Less, but better

• The thesis: 2026 is the year Luxury stops growing as before, and the most useful AI becomes the one that helps produce right. The lever is no longer sales volume, it is precision: fewer unsold goods, less blind replenishment, less wasted material. This protects margin — and, through it, future growth. Telling a model (which measures and advises) from an agent (which decides and executes) is the key to knowing what to entrust and what to keep.
• The vineyard is not a sanctuary free of AI — quite the opposite. Hennessy has run Grapp for years, a data platform for its partner cognac growers: plot-by-plot weather radar, calendar, tracking of sustainable practices. At VivaTech, Moët Hennessy adds Genesis, a model that scores soil health. Data is well established in the vineyard; but these tools measure and recommend — the agronomist decides. A model, not an agent.
• Cutting unsold stock protects margin better than one more campaign. Unsold goods account for 10 to 40% of global fashion production. Fairly Made, awarded at the LVMH Innovation Awards and deployed across 14 of the group's Maisons, tracks material along the chain and flags compliance gaps on its own. The agent that removes the surplus is worth more, margin included, than the one that inflates the basket.
• Agents share tasks among themselves. Perfect Corp (on Claude) shows how one brand's agent hands a task to a partner's beauty agent without opening access to the client. Digital Luxury Group makes 1.4 billion Luxury data points queryable by these agents. The Maison coordinates without doing everything itself.
• The real battle in AI commerce is not visibility, it is getting paid. With Adyen, Sézane lets a shopping agent close the sale while remaining the one who receives the money and keeps the client link. When a customer buys via ChatGPT or Meta, the decisive question is: who collects, the Maison or the platform?
• Three signals outside Luxury: in hospitality, an AI ranking screens at the door and 84% of hotels never appear in it (HotelWorld AI, June 16); Anthropic files for an IPO at 965 billion dollars; the rules of agent-led purchasing are converging on common standards.
• The projection: Sézane, an accessible brand born online, leads the way; a great heritage house will follow where the client relationship weighs most. Watch mid-July, when LVMH and Kering report their first-half results.

An AI agent has value only through what it can call on elsewhere. To answer, it hands a task to a specialised agent, queries a database it does not own, plugs into a partner's service. Left to itself, it does almost nothing.

Two facts from the week show it. Since the start of the year, Digital Luxury Group has made its LuxuryIQ base queryable by AI agents: 1.4 billion Luxury data points that these programs read and use to act, without crawling sites one by one. At VivaTech, Perfect Corp presented its way of making agents talk to one another, and its founder and CEO Alice Chang sums it up: "VivaTech 2026 is a key milestone for Perfect Corp with the presentation of our Agent-to-Agent integration framework." Concretely: one brand's agent hands the beauty task to Perfect Corp's agent, gets the result back, and never opens direct access to its client.

For an industry that learned to make and control every link itself — its tanneries, its supplies, its distribution — relying on an outside agent is a new reflex. It is not about handing over everything: calling on a partner agent forces neither opening your data nor entrusting your client. That is the whole question, and it is concrete — deciding, agent by agent, what you let it do and what you keep for yourself.

Sources:

• DLG LuxuryIQ MCP · EQS News — LuxuryIQ MCP launch · Perfect Corp — Agent-to-Agent at VivaTech.

Three dated Luxury signals this week — a model that advises, an agent that assists, an agent that watches — each read for the Maison.

In the vineyard, an AI model measures and advises; the agronomist decides. At LVMH's DreamGallery during VivaTech, Moët Hennessy presented its partnership with Genesis, a French startup that scores soil health: an index combining stored carbon, biodiversity and water, with recommendations on regenerative practices — when to act, how. It is a model, not an agent: it computes a score and proposes, it executes nothing on its own. And contrary to common belief, the vineyard is not the last domain spared by AI. Hennessy has run Grapp for years, a data platform reserved for its partner cognac growers, with weather radar forecasting plot by plot, a calendar, tracking of sustainable practices. Data is already deeply settled in the vineyard. What is not delegated is the final decision, because the quality of a vintage depends on variables — April frost, August rain — that no model controls. AI informs the gesture there; it does not replace it. The backdrop sets the urgency: Hennessy's US sales fell 6.5% by volume in 2025, and the first-quarter 2026 rebound — Moët Hennessy at +5% organic, its best quarter since 2022 — owes mostly to a favourable calendar effect from the Chinese New Year, shifted to February 17. The group itself warned that this rebound should not last into the second half.

Dior puts an agent to work for its advisors, not for its clients. Also at VivaTech, Dior presented DIVA, an AI agent for its clienteling app. It is not turned toward the client: it works for the advisor. It queries stock in real time, pulls up internal procedure, answers across every country and product category, to free the salesperson from searching and return them to the relationship. This is delegation applied to operations: the agent searches and proposes, the advisor decides and embodies. The doctrine is clear, and I believe it right: the agent backstage, the human on stage. The real governance question remains: if the advisor follows DIVA's recommendation nine times out of ten, who still decides — them or the agent?

Fairly Made turns compliance into continuous monitoring. Bernard Arnault personally handed the Best Impact prize at the LVMH Innovation Awards to Fairly Made, already deployed across 14 of the group's Maisons, including Louis Vuitton, Dior and Céline. It is not one more digital passport: it is an agent that tracks material along the chain and raises an alert on its own when a batch strays from the rules or a supplier steps out of line — with no human having to comb through a dashboard. The regulatory deadline is approaching: the European digital product passport, which will require tracing each product across its life, is rolling out in stages; the act covering textiles is expected for 2027, with application from 2028 at the earliest. For a Maison, compliance stops being a file handled after the fact: it becomes monitoring that runs continuously. "There was no debate in the jury," says Gonzague de Pirey. This is delegation on the most sensitive subject of all: control of one's own manufacturing chain.

Sources:

• LVMH — VivaTech 2026 · LVMH — Maisons × Tech 2026 · Hennessy Viti — Grapp · The Spirits Business — LVMH Q1 2026 · Just Drinks — W&S Q1 2026 rebound · European Commission — ESPR / digital product passport.

Three signals you will find nowhere else in this edition, sketching the ground Luxury is already moving on.

1. In hospitality, an AI ranking screens at the door — and most hotels are not in it

On June 16, at HITEC, HotelWorld AI released the second edition of its study on AI visibility: 2.36 million data points, 2,100 hotel brands, 141,000 properties, 40 countries. The verdict is blunt: 84% of hotels never appear in AI-generated recommendations, and the threshold to enter the global Top 25 jumped 25% in a single quarter. Filip Boyen, co-founder of HotelWorld AI and former head of Small Luxury Hotels of the World, puts it plainly: "Travellers are no longer scrolling through pages of results. They are receiving a handful of recommendations. If your hotel is not part of that shortlist, it may never enter the guest's consideration set." Why it matters: hospitality is the antechamber of experiential Luxury, and it is already living, at full scale, what jewellery and wines will discover in turn — software that selects offers before the customer even searches. When it chooses who appears, absence is no longer a mere lack of visibility: it is elimination by default. A telling detail: Aman and Auberge enter the ranking not through the size of their estate but through the coherence of their brand narrative — proof that one can exist in AI answers without volume, provided a dense editorial corpus.

2. Anthropic has filed for an IPO — 965 billion dollars

On June 1, the maker of Claude confidentially filed its registration with the US regulator, in the wake of a 65-billion-dollar round that values it at 965 billion dollars. Why it matters for a Maison: Claude already runs Perfect Corp's beauty agent, and tomorrow perhaps your own agents. The day the vendor of the AI that powers your operations must report every quarter to public-market shareholders, its priorities become theirs, not yours. The question is no longer only technical. It is one of governance: part of your operations would depend on a player bound by a quarterly market logic that is not yours. That is the concrete argument for demanding, today, an exit and an alternative to any strategic agent.

3. The rules of agent-led purchasing are taking shape

The sector is aligning on common standards. The big platforms — OpenAI, Google — and software-to-software payment systems are gradually adopting the same technical rules, so that shopping agents can transact from one service to another. Specialists such as Scot Wingo, in his Retail Gentic newsletter, track this closely: the question is no longer whether agents will buy, but on which standard. Why it matters: when a common rule takes hold, it opens the market to everyone at once. The Maisons waiting for things to "settle" will find the standard was written without them — and that they will have to comply after the fact, instead of weighing in while it is being written.

Sources:

• Hospitality Net, June 16 · Searchless.ai — AI Visibility Index Hotels 2026 · Fortune — Anthropic, June 1 · New York Times, June 2 · Retail Gentic — Scot Wingo.

Yesterday, AI promised growth: more conversion, more customers, more of everything. Now that the Luxury market is slowing, the most useful agent does the opposite: it helps produce less, but right. And that is where margin is defended.

A conviction that goes against the grain: in today's Luxury, the most profitable AI is not the one that sells more, it is the one that helps produce right. The double-digit growth that carried the sector for fifteen years has stopped. Hennessy's US sales fell 6.5% by volume in 2025, Luxury fashion slowed at the start of the year, and the Business of Fashion recently devoted a whole framing to those who will survive the end of this easy growth. Yet the entire promise of AI, for three years, has been calibrated to make things bigger: recommend more, convert more, personalise to sell more. That AI has not disappeared, and it remains useful. But it is no longer the centre of gravity. The Maison's subject is no longer only to bring more people into the sales funnel. It is also, and first, to defend margin from below: fewer unsold goods, less blind replenishment, less wasted material.

A clarification, to avoid the misreading. Producing right is not producing less on principle, nor giving up selling. It is matching production to real demand, piece by piece and market by market. A Maison that cuts its unsold stock by a few points frees up margin; that margin funds creation and store openings; so today's precision feeds tomorrow's growth. The agent that removes the surplus and the one that supports the sale do not clash: they serve the same profitability from two ends.

We already see it on raw material. Fairly Made — awarded at the LVMH Innovation Awards, deployed across 14 of the group's Maisons — tracks leather and fabric along manufacturing, hunts waste at the source and flags a non-compliant batch on its own. On quantities, the same principle is gaining ground: replenishment agents adjust restocking on their own, store by store, according to real sales, so as to stop piling up pieces that will not move. The stakes are vast — unsold goods represent, by industry estimates, 10 to 40% of global fashion production. That is where the clearest margin reservoir lies today.

Three reading keys to gauge what this changes.

Economics

You no longer defend your margin by selling a lot, but by producing what will sell. When the market slows, every piece made and then left unsold costs twice: in production, then when it is discounted — which damages the price image. An agent that removes a few points of unsold stock protects margin better than a campaign that adds a few sales. It is less glorious than triumphant growth. It is more solid.

Sociology

Luxury lives on scarcity, and a forecasting agent can produce two opposite kinds of scarcity. There is the scarcity you organise — making just enough to sustain the waiting list and desire. And the scarcity you suffer, when a Maison cannot keep up with demand. Handing this calculation to an agent means being able to steer the first instead of enduring the second. The risk is in the nuance: a scarcity too well optimised stops being a mystery and becomes a mechanism, and the client ends up sensing the calculation behind the desire.

Philosophy

"Weniger, aber besser" — less, but better: the phrase is Dieter Rams's, the tenth of his ten principles of good design. I am an unreserved admirer — these ten principles still guide designers today, and Jony Ive drew from them the essence of the language that made Apple, never hiding it. Rams spoke of objects; I divert his rule toward the decision to produce, and I own the diversion — it is for a good cause. The agent that helps produce right finally gives this phrase an industrial reality, and reveals its limit. An agent can remove what does not sell; it cannot give up what would sell too well. Deciding not to produce a highly desirable piece, to preserve the Maison's prestige, remains a human decision. The sociologist Hartmut Rosa contrasts availability — making the world measurable, predictable, optimisable — with resonance, that living relation nothing guarantees. An agent excels at making production available; it does not manufacture the resonance that makes an object desired. The useful distinction is therefore not "to delegate or not": it is to entrust the agent with the calculation of what does not sell, and to keep your hand on the gesture of restraint that makes prestige.

Sources:

• FashionNetwork — reducing losses, June 2 · Business of Fashion — the end of easy growth (subscriber access) · LVMH — Fairly Made, VivaTech · ABC17 / CNN — cognac, June 1.

Luxury spent fifteen years removing the middleman. The department store, the wholesaler, the multi-brand retailer: one by one, the Maisons took them back in hand to speak to their client directly, collect the money themselves, keep the data. Direct-to-consumer was the great conquest of the decade. Yet a new middleman returns through another door — and this one, the Maison does not own.

This middleman is called ChatGPT, Meta, Google, Microsoft, Amazon — and tomorrow, no doubt, Tesla or SpaceX. When a customer asks a shopping agent to find them a dress, a bag, a perfume, and that agent compares, selects then closes the order, the sale no longer goes through the Maison's boutique or site. It goes through the agent. And the real battle, contrary to what has been repeated for eighteen months, is not appearing in the AI's answer. It is remaining the one who gets paid — the one with whom the customer saved their card, the one who keeps the direct link — when the sale passes through a platform that is not the Maison.

That is exactly what Adyen Agentic sells, and why it is worth dissecting. Three layers, three functions. The first, Agentic Feed, sends the shopping agent, live, the catalogue, prices and stock — and it is the Maison that chooses what it may show. The second, Agentic Cart, handles everything between the agent validating the basket and delivery: checkout, taxes, order processing, stock updates. The third, Agentic Payments, is the most important. It verifies that the payment triggered by the agent is legitimate, protects against fraud, and above all ensures that it is the Maison — not the platform — that actually receives the money. The result: even when an agent pays via ChatGPT, it is the Maison that gets paid. The customer buys through the agent, but they buy from the Maison.

Now, it is not a heritage institution among the first to use it: it is Sézane, born online, positioned in accessible-premium. Nicolas Benoist, its chief technology officer, explains the choice plainly: "The rapid evolution of Agentic Commerce protocols means retailers need a trusted partner who stays ahead of the curve. With Adyen Agentic, we were able to build on proven, battle-tested foundations to deploy quickly and confidently." And here is a nuance that counts. If accessible-premium leads the way, and not the great heritage houses, it is because it has less to lose: no two centuries of tradition to protect, a client already used to buying in one click, a brand whose value does not rest on the ceremonial experience of a boutique. The great heritage houses, for their part, watch while others move. The technical plumbing, besides, is already in place among the majors: Adyen equips Kering across 27 markets. The day one of these groups switches its infrastructure to agent-led commerce, the first great Maison able to be paid automatically by agents will be born. That moment has not yet come — but technically, nothing holds it back any longer.

There is the pendulum. A payment agent can mind the till: verify, secure, remit. It will never hold what turns a buyer into a client: recognition in the boutique, the handwritten note, the attention that precedes desire. Technology speeds up the transaction; it does not manufacture the relationship that brings people back. The decision point, in agent-led selling, is exactly there: you can let an agent collect, provided you remain the one who receives the money and keeps the contact. You never entrust it with what brings a client back. Who gets paid when an agent buys? The Maison that will have understood, in time, that what counts in 2026 is no longer only appearing in AI's answers — it is remaining the one who collects the sale.

Sources:

• Adyen Agentic · Option Finance, June 16 · FFNews — Adyen × Kering, 27 markets · Retail Gentic — ACP / UCP / AP2 protocols.

To a Maison's leader.

Do you know, in your Maison, what your AI agents already decide on their own — and what they merely recommend to a human who rules? And who, in your House, decided that split: a named person, or no one, deployment after deployment, without it ever being arbitrated?

The question is not rhetorical. In most Maisons I observe, no one has set this split: it shifts on its own, with each agent plugged in, each recommendation followed, until the day you realise you have delegated far more than you thought. Deciding it consciously is the act of governance that separates the Maison that decides from the one that submits.

This is exactly the work of the Agentic Delegation Plan: reviewing, function by function, what a Maison can entrust to an agent and what must stay decided by a human — then writing it down, instead of letting the split happen on its own, agent after agent.

The Agentic Delegation Plan →

An agentic technology is maturing fast among the pioneers, and Luxury has not yet seized it: the agent that negotiates. Not the one that recommends a product nor the one that collects payment — the agent that discusses a term, a price, a deadline, an availability, on behalf of whoever employs it.

The building block already exists. The major AI vendors describe agents able to run a simple negotiation: propose, take a counter-offer, adjust, close within a pre-set range. On the mass-retail side, Bain documented in June shopping agents that compare prices from one retailer to another and fragment the basket — a sales matter, on the customer side. But for Luxury, the most mature ground is elsewhere: in buying, where margin is decided.

Three areas of work, all business-to-business. The negotiation of commercial leases first: a flagship rent on a great avenue runs into the millions, and an agent that prepares the scenarios, gathers comparables and simulates a renewal facing the landlord shifts the balance of power. The purchase of raw materials next — leather, cashmere, precious metals, stones: an agent can track prices, pre-negotiate volumes and deadlines within a validated range, and keep the pressure on suppliers often in a position of strength. Terms with subcontractors and retail partners last — façonniers, department-store corners, wholesale — where a point gained on a discount or a deadline ripples across an entire season.

There remains a more visible angle on the customer side, but secondary: resale, where an agent could appraise a second-hand piece and hold a floor price instead of negotiating everything by hand. The essential is not there. The real subject is not the agent that marks down — that would be a misreading — it is the agent that defends value, in buying as in leasing. The real question: the day a Maison's landlords and suppliers negotiate with agents, will it have its own, set to its own rules, or will it let the conversation happen without it?

Sources:

• Bain — pricing for humans and bots, June 15 · Bain — agentic operating model, NRF APAC, June 9 · Retail Gentic — agentic purchasing protocols.

• Who Will Survive Luxury's Post-Growth Era, Business of Fashion. The sharpest framing of the current break: the end of easy growth reshuffles the deck, and useful AI is no longer the one that makes things bigger. Read it to understand why the agent that cuts unsold stock becomes the priority. (Subscriber access.)
• How data and AI can reduce losses in the fashion industry, FashionNetwork, June 2. Unsold goods represent 10 to 40% of global fashion production. The article shows how tools that forecast sales better cut these losses upstream. Read it to understand, figures in hand, why agents that calibrate quantities become a priority.
• Online Grocery Pricing for Both Humans and Bots, Bain & Company, June 15. The best current analysis of shopping agents that compare prices and fragment the basket — with a countermeasure directly transposable to Luxury: the exclusive assortment agents cannot find anywhere else. The seed of this week's Bêta Luxury.

The next two weeks' agenda comes down to a few dated appointments.

• July 6–9: Haute Couture Autumn-Winter 2026-2027 in Paris, 30 Maisons, 30 shows. Schiaparelli opens on the 6th, Chanel presents Matthieu Blazy's second couture on the 7th, Balenciaga by Pierpaolo Piccioli and Jean Paul Gaultier by Duran Lantink show on the 8th. Watch for: the exact place these Maisons leave — or not — for AI in the communication of their shows.
• July 28: LVMH reports its first-half 2026 results. We will read whether the first-quarter cognac rebound held, or faded once the Chinese New Year effect passed.
• July 29: Kering reports its own, the first full half-year accounts under Luca de Meo. The market will look for the first effects of the ReconKering plan.
• By 2027: the European act on the digital product passport covering textiles is expected, with application from 2028 at the earliest — the deadline that turns traceability from a project into an obligation.

This year, Luxury learns a gesture it never had to make: entrusting part of its decisions to agents, without entrusting those that define it. Replenishment, forecasting, traceability, it can delegate without betraying itself. The gesture of the atelier and the relationship that brings the client back, never. Producing right is learning this distinction and holding it. The market will no longer reward the one who produces the most, but the one who produces what is needed — and stays master of what is decided in its name. Luxe oblige.

Sources:

• Fashionista — Haute Couture July 2026 schedule · LVMH — financial calendar · Kering — financial calendar.