News + Thought
Who decides what you desire? Luxury and the rise of agentic commerce
There’s a version of this article I could have written six months ago, and it would already be wrong. Back then, the story everyone was telling about agentic commerce was simple: AI assistants would soon do your shopping for you, end to end.
Find the bag, judge the fit, complete the purchase, all without you leaving the chat. The checkout was supposed to disappear into the conversation.
Then the most-watched experiment in the category fell apart in public. Walmart spent several months from late 2025 testing OpenAI’s Instant Checkout inside ChatGPT, and in early March, its AI lead told an investor conference that purchases completed in the chat converted at roughly a third of the rate Walmart sees on its own site.
Carts were populated with the wrong items; there was no working system for US sales tax. Within weeks, OpenAI had retired Instant Checkout altogether and repositioned its protocol around product discovery, letting retailers handle their own checkout. Walmart’s response was telling: it embedded its own assistant, Sparky, inside ChatGPT instead and early figures put conversion back up around 70% of its website rate, simply by changing who controlled the transaction.
Looked at from the systems level, that failure was predictable and the lesson inside it is the most useful thing a brand can carry into the next two years.
Discovery and transaction are splitting apart, and they are not splitting evenly.
It was never about the checkout
It’s tempting to read the Instant Checkout retreat as “agentic commerce was overhyped,” but that’s the wrong conclusion. Discovery through AI surfaces is real and growing quickly; it’s the transaction layer that proved harder than the AI platforms assumed. Checkout, inventory accuracy, tax, fraud, returns – these are competencies retailers spent decades building, and they don’t transfer to a language model just because the conversation is fluent.
So the part that’s really moving is further upstream, and for luxury, it’s the part that matters most. For decades, a luxury brand controlled the first impression. The boutique, the website, the window, the client stepped into a world the brand had designed, and desire was shaped inside that world. McKinsey frames the change well: as AI systems increasingly interpret, compare and curate on a shopper’s behalf, decision-making shifts upstream, from where a product is bought to where intent is first expressed. If a meaningful share of early shopping behaviour now happens inside a general-purpose assistant, then the first interpretation of a customer’s desire is occurring somewhere the brand doesn’t own, can’t see, and didn’t design.
For luxury leaders, the sale was never really the contested ground. The contested ground is interpretation: who reads the customer’s intent, and shapes it, before that customer ever reaches you.
The plumbing, briefly
It’s worth understanding the infrastructure underneath, because it explains why no single brand strategy will hold.
There are, in effect, three layers being standardised at once. There’s:
- a discovery and commerce-journey layer, Google and Shopify’s Universal Commerce Protocol (UCP), launched in January 2026 with backing from twenty-plus retailers and payment networks.
- a transaction layer, OpenAI and Stripe’s Agentic Commerce Protocol (ACP), now refocused on discovery after the Instant Checkout retreat.
- And underneath both sits a connective layer, the Model Context Protocol (MCP), originally from Anthropic, which is simply how an AI agent plugs into your systems and reads your data.
I won’t pretend to know which of these “wins,” and I’d be sceptical of anyone who claims certainty this early. We are roughly at the stage the web was at before anyone had agreed on what the padlock icon meant. What I can tell you is that they are not really competitors at the level that matters to you. They’re distribution channels for the same underlying asset: your product data. One analysis found that merchants implementing both UCP and ACP captured materially more agentic traffic than single-protocol stores. Platforms like Shopify now switch many of these channels on by default; newer entrants such as Swap take the opposite bet, giving a brand its own conversational storefront on a dedicated site so it keeps the journey and the data. Both bets are reasonable. Picking one protocol and declaring victory is not.
Why luxury is exposed differently
A comparison engine is, by design, a flattening device. It strips a category down to attributes it can rank: material, price, dimensions, reviews, and surfaces a shortlist. For a mass-market product that’s mostly helpful, spec and price are much of the story. For luxury, it’s a threat to the core proposition, because the things that justify the price discretion, scarcity, heritage, the high-touch service around the object are precisely the things an attribute-matching agent discards. If your £4,000 coat is reduced to “wool overcoat, beige, 200 GSM” and ranked beside a £400 one, the agent has quietly redefined your category on terms that erase your value.
The same technology that flattens can also amplify, if you feed it well. Brands that supply rich, accurate, structured product data, provenance, craft, materials, care, and the actual story give the agent something truer to surface than a scraped spec sheet. Done properly, agentic discovery becomes a way to scale adviser-grade guidance to a generation of clients who’ll never set foot in the flagship first. Kering’s KNXT and its assistant Madeline, or Zalando’s conversational styling tools are early attempts at exactly this: the personal-shopper relationship, rebuilt for a chat interface.
The caveat: Measurement
There is one unsolved problem: attribution in agentic commerce is genuinely broken. When discovery happens inside an assistant the brand can’t see, and the sale completes days later on the brand’s own site, last-click measurement quietly miscredits the whole journey. AI-referred traffic is rising fast while converting, on reported figures, well below traditional affiliate channels, and part of that gap is almost certainly mismeasurement rather than poor performance. Anyone who tells you they’ve fully solved AI commerce attribution today is selling something.
A few calls I’m willing to make
Predictions are cheap, so let me be specific enough to be wrong.
First: in-chat checkout returns, but not soon and not universally. Once agent payment and identity infrastructure mature, probably 18 to 24 months out, and even then, I’d expect luxury to keep checkout on owned property far longer than mass retail.
Second: multi-protocol support becomes table stakes, not an edge, by the end of 2026.
Third: the discovery surfaces will not stay purely “merit-based”, paid placement is coming to AI shopping, because the economics demand it, and brands should plan for a world where visibility is partly bought.
What I’d do now
If you take one thing from this, make it this: the work that compounds across every protocol, platform and prediction above is the same single thing, your product data. So, concretely:
- Get your catalogue agent-ready first. Factual, structured, complete attributes beat marketing copy in an agent’s eyes, every time, and for luxury, that means encoding provenance and craft as data, not just prose.
- Adopt a multi-protocol posture rather than betting on one platform.
- Keep checkout on property you control while investing seriously in your discovery presence on the AI surfaces.
- Decide deliberately who owns your conversational layer, a platform default or your own agent, rather than letting it be decided for you.
- Start building agentic-aware measurement now, before the gap between what you see and what’s actually happening grows any wider.
Frequently Asked Questions
What is agentic commerce?
Agentic commerce is a model of online shopping in which AI agents act on a shopper’s behalf across the buying journey, interpreting intent, researching and comparing products, surfacing recommendations and, in some cases, completing the purchase. In practice today, the discovery part of that journey is moving into AI assistants quickly, while the transaction still tends to be completed on the brand’s own site.
What is the difference between UCP, ACP and MCP?
They operate at different layers. The Universal Commerce Protocol (UCP), co-developed by Google and Shopify, covers the wider commerce journey from discovery through post-purchase. The Agentic Commerce Protocol (ACP), from OpenAI and Stripe, began as a checkout standard and is now refocused on product discovery. The Model Context Protocol (MCP), originally from Anthropic, is the general-purpose connective layer that lets an AI agent plug into a merchant’s systems and read its data. For most brands, they are complementary distribution channels for the same product data, not a single choice to be made.
Did in-chat checkout fail?
The first major implementation stumbled. Walmart tested OpenAI’s Instant Checkout inside ChatGPT and reported it converted at roughly a third of its own website’s rate, with accuracy and tax-handling problems. OpenAI retired Instant Checkout in March 2026 and repositioned around discovery, while Walmart embedded its own assistant and recovered conversion to around 70% of its site rate. The takeaway is not that agentic commerce failed, but that discovery and transaction are separating, and brands convert best when they keep control of the checkout.
How does agentic commerce affect luxury brands specifically?
It changes who interprets the customer’s desire first. Comparison-driven agents tend to flatten a category down to rankable attributes like material and price, exactly the dimensions that strip out the discretion, scarcity, heritage and service that justify a luxury price. The risk is being ranked beside cheaper goods on terms that erase your value. The opportunity is the reverse: brands that feed agents rich, structured data about provenance and craft can scale adviser-grade guidance rather than be reduced to a spec sheet.
What should brands do to prepare for AI-driven shopping?
Start with product data, because it is the asset that compounds across every platform and protocol: make catalogue attributes factual, structured and complete, and for luxury, encode provenance and craft as data rather than only as prose. Beyond that, adopt a multi-protocol posture instead of betting on one platform, keep checkout on owned property while investing in discovery presence on AI surfaces, decide deliberately who owns your conversational layer, and build agentic-aware measurement early.
Will AI shopping recommendations stay free of paid placement?
It is unlikely to stay purely merit-based. The economics of the discovery surfaces point toward paid placement entering AI shopping over time, much as it did in traditional search. Brands should plan for a future in which visibility within AI recommendations is partly earned through data quality and partly bought.
How agent-ready is your brand?
Skywire helps luxury and lifestyle brands stay visible and keep selling as discovery moves into AI. If you’d like a read on how agent-ready your catalogue and commerce stack actually are, get in touch.
Speak to a specialist
