A layer of software agents is settling between brands and their customers. The product isn't threatened. The buying relationship, however, changes hands. And that's where value is at stake.
Published on 2026-06-01 by Nathalie Lamborghini Dumas.
A layer of software agents is settling between consumer brands and their customers. The product isn't threatened. The buying relationship, however, changes hands. And that's where value is at stake.
Many consumer brands believe they are safe from artificial intelligence. Their product is physical, sensory, made by hand and experienced in person. No machine reproduces it. The reasoning holds for the product. It falls apart as soon as you look elsewhere, where AI doesn't touch the object but inserts itself into the way it is bought.
A layer of software agents is slipping between brands and their customers. These agents make nothing. They search, compare and pay in the buyer's place. For a house that built its value on desirability and on the direct bond with its customers, it is through this discreet channel that the question arrives.
The movement has a name, agentic commerce: AI programs that carry out a purchase end to end, from search to payment. At its I/O conference in May 2026, Google unveiled Universal Cart, a basket that works across the entire web, backed by an open commerce protocol built with Amazon, Shopify and Walmart, and designed so that an agent can ultimately buy on its own, within the limits set by the user. Amazon, for its part, is testing a system that orders on third-party sites on the customer's behalf. Beauty and fashion are leading the way, food follows, the rest gradually falls into line.
The consumer no longer browses pages, no longer compares prices, no longer reads reviews. They hand an intention to an assistant, which takes care of the rest.
In online commerce as we know it, a brand fights for the consumer's attention, through advertising, design, search ranking, placement. In a journey led by an agent, it fights to be retained by it, on the basis of its data, the completeness of its catalogue, its availability, its price.
The shift matters for a premium retailer. An agent ranks an intention according to machine-readable criteria. It does not feel the aura of a house. The brand premium, that attachment which justifies paying more, does not pass intact through an assistant that files options into a list. Selection plays out on the legibility of the offer; affective familiarity carries no weight there.
This preparation for agents has already become a profession. In May 2026, Performics and Publicis Sapient launched an "Agentic Ready" offering to make brands discoverable and purchasable by agents, by restructuring content and catalogues into data that machines can read. As the president of Performics France sums it up, "AI doesn't read the web like a human": it looks for structured data.
This shift does not strike the company uniformly. It cuts value in two.
Everything that goes through a mediated channel, home delivery, discovery ("what is recommended to me nearby"), sales on third-party marketplaces, becomes exposed. The agent can reduce a brand there to a line in a comparison and favour the one that is big, fast and cheap. This is the terrain of disintermediation, where the software intermediary captures the relationship and relegates the supplier to the rank of commodity.
Everything that goes through the direct and embodied bond follows the opposite path. An in-person exchange, a gesture, a place, a moment cannot be delegated to an agent. This part resists, and the abundance that AI pours out everywhere else makes it rarer, and therefore more sought after. The economist Alex Imas, of the University of Chicago, has shown this experimentally. People pay about twice as much for a good when they know that not everyone will be able to have it: scarcity creates desire. And this premium disappears as soon as they think an AI made it, because they then judge it infinitely reproducible. What is experienced in person, or what a human hand shapes, cannot be reproduced: it keeps that premium.
The same technology commoditises delegated distribution and raises the value of direct experience.
For an agent to retain a brand, you have to speak its language: legible price, availability, structured catalogue, normalised data. Making yourself comparable in this way means entering a competition where the brand matters less and where price and speed decide. The more you optimise to be chosen by agents, the more you capture the volume that will flow through them, and the more you erode what the brand drew from being desired for itself.
The other path also costs. Keeping the point of control, that is, remaining the entry point the customer chooses directly, rather than a supplier an agent ranks among others, means favouring your direct channels, stores, website, loyalty, and accepting to remain less visible where agents will distribute a growing share of purchases.
Both options are legitimate. Making yourself legible to machines captures the volume that flows through them and prevents others from occupying that place alone; the price to pay is entry into a comparison that dilutes the brand premium. Staying out of it protects that premium, along with the point of control, but leaves that volume to those who will have played the game; they win on price and speed, where differentiation is not rewarded, and wear theirs down there. The market splits in two, a mediated segment where you compare yourself and a direct segment where you are chosen, and no brand holds both fully.
What places a brand on one side or the other comes down to one thing: is it demanded for itself? When a customer asks for a retailer by name, the agent must propose it, on pain of disappointing them. The one that is desired thus keeps its place, even when the purchase is delegated. Points of sale help maintain this bond and see what the clientele wants; they do not create that desire on their own. An already established retailer starts with that advantage.
The same divide takes shape outside commerce. The Economist is preparing two versions of its content, one rich for readers, the other structured for agents. Alessandro De Zanche, a media consultant, draws a simple rule from it: optimisation for machines becomes a defensive baseline that everyone will eventually build, but it is audience trust that decides who survives.
One caveat, however. Visibility in this layer of agents is not neutral. In the press, titles that have signed an agreement with OpenAI feature more prominently in ChatGPT's answers: according to an analysis by INA, Le Monde alone captured a quarter of the visits the assistant sent back to French media in 2025. Being desired is not always enough; the one who holds the door must also let you through.
None of this is imminent for an impulse purchase made in store. Beauty and fashion are affected first, food comes next, and a purely in-person retailer remains out of reach for now. A brand's exposure depends on its mix of channels: the more it sells through delivery and online commerce, the closer the question gets. A signal to read early, not a project to wrap up within the year.
The right answer depends on the share of sales an agent can decide in the customer's place. That share remains small for those who sell mainly in person, or who are asked for by name; it grows with marketplaces, delivery and, yes, purchases an agent triggers right on the brand's own website. That is what I would look at first.
On what share of our customer relationship do we accept becoming one option among others for an agent, and on what share do we want to remain the choice the customer makes themselves?