The Real AI Fight Is Moving to the Checkout
For the past two years we have told ourselves a simple story about where the value in AI lives. Labs chase the lead, founders bolt the best model onto their product, and the frontier effectively becomes a proxy for power. It is an easy story to hold onto, because it is the one the industry keeps repeating. But it misses where the real leverage is moving. This week OpenAI made that shift visible with a small, easy to miss feature called ChatGPT Wallet, and it is worth pausing on what it signals before everyone rushes past it.
ChatGPT Wallet lets an OpenAI agent, a dot, decide what you should buy and then complete the purchase on your behalf. It is built with American Express, Mastercard and Visa, uses a payment token in place of your actual card number, and asks for a biometric check before each purchase. For now it is limited to Pro subscribers in the United States. The checkout itself is almost the least interesting part of the announcement. What is interesting is the seat the assistant now occupies. It has moved from answering questions and drafting email to being the place where a shopper is introduced to a product, compares options, and pays. That is a different position in the economy, and it changes who owns what comes next.
Consider what a merchant currently owns. A storefront, a brand, a set of product pages, the discovery that brings a customer in, and the relationship that brings them back. A customer who orders through a dot may never see any of that. The agent reads the product feeds, compares the options, chooses, and checks out. The merchant still fills the order, but it has surrendered the front of the transaction. That front has been the most valuable real estate in retail for a century. The shopwindow, the homepage, the search result, and the loyalty that comes from a repeat customer knowing your name. Most of it now sits on the other side of a conversation the customer never had with the merchant.
The most revealing part is the guidance OpenAI published for businesses on the same day, its Commerce Partner Playbook. It reads as an instruction sheet. Let signed agent traffic through, keep your product feeds current, work with risk providers so agent payments do not get declined. Read it closely and you notice the reversal. The merchant, who used to set the terms of how customers found and bought, is now being told how to make itself visible and acceptable to somebody else's agent. The business that once had only to be good at selling is now being asked to be good at being found by a machine. That is a quiet change in who controls access, and it is worth not missing.
There is an irony here that connects to an argument I have made all year. Most of the anxiety about AI has been framed as dependency on a model provider, running a business on an engine you do not control. But this is dependency of a more intimate kind. It is not about chips or weights. It is about the customer relationship. Businesses spent decades with their stores, their names, and their service as the thing a customer trusted. If the assistant becomes the default way a customer decides and buys, then that trust, and with it the ability to bring the customer back, migrates to the assistant. The dependency moves from compute to distribution, and distribution has always been the harder thing to rebuild.
I should be honest about how early this is. The launch is limited, gated to Pro users, and OpenAI has been down this road before. Its earlier attempt at a quick checkout was rolled back within months, and its first real push into shopping had to be redone. Most people will keep buying the old way for years, and agentic commerce in general is far from proven. It is entirely possible this particular feature becomes one more abandoned experiment, and I would not stake the future of retail on the first version of it. The point is not that this exact product will win. The point is the direction it points and where it is taking the money.
The direction is that the transaction is becoming the place where AI captures value. We have watched this happen with search and with apps. Whoever sits at the point where a buyer decides and a seller gets found owns the toll. The app store took a cut of almost every sale. Search took control of discovery. The card networks took a small percentage of nearly all of it. Each of those middlemen ended up more valuable than most of the businesses that depended on them. The pattern repeats because the middleman sits at the least replaceable spot, the one place both sides have to pass through. An assistant that shops for you is exactly that spot.
Notice that OpenAI says it is not taking a cut of transactions to start. That sounds generous, and it is worth reading the way business people always read that phrase. Every platform that eventually took a toll began by making itself too convenient to refuse. The card networks did not start by charging merchants. They started by making payment frictionless and then built the toll into the market. The app store subsidized mobile until mobile was the only way in, then collected rent on the way through. When a company with this much reach tells you it will handle payments for free, the sensible reading is not charity. It is acquiring the rails now so the toll, in whatever form, can be switched on later. I am not predicting the fee. I am describing the familiar shape of the move.
If the pattern holds, the surprising winners are not necessarily the labs with the best models. They are the ones that hold the relationship at the point of purchase. For a merchant, the practical consequence is that the customer relationship it felt it owned, the store, the brand, the repeat visit, largely moves behind a curtain it cannot reach. The loyalty a business used to earn through experience and service now belongs to whoever the customer's agent trusts, and that party may not know the merchant's name. This is not a claim that e commerce is collapsing. It is a claim that the spot where value sits is moving, and it is moving to a place no individual merchant controls.
The fair thing to acknowledge is that this cuts both ways. A small merchant is largely invisible in a world where discovery runs through advertising budgets and the algorithms of a few giant platforms. An agent that reads your product feed and genuinely weighs price, quality, and fit could be a genuine equalizer, the way a customer finally finds your shop because it was the best answer rather than because you won the auction. There is a version of this I like. The worry is that the comparison still runs on somebody's rails, and whoever owns the comparison owns the filter. The storefront disappears either way. The real question is who decides what the customer sees, and that question now has a fresh, quiet answer.
There is a clean parallel in how distribution has always eaten the merchant's relationship. Catalogs, then marketplaces, then search, each stood between a business and its customer. Each was convenient for the buyer and quietly taxing for the seller. Mail order made shopping easier and moved loyalty from the shopkeeper to the catalog. Marketplaces did it again. Search did it a third time. Every wave made a middleman richer and made the individual seller more dependent on access it did not control. The AI assistant is the next catalog, except it does not merely show you the goods. It decides, negotiates, and pays. It is the concentrated version of everything the middlemen before it did one at a time.
I am not trying to tell anyone what to do, and I am not predicting the exact shape this market takes. What I am saying is that the question founders used to ask about AI, whether to build on this model or that one, is rapidly becoming the wrong one. It matters less which engine your product runs on than whether your product holds a relationship a customer comes back to on their own. If that relationship sits with an assistant somebody else owns, then the margin in your business follows it out the door. The moat in this era is not the model and not the code. It is being the thing a customer trusts enough to return to without a middleman in between.
None of this means I think the wallet is a disaster or that every merchant should panic. I have argued before that AI makes it cheaper than ever to build a product, and that the scarce resources are distribution and trust. This week is a reminder that the same is true when the product is a purchase. The assistant is not the villain and the convenience is real. But the clock on the customer relationship started the day the assistant learned to check out, and it does not care how good your website is.
Every era hands the most trusted intermediary the keys to the checkout. The AI era is only deciding who that intermediary is, and the window to be that person is short. The customer belonged to whoever they trusted at the counter. The interesting question for the next decade is not what that counter sells. It is where it moved, and whether it moved somewhere you can still stand behind.