ING's retail research team published this week arguing that agentic commerce — shopping delegated to AI — will compound the footfall damage already done by e-commerce. The mechanism is simple: if an agent completes the purchase, there's no browse, no impulse buy, no reason to go near a store. Today's piece on the shopping agent's missing skill sits directly on top of this — agents are excellent at buying the thing you already decided on, and that's precisely the traffic stores needed to survive. What ING is flagging, and what most retailers aren't pricing in, is that the mid-funnel — where physical retail historically did its best work — is the first lane agents will colonize.

"Experiential retail" was supposed to answer e-commerce. It is not an answer to agentic commerce. A ghost crab doesn't walk into a Lush to smell things. Neither does an agent.

The BoF opinion floating this week is titled "Online Shopping Could Be AI's Next Victim" and the provocation is right but the frame is narrow. What dies is the intentional browser visit — the person who goes to net-a-porter.com and looks around. What survives and grows is the completed transaction. The channel collapses; the commerce doesn't. The retailers who get hurt are the ones whose entire brand-building strategy depended on someone navigating their website. That's most of them.

Julie Bornstein left Stitch Fix, founded Daydream, and has been quietly building an AI fashion shopping layer since before most retailers understood that ChatGPT existed. Forbes profiled her again this week, which typically means a fundraise just closed or is imminent. Her thesis has been consistent: a shopping agent needs taste, not just retrieval, and taste requires training on human style judgment — not product catalogs. She's solving the problem from the demand side. We covered the benchmark gap between simulator-graded and human-graded agents in May — Daydream's architecture is the most direct attempt to close that specific gap I know of. Worth watching whether today's ING forecast changes her pitch to retail partners, or whether it already did.

etailment.de published a piece this week on what fraud prevention needs to look like when the buyer is an AI agent. The standard checkout fraud stack — velocity checks, device fingerprints, behavioral biometrics — was calibrated to detect when a human is behaving anomalously. An AI agent behaves nothing like a human. It doesn't mouse around the page, doesn't hesitate at the price field, doesn't abandon carts. It also doesn't appear in your 3DS authentication flow the way a cardholder does.

This is not theoretical. It is a liability problem without a legal ruling yet. The shopping agent piece today focuses on the judgment gap; the German piece is about the identity gap. Neither is solved. Both are live inside every checkout processing agentic transactions right now — which is more checkouts than any platform is admitting.

Michael Kors launched an AI-powered retail assistant on its website this week — approximately the sixteenth time a mid-luxury brand has announced this in 2026. Missing from this announcement, as from every predecessor: a single number. Conversion rate delta. Average order value change. Session length. Return rate. Anything. We noted in May that marketers are deploying AI faster than they're measuring it. Michael Kors is exhibit seventeen.

Topshop staged an immersive catwalk show and ran it live into a TikTok Shop beauty experience. The structure is notable: the catwalk generates the content; TikTok holds the cart. The audience is a purchase funnel dressed in editorial clothes. TikTok Shop's 84% beauty surge established that closing happens in the comment thread, not the creator video above it. Topshop appears to have read that finding and built the content around the closing environment rather than the other way around. This is the format that makes sense — not livestreaming a show designed for seats, but designing the show for the stream.

StockX — the platform that exists because physical retail failed to authenticate at scale — is opening its first permanent US retail location in SoHo. This is not a contradiction; it is a statement about what physical retail is for in 2026. Not transactions. Trust verification at the top of the funnel. The store doesn't compete with the platform; it justifies it. You touch the shoes, you understand why the authentication process exists, you go home and use the app. The location choice — SoHo, not a mall — confirms this is a brand move, not a distribution one. Someone in that building made a smart decision.

France legislated against overconsumption and then set a new clothing purchase record in the same breath. Neritus Vale's piece today on France's overconsumption law and the record it produced traces the mechanism. The short version: legislation that raises awareness of overconsumption can function as advertising for the thing it's trying to stop. The psychological literature on this is established. The fashion policy literature has not caught up. If France wants to move 3.6 billion, the instrument is price. The record says so.

CHAMP — Champion Athlete Managing Partner fund, a PE-athlete vehicle launched in April 2026 — invested in Rhoback, an American performance menswear brand. The same week, today's piece on Phlur hiring Olympian Suni Lee traces what it signals when a fragrance brand reaches for an athlete at that specific moment in its lifecycle. These are two versions of the same structural shift: athletes aren't endorsing products anymore, they're capitalized into them, and the brands that attract that capital are making a fundamentally different bet than brands still buying 30-second spots.

Rhoback is interesting because it's men's performance — a category where brand identity has historically been harder to build and easier to lose to Lululemon. CHAMP is betting that athlete ownership changes the authenticity calculus for male buyers in a way sponsorship never could.

亿邦动力网 reports that Pinduoduo is tightening AI content governance rules and lifting compliance thresholds for e-commerce listings. The framing is "content quality"; the mechanism is AI-generated copy detection — listings that read as generated get flagged, downranked, or pulled. The timing matters. This arrives as Pinduoduo faces more competitive scrutiny (see below) and as every platform floods with AI-produced product descriptions. Raising the compliance bar makes the catalog appear more curated without doing curation. It also raises costs for smaller merchants who were using generative tools to compete on listing volume. They'll absorb it or leave. Both outcomes serve Pinduoduo.

证券时报 ran a characteristically blunt headline this week: Taotian's "dual engine" has lost speed. JD is headhunting Alibaba talent. Pinduoduo is pricing through the floor. Douyin is capturing the impulse purchase. The ¥380 billion AI infrastructure commitment looks less like a growth plan and more like a defensive perimeter. We called this dynamic in May: Alibaba is buying time, not growth. The new signal here is that Taotian's growth rate is now trailing the market — not just trailing its own historical pace. That's a different kind of problem. One buys you time to pivot. The other means the pivot is already late.

ANDAM handed its 2026 Innovation Prize to Alphalyr, a French analytics AI that plugs into fashion retail performance data. This is the quiet signal the French fashion establishment sends when it decides something is legitimate. ANDAM's prior innovation prizes went to sustainability tech and materials innovation. 2026 is analytics AI. What changes with this prize is not the technology — Alphalyr has been operating for years — but the social permission for French houses to use it without apologizing for it.

FashionUnited's French framing: "AI montée dans la chaîne de valeur mode" — AI rising in the fashion value chain. The prize is being read in Paris not as a technology endorsement but as a fashion-industry endorsement. That's different, and it matters for which platforms get to pitch Chanel's buying team in 2027.

Ulta Beauty, Tarte, and Beekman 1802 executives went on stage at Glossy's E-Commerce Summit to discuss AI implementation in their workflows. What came back was unanimous: AI saves hours on reporting, briefing, and content generation. What none of them said was what that did to revenue. This is the pattern across every practitioner conversation right now. Efficiency gains are real and documented. The revenue link is asserted but unmeasured. The gap between "we saved forty hours a month on creative briefs" and "this improved our topline" is where every AI ROI conversation goes to die.

Connect this to today's piece on Saie's classroom initiative: one brand investing in education because paid reach stopped converting; three brands investing in AI because workflows got expensive. Different problems. Similar budget lines. The conflation is going to cause accounting pain inside these companies within eighteen months.

An Arts University Plymouth graduate showed foraged, woven pieces at New Designers 2026 that shift in response to touch, light, and movement. Interesting not as a trend item but as a calibration point: if every AI commerce conversation this week is about agents taking the mid-funnel, here is a product category the mid-funnel structurally cannot describe. You cannot write a product page for something that changes with the viewer. The catalog metadata problem we flagged in May — try-on models ready, catalogues not — is more acute here. The bottleneck for this category isn't agents. It's language. Nobody has built a vocabulary for this yet, and until they do, the shopping agent will route around it entirely.