Every year about now, retailers spend a fortune on the last three feet of the sale. The countdown clock. The “only 2 left in stock.” The photo of a mug of cocoa sitting next to the throw blanket you have been circling for a week. All of it aimed at one target: a tired person on a phone late at night, a single thumb-swipe from closing the tab. Move that person and you have made your quarter. The strange thought hanging over this holiday season is that the person doing the shopping might not be a person.
From Describing To Doing
For two years, AI in retail has mostly meant better talking. Chatbots that field sizing questions. Software that churns out product blurbs by the thousand. A search box that finally parses “something warm for my dad, who runs cold, but nothing itchy.” Useful, all of it, and completely passive. The machine described; you decided; you clicked buy.
That gap is where things are starting to move. In October, Anthropic released a beta of something it calls Computer Use — a version of its Claude model that operates a computer the way we all do. It looks at the screen, moves the cursor, clicks, types into boxes, and strings those steps into an actual task. It is slow and it fumbles, which the company cheerfully admits. The point is not that it is good yet. The point is what it treats a website as: not a page to summarize, but a place to get something done. That is the piece an errand-running shopping agent always lacked.
The commerce-flavored versions are arriving alongside it. Perplexity, the AI answer engine, has started pulling shopping straight into its results instead of handing you off to a list of blue links. Amazon has spent the year growing Rufus, the assistant tucked inside its app, from a curiosity into something that increasingly plants itself between you and the buy button. Everyone else — Google, the Shopify crowd, a swarm of startups — is coming at the same idea from another angle.
None of it is a hands-off “just buy it for me” machine yet. But for the first time you can walk into Black Friday, say “the AI found it and set the whole thing up,” and mean something real.
We Have Met This Bot Before
If this feels like it has happened before, that is because a version of it has. In the late 1990s a small zoo of shopping bots — BargainFinder, Jango, mySimon and the rest — promised to scour the young web and hand you the lowest price on a CD or a paperback. The technology worked well enough to frighten the merchants, and some of them did the rational thing: they blocked the bots outright, refusing to be flattened into a single number on a comparison table. It turned out that shopping-as-a-race-to-the-cheapest-price was a game stores could quietly decline to play. The bots faded. Price comparison shrank into a tidy corner of the web instead of swallowing the whole thing.
What is different this time is that the agent does not stand outside the store scraping prices. It walks in the front door, indistinguishable from a customer, and does what a customer does. You cannot block it the way you blocked a price crawler without also blocking the sale. And it is not only comparing on price — it is comparing on whatever its owner asked for, which might be the delivery date, or the return policy, or “the one that will not fall apart,” a judgment stitched together from a hundred reviews the shopper would never have sat down to read. That is a much harder thing to refuse, and a much harder thing to game.
What Should Worry Retailers
Here is why that should worry retailers, and it is not the obvious reason. An agent does not get tired at 11pm. It does not feel the countdown clock ticking down. The cocoa photo does nothing for it. It will not abandon a full cart because the checkout button was an ugly shade of orange. It reads the specs, weighs whatever it was told to weigh, and executes. So the whole apparatus of urgency and nudging and lifestyle photography — the stuff stores have spent a decade tuning — is now aimed at a customer who has quietly left the room. It is a pitch delivered to an empty chair.
Follow the thread and it gets worse for the incumbents. When someone tells an assistant “find the best-rated cordless vacuum under $300 that will arrive by Friday,” the winner is chosen inside the assistant’s head. There is no results page to buy your way to the top of, because nobody is looking at a results page. What wins instead is being clean and honest to a machine: real stock counts, real shipping dates, reviews it can trust, specs that are correct and complete.
These are unglamorous things. Nobody’s marketing team gets promoted for making sure the product feed lists the right cord length, the accurate weight, whether the thing works with last year’s model. That work has always lived half-done in a back office somewhere because a human shopper would forgive a blank field and fill the gap with the pretty photo and a bit of hope. An agent will not. To an agent a blank field is just an absence — a reason to pick the competitor who bothered to fill it in. The boring back-office data, the stuff no customer ever once praised, quietly becomes the storefront.
There is a version of this that is good news for the little guy and a version that is not, and it is too early to say which one lands. In principle the agent is a great equalizer: it does not care that your shop has four employees and no ad budget, only that your data is right and your product is good, and a small merchant can absolutely get those right. In practice, the agents most people will actually use are being built by the same handful of giants who already own the shelf. If the assistant the average shopper reaches for lives inside Amazon, then “optimize for the agent” mostly resolves to “rank well on Amazon,” which is neither a new instruction nor a liberating one. The tool that could flatten the field is being handed out by the people with the most to lose if it ever does.
Whose Side Is The Bot On
Which raises a question: what is an ad worth when the thing looking at it has been told, in plain English, to ignore ads and get the user the best deal? Nobody has an answer. A great deal of expensive infrastructure is riding on there not being a good one.
Close behind it sits a less funny question: who does the agent actually work for? It feels like it works for you — you asked, it fetched. But somebody built it, somebody pays to keep it running, and somebody, sooner or later, is going to notice that whichever product the agent picks is worth a fortune to the maker doing the picking. The gap between “the assistant found you the best vacuum” and “the assistant found you the vacuum whose manufacturer cut a deal with the assistant’s owner” is completely invisible from where the shopper sits. We spent twenty years training ourselves to squint at search results and guess which ones were paid for. That reflex does not carry over, because the agent does not show you the results. It just hands you the box.
Who Owns The Last Gate
Let us not oversell where this sits November 2024. It is not running your Christmas list unsupervised, and it should not be. Handing software your card and permission to spend it is a big ask, and the ways it goes sideways are not exotic: a misread price, the wrong quantity, last year’s shipping address, a checkout flow that ties it in knots. A bad recommendation you shrug off. A bad purchase you have to phone someone about. So every sensible version being built right now keeps a human at the final gate — let the agent do the tedious hunting and comparing and cart-filling, then stop and wait for a person to get the actual money. Agents pull friction out of shopping everywhere except the one spot it belongs.
There’s also a whole category of shopping the agent is quietly bad at, and the holidays are made of it. Nobody delegates a present. You can hand off “reorder the dishwasher pods” without a flicker of feeling, because there is no meaning in a dishwasher pod. A gift is a different animal. Half the point of the thing is that you chose it — that you spent the attention, noticed the person, took on the small risk of getting it slightly wrong. An agent that selects the statistically optimal sweater for your sister has produced a correct object and missed the entire occasion. So the season fuelling all this agentic momentum is also the one most stubbornly resistant to it. The printer ink and the phone chargers and the laundry detergent will go to the machines first. The things we actually get sentimental about will be the last to fall, if they fall at all.
What Are The Real Changes
Whether people eventually relax that final gate is the real question, and it is a question about us, not the software. Convenience usually wins these fights. It rarely wins them all at once, and never quite on the timetable the people selling it would prefer.
That leaves stores in an uncomfortable spot with no clean exit. Lean hard into serving agents and you gut the brand — the images, the story, the loyalty — that still moves the actual humans who are essentially all of your sales today. Ignore agents and you risk going invisible in a channel that is tiny now and growing on a curve that has embarrassed skeptics before. “Do both” is the sane answer and also the expensive one because it means running a single storefront for two customers with opposite tastes: one who wants to be seduced, and one who wants the spec sheet and to be left alone.
How any of this lands at the register, nobody knows. The season has not happened. Anyone quoting you a figure for how much of this Christmas the robots did is making it up.
What has changed is not the sales numbers. It is that the demos stopped being demos. A month ago you could argue about whether software could push a real purchase through a real checkout on someone’s behalf. You cannot really argue about that now. What is left to find out is whether we will let it — and how fast the stores built for human eyes can rebuild for a customer that does not have any.
About The Author:
Maxim Kozlov is a specialist in e-commerce with years of experience building the online systems retailers rely on to sell their products and manage customer relationships. As artificial intelligence reshapes the industry at pace, his current focus is the convergence of AI and commerce technology — and the new patterns of buying and selling starting to emerge from it. He writes about where those two worlds meet, and what it means for the businesses caught in between.
