rights over your own preferences is the practical form of that freedom. A market in which you cannot take your data and leave is not a market. It's a company town.

Back to the Kettle

Go back to the woman in Parramatta. She has a new kettle and a good Friday, and she has no idea how many hands it passed through to get there. Amazon may have packed it. A third-party seller may have made the margin. And a piece of software she trusts made the only decision that mattered.

The future is not necessarily a world without Amazon. It may be a world where Amazon's logistics, selection and merchant network endure while the front door, the bit that made it powerful, becomes optional.

The history of disintermediation is not a story of middlemen disappearing. It is a story of power moving to whoever makes complexity feel simple. For a century that was the shopkeeper, then the catalogue, then the search bar, then the marketplace. In the age of agents it may be none of them. It may be the system you've entrusted to act for you.

And that system's loyalty, the one question nobody is asking loudly enough, is up for sale. You had better find out who is bidding.