← Newsletter Edition #3 · March 12, 2026

The workshop threshold

Where AI stops, where creation begins. Agent-to-agent commerce. The bottle law.

SECTION

TL;DR (Too Long; Didn't Read)

  • Burberry unveils the numbers behind Penguin, its AI clienteling platform: +24% average transaction value, 100+ advisors deployed, DataIQ "Most Innovative Use of AI (Global)" award
  • Demna publishes AI-generated images on Gucci's official Instagram account: massive backlash ("AI slop"), reigniting the debate on AI in fashion creation
  • Shiseido launches the first beauty product co-formulated by an AI agent: its VOYAGER platform explored 500,000+ formulations autonomously
  • Meanwhile, fashion responds with craft: on Fashion Week runways, Prada, Bottega Veneta, Loewe, and Miu Miu celebrate imperfection as a marker of authenticity
  • Mastercard unveils "Verifiable Intent": a cryptographic framework proving that an AI agent purchase was authorized by a human
  • Pernod Ricard structures its data for agent-to-agent commerce: Harvard Business School case, 85% internal adoption, rising sales
SECTION

WHAT'S MOVING

Burberry Penguin: +24% average basket thanks to AI clienteling

On February 25, at the DataIQ Awards 2025 in London, Burberry won the "Most Innovative Use of AI (Global)" award for Penguin. Not a customer chatbot, not a showpiece: an AI assistant reserved for in-store sales advisors. The salesperson describes what their client is looking for ("a trench for a wedding in Tuscany") or photographs a piece brought in by the client, and Penguin, an LLM (Large Language Model) fine-tuned on Burberry's entire catalog, instantly proposes complete matching looks, available in stock. 100+ advisors deployed across all markets. Result: +24% average transaction value. The first published financial KPI (Key Performance Indicator) of an AI clienteling tool in luxury. Not a POC (Proof of Concept), not a pilot: a full-scale deployment confirming what the first two editions of this newsletter argued.

Source: DataIQ Awards, Feb. 2026

Prada x Meta: when the AI agent becomes a luxury accessory

Mark Zuckerberg in the front row at the Prada Fall/Winter 2026 show in Milan, seated between Andrea Guerra (CEO of the Prada Group) and Lorenzo Bertelli (executive chairman of Versace). When asked backstage about a potential AI glasses collaboration with Meta, Miuccia Prada offered: "Maybe. Who knows." The Ray-Ban Meta smart glasses (connected glasses with built-in AI assistant: hands-free photo/video capture, real-time translation, object recognition, voice commands) sold 7 million units in 2025. EssilorLuxottica, which manufactures them, also holds the Prada license for ten years. On the other side, Kering Eyewear x Google is formalizing its own AI glasses under Android XR. Two rival tech ecosystems, two rival luxury groups, same bet: turning a tech gadget into a fashion accessory.

If Prada signs, it would be the first time a luxury maison integrates an AI assistant into a product worn on the face. AI leaves the screen, the back-office, the communications department. It enters the product.

Source: TechCrunch, Feb. 2026

Mastercard: a trust framework for AI agent purchases

Without proof of human authorization, no maison will accept an agent buying a 3,000-euro bag on its behalf. Mastercard just created that proof: "Verifiable Intent", a cryptographic framework certifying that an AI agent purchase was explicitly authorized by a human. Protocol-agnostic, compatible with the three agentic commerce protocols in deployment: UCP (Universal Commerce Protocol, Google), ACP (Agentic Commerce Protocol, OpenAI/Stripe), and AP2 (Agent Payments Protocol, Google Cloud). On the consumer side, nothing changes. On the infrastructure side, it is the trust layer that was missing for brands that sell trust as much as products.

Source: Mastercard, March 2026

OpenAI pulls checkout from ChatGPT: the infrastructure remains, usage retreats

OpenAI abandons direct payment in ChatGPT, five weeks after launching it. Five weeks. Even a pop-up store lasts longer. Near-zero conversion, a dozen Shopify merchants live out of millions of stores. The ACP (Agentic Commerce Protocol), co-developed with Stripe, remains active, but ChatGPT refocuses on discovery: recommending products rather than selling them. In edition #2, we analyzed the record $110 billion fundraise as the signal that agentic commerce infrastructure was in place. This retreat refines the picture: the infrastructure exists, usage does not follow. For luxury, this is a reprieve. You still have time to structure your data for agents. Less than you think.

Source: Skift, March 2026

How far does AI enter? Estee Lauder creates the role that decides.

This edition raises a question: how far do we let AI into the value chain? But in most luxury groups, no one has the mandate to decide. The CTO manages infrastructure. The CDO manages e-commerce. The artistic director protects creation. Each defends their turf. In the boardroom, it goes by a nice name: collegial governance.

Estee Lauder Companies (La Mer, Tom Ford Beauty, Jo Malone London, Clinique: $15.9 billion in revenue) just created the answer: a Chief Technology, Data & Analytics Officer, a role unprecedented in luxury beauty, entrusted to Brian Franz (ex-Diageo, ex-PepsiCo International). A single C-suite mandate covering tech, data, and analytics. To mark the shift, the group transferred a $500 million contract from Wipro (Indian IT provider) to Accenture. Without this role, who decides whether AI enters R&D, clienteling, or communications? The three cases in this edition show that the answer cannot remain implicit.

Sources: BoF, March 2026 | Business Standard, March 2026

SECTION

PARADOX OF THE WEEK

On February 27, Gucci publishes AI-generated images on Instagram. Immediate backlash: "cheap," "lazy," "AI slop." At the same time, Shiseido announces the first cosmetic co-formulated by an AI agent. Complete silence. No controversy.

Both maisons use AI in their creative process. One shows it, the other does not. The first is sanctioned by its own customers. The second is praised by its R&D department. You don't say "AI-generated," you say "formulation innovation."

The paradox: in luxury, the problem has never been AI itself. It is where you make it visible. Luxury has always shown its process: Hermes stages its saddlers at the Festival des Metiers, LVMH opens its workshops during the Journees Particulieres, Chanel films the petites mains of haute couture. The hand, the gesture, human imperfection ELEVATE the product. But the algorithm? Kapferer and Bastien formalize it in The Luxury Strategy (3rd edition, Kogan Page): luxury sells a dream, not an industrial process. Yet AI is perceived as reproducible, universal, accessible to all. Showing the hand proves uniqueness. Showing the algorithm destroys it. The Exception invariant collapses: if an AI can do it, it is no longer unique.

SECTION

DEEP DIVE

The workshop threshold: when AI agents move from the warehouse to creation, luxury must draw a boundary it has never had to define.

In edition #2, I defended a thesis: the AI that delivers results in luxury is the one you cannot see. The RealReal, $42 million in EBITDA thanks to warehouse AI. What happened over the past two weeks forces us to go further. AI is no longer content with optimizing operations. It is pushing through the doors of creation. And there, no one knows who is holding the door.

But "creation" in luxury often encompasses distinct realities. The studio: the artistic director draws, explores, imagines. The workshop: the artisan shapes, sews, sets. The manufacturing unit or production facility (it exists in almost every maison, even if no one talks about it): the product is manufactured in series, however small. Luxury has always showcased the workshop and the studio. Hermes films its saddlers. Chanel opens its haute couture ateliers. Showing the human hand proves excellence. Showing the algorithm? No one has yet figured out how to make that noble.

The philosopher Walter Benjamin called "aura" this irreducible quality of a work: tied to a hand, a place, a moment (The Work of Art in the Age of Mechanical Reproduction, 1935). Kapferer speaks of the "dream component." Two vocabularies, one same intuition: what can be reproduced identically loses its value.

The hand nurtures the aura because it is unique, irreproducible, slow. The algorithm dissolves it. AI in manufacturing? No one visits the manufacturing floor. AI in the workshop? The aura is at stake.

Three maisons, in two weeks, drew three radically different lines on the same question: how far do we let AI in? In cosmetics, the customer never sees the lab. In haute couture, the workshop IS the product. In leather goods, the artisanal gesture is the communication. The threshold for AI is not the same depending on whether you are Shiseido, Valentino, or Hermes.

The cost of transparency without value

This edition's PARADOX illustrates it: Gucci sanctioned, Shiseido praised. Same sequence in December 2025 when Valentino published DeVain, a campaign entirely generated by AI. Rebecca Swift, VP Creative at Getty Images (the world's leading image bank), offers the diagnosis:

Consumers consider AI-created works less valuable than human-made images. For expensive brands, standards are higher.

-- Rebecca Swift, Getty Images (Glossy, Feb. 2026)

My takeaway: AI is already in the studio. NXN Labs (a Stanford and Microsoft alumni startup) has signed paid contracts with European luxury maisons to automate visual production via generative AI. Heuritech predicts SS27 trends at Premiere Vision with 91% accuracy. But the maisons that embrace AI in their communications rather than in their processes pay the cost of transparency without reaping the value. It is not AI in the studio that shocks. It is AI in the storefront.

AI in the bottle: the Shiseido precedent

At the opposite end, Shiseido quietly announced the first beauty product co-formulated by an AI agent: a suncare mist developed via VOYAGER, its augmented R&D platform. VOYAGER explores 500,000+ formulations autonomously, evaluates interactions between ingredients, and recommends combinations that a human researcher would not have tested. 80% of Shiseido researchers use it daily. The product will be marketed this summer. No negative comments. No controversy. R&D does not have an Instagram account.

I am convinced this is an underestimated turning point. VOYAGER is not an image generator. It is an R&D agent: an autonomous system that navigates a solution space, evaluates, iterates, without human intervention at every step. AI does not enter communications. It enters the bottle. And the law of silence holds, because the customer only sees the result, never the process.

This is what I call the bottle law: in luxury, the AI that matters is the one you put in the product, not the one you display in the storefront.

Shiseido was able to deploy VOYAGER because its R&D was already data-driven: structured ingredient database, digitized history. The same prerequisite as Pernod Ricard in our STORY: structured data makes AI possible, not the other way around.

Another unexpected illustration: in January, couturier Alexis Mabille presented his Haute Couture SS26 collection at the Lido on entirely virtual mannequins, AI-generated from his own designs. His justification: "Many of my clients ask to see the pieces remotely, without coming to Paris for fittings." The AI simulates the garment on the client's actual measurements, in private. No one cried AI slop. Because AI does not enter the brand's communication. It enters the relationship with the client. The bottle law, applied to fitting.

The luxury of imperfection

At the other end of the spectrum, Hermes. The saddler has formalized an AI governance committee. Its position is crystal clear: artisanal processes are excluded from AI's scope. The 18 hours it takes to make a Birkin remain human, by strategic choice. On communications, the reality is probably more nuanced, but the message is clear: Hermes's value lies in the hand, not the algorithm. And this week, in Paris, fashion proves them right.

Fashionista (American fashion media) documents an emerging movement on the runways of the Milan and Paris Fashion Weeks: Prada displays "intentional abrasions" on deliberately aged fabrics, Bottega Veneta and Loewe center their campaigns on visible craft, Miu Miu unveils "Making of Old," a project celebrating time and wear as markers of authenticity. And this week in Paris, Pieter Mulier signed his final show for Alaia before joining Versace. His show philosophy: "Reduce, reduce." Zero accessories, zero jewelry. Just the garment on the body. The last gesture of an artistic director leaving a maison is to strip away everything except the hand. While AI floods feeds, fashion responds with restraint. The most convincing antithesis is one that is worn. And that makes money: Hermes's operating margin exceeds 40%.

This is the most counterintuitive lesson of these two weeks: in a world saturated with AI, imperfection becomes a luxury. The Japanese have a word for it: wabi-sabi, the aesthetics of the imperfect, the transient, the incomplete. Wear, time, the trace of the hand: everything an algorithm cannot reproduce. I call this the luxury of imperfection: the moment when the artisanal flaw becomes the ultimate marker of authenticity against algorithmic perfection. Luxury does not need to fight AI. It needs to make it invisible where it adds value (the bottle law), and to reclaim the hand where it creates meaning.

The threshold is about identity, not technology

The EU AI Act (European regulation on artificial intelligence) comes into force on August 2, 2026. Luxury brands are classified as "deployers"; personalized recommendations and AI clienteling fall within a scope that must be documented (Fibre2Fashion). Will we have to declare the share of AI in a perfume's formulation tomorrow? If an agent co-formulates a product, who holds the patent? Regulation will come. But it will not answer the question that matters: in my maison, what cannot be delegated to an agent?

Every maison will have to decide where it places the door. Manufacturing? Settled. The R&D lab? Shiseido shows it works. The artistic director's studio? Valentino shows the customer is not ready. The artisan's workshop? Hermes says no. The workshop threshold is that door each maison places differently according to its identity and what it promises its clients. Neither Google, nor OpenAI, nor Brussels will place it for you.

SECTION

THE STORY

From the cellar to checkout: what spirits teach luxury about agent-to-agent commerce

The context. Pernod Ricard (the world's second-largest wine and spirits group: Absolut, Jameson, Martell, Mumm, Perrier-Jouet) is not a fashion house. But its approach to AI illustrates a principle the entire luxury sector should study: you do not deploy agents on a fragile IT system.

The sequence. Everything starts with data, but not in the generic sense. What Pernod did specifically: structure its product data so it is readable by AI agents, not just by humans. In practice, the team of Gokcen Karaca (group digital and design director at Pernod Ricard) regularly queries the leading LLMs (ChatGPT, Gemini, Claude) with consumer queries about the group's brands, catalogs the responses, identifies positioning errors, then adjusts web and advertising content until the models return the right message. Iterative optimization work, brand by brand, model by model. The next step: adopting standards like llms.txt, a machine-readable format designed for AI agents to parse and prioritize a brand's product information. Cloudflare, HubSpot, and Stripe have already implemented it, with up to +25% organic traffic within two weeks. In parallel, three internal tools deployed for the sales force and marketing, not for the end consumer: D-STAR (field recommendations for B2B sales reps: which SKUs to push in which point of sale, how often to visit), Matrix AI (near-real-time marketing budget reallocation across brands and channels), and Genie (marketing content generation, in pilot). Internal adoption reached 85% across deployed markets (a threshold imposed before any new geographic deployment), with a measured impact of +1.5% to +4.5% on sales depending on markets and up to +15% marketing efficiency.

From monitoring to transaction. The HBR (Harvard Business Review) March-April 2026 case ("Preparing Your Brand for Agentic AI") reveals that Pernod is now structuring its brand data in machine-readable format. The objective: prepare for a world where a sommelier agent will converse directly with a buyer agent to recommend a whisky to a bar, without a human intervening in the transaction. Agent-to-agent commerce. Why? Because the end consumer has changed their search channel. Karaca discovered that two-thirds of Gen Z now use LLMs to research products before buying. And the models' answers were often wrong: an LLM ranked Ballantine's as a "prestige" product when it is a mainstream scotch. If the consumer asks "what cognac should I give for a birthday?", the agent's answer depends on the quality of the brand's data, not the quality of the model.

Data before agents.

The 85% adoption rate proves the bottleneck is not technological but organizational. Agent-to-agent preparation is a strategic signal: when your B2B clients (business-to-business: bars, restaurants, retailers) delegate to agents, your data must be accurate. Otherwise, the agent will recommend a competitor. And the infrastructure subtext is the same as for Shiseido: Pernod could deploy because its data was already structured. It is the data that enabled AI, not the other way around.

The compass: 3 questions

Is your product data structured to the point where an external AI agent could read and interpret it correctly? If you discover that ChatGPT mispositions your brand, the problem lies in your data, not in the model.

What is your internal adoption rate for AI tools already deployed? If the answer is "I don't know," then nobody is measuring it. Pernod will not deploy a tool in a new market below 85%.

Can your IT system dialogue with an external agent? The question of interoperability is not an IT issue. It is a question of commercial survival in a world of buyer agents.

Sources: HBR, Dec. 2025 | HBR, March 2026 | CIO, 2025

SECTION

MY INDISCREET QUESTION

In your maison, how far do you let AI enter: into the manufacturing floor, the lab, the artistic director's studio, or the artisan's workshop?

SECTION

ON MY READING LIST

Capgemini, "From Hype to How: Retail AI Trends 2026"
Only 8% of consumers would pay for an AI buying agent. 74% value humans in-store. 76% want clear rules about what the agent can do. The report's title could be this newsletter's subtitle.

WWD, "How AI Agents Are Transforming Fashion Brand Discovery"
How AI agents are rewriting fashion brand discovery: LLM biases, positioning hallucinations, and what this means for maisons that have not structured their data. Cross-reference with the Pernod Ricard case in our STORY.

DLG (Digital Luxury Group), Luxury AI Adoption Study 2026
250+ luxury executives surveyed. 71% say adoption "can no longer wait," but 55% are still in exploration mode. The number-one obstacle is not budget: it is data fragmentation (37%).

Deloitte, "The Future of Luxury During the Rise of AI"
The thesis in one sentence: "artisans and algorithms must work together." AI adoption in luxury must align with brand values. Leaders build the data platform first, then the use cases. The exact reasoning of our Pernod Ricard STORY.

WWD x CFDA, NYFW Dinner: Tech and AI's Role in Luxury Fashion
An insider dinner in New York during Fashion Week. Consensus among the artistic directors and CEOs present: "tech should amplify, not dilute, personal connections." Designer Danielle Frankel explores AI in production, Kate Barton the AI-powered digital show. The same tension as our DEEP DIVE, seen from backstage.

AI & Creativity: shotgun wedding or love match?, Mickaël Tsakiris
In June 2024, I wrote about the tension between Data and creation. The thesis held even then: AI makes creation smarter, but humans give it meaning. Eighteen months later, Gucci and Shiseido illustrate exactly the two scenarios I described. I'd like to think this piece has aged fairly well...

The Adolescence of Technology, Dario Amodei
A 20,000-word essay I recommend to anyone deploying AI agents in a professional context. This is not a CEO doing crisis communication: it is a structured analysis of the systemic risks AI poses to economies, democracies, and labor markets. Uncomfortable to read. Essential.

Full annotated text in French, Le Grand Continent
If you don't have time to read Dario Amodei's 50-page essay in English, Le Grand Continent offers an annotated French version. The editorial commentary adds genuine depth; this is not a simple translation.

SECTION

COMING NEXT

Edition #1: AI agents are searching for your brand. Edition #2: they are optimizing your warehouse. Edition #3: they are pushing through the workshop doors. Week after week, the frontier draws closer to the core. Next up: the customer, and what AI agents change in the in-store experience, clienteling, and the direct relationship. In parallel, L'Oreal will present its AI-powered formulation optimization approach at the Nvidia GTC on March 17. The bottle law is gaining new followers. And every week, with you, I will try to redefine it. Luxe oblige.

LUXE ÆTERNAI | Every Thursday morning in your inbox.

Mickaël Tsakiris