Sephora has integrated its product catalog, reviews, and advisor data into Google's AI shopping ecosystem — meaning when Google AI Overviews, Lens, or Shopping Graph surfaces a beauty recommendation, Sephora's infrastructure is upstream of the answer. The Glossy framing is that this extends Sephora's "beauty advisor" model into AI. That's the press release version. The real version: Sephora is paying to be the default answer in AI-mediated beauty discovery, the same way it once paid for premium gondola placement at department stores.

We noted in May that every prestige beauty retailer had converged on the same curation pitch. What differentiates them now isn't positioning — it's whose data is upstream of the AI. Ulta's next move will be worth watching. Compare to John Lewis, which hedged between ChatGPT and Google integrations from the same budget. Sephora picked a lane. And note: this announcement lands the same week Pinterest is hiring a Chief Shopping Officer (more below). The commerce layer of AI is being built in real time and the race to be the preferred data source has already started.

JD.com is reportedly willing to pay around £2 billion for Very Group, the UK retailer behind Very.co.uk and Littlewoods, which has roughly 4 million active customers and a significant buy-now-pay-later book. This would be JD's most consequential Western bet since its Ochama dark store experiment in the Netherlands. Very isn't prestige — it's a mass-market, credit-driven British fashion and homeware retailer with deep catalog depth and a loyal older customer. That is exactly the demographic JD needs if it wants UK mass-market legitimacy rather than just a European logistics outpost.

The BNPL book is either the prize or the liability depending on your view of UK consumer credit conditions. And consider the context: Zalando absorbed ABOUT YOU for €1.13 billion to consolidate European fashion. JD buying Very would be China acquiring European payment infrastructure as much as retail reach — a very different kind of transaction than it appears.

Also worth connecting: Chinese financial press today reports that JD is specifically poaching AI and recommendation engineering talent from Alibaba (see below). A JD that is talent-rich, holds a UK mass-market platform, and is scaling European logistics is a different animal than the JD of two years ago.

Pascal Wördehoff, writing in German trade outlet etailment, asks the question nobody in agentic commerce wants to answer: when an AI agent completes a purchase on your behalf, every fraud detection system on the internet sees a bot. Behavioral biometrics — typing cadence, mouse movement, device fingerprinting — are calibrated on human signals. A programmatic agent buying at machine speed looks like an attack. Chargebacks, liability, identity verification — none of this infrastructure was designed for authorized non-human buyers.

This is the unsexy underside of every agentic shopping demo you've watched this year. We noted in May that retailers shipped AI shopping apps faster than researchers could build evaluation frameworks. The fraud credentialing gap is the same failure mode: build the frontend, inherit the backend consequences later. The irony is that Sephora's new Google AI integration (above) is precisely the kind of agentic-adjacent infrastructure that makes this problem more acute. When AI agents are completing beauty purchases on behalf of users, who authenticated the agent in the first place?

Someone — Shopify, Stripe, some identity infrastructure layer — is going to have to ship agent authentication tokens. Whoever does it first owns a piece of commerce plumbing that every retailer on the planet will need.

Chinese retail intelligence outlet 亿邦动力 reports that AI e-commerce shopping entry points have now "taken shape" (成型) and that platform capability is driving end-to-end transaction closed loops. Translation: the experiment phase is over. On Chinese platforms, AI is no longer a discovery feature or a chatbot — it is a checkout pipeline. The piece frames this as infrastructure maturity, not individual feature launches.

This matters globally because China is running twelve to eighteen months ahead of Western markets on agentic commerce deployment. While Western retailers are still shipping apps faster than they can be evaluated, Chinese platforms are reporting closed loops. Also: today Sir John Crabstone is publishing on Pinduoduo's voluntary AI content governance — a platform that has closed the transaction loop is now also taking responsibility for what runs through it. Those two moves together look less like product maturity and more like preparation for regulatory scrutiny.

Chinese financial press is asking whether Alibaba has fallen behind: the Securities Times documents Taobao/Tmall's "dual engine" (content + search) losing momentum while JD poaches AI engineering talent, Pinduoduo's pricing grinds, and Douyin captures impulse spend. We covered this dynamic in May — the ¥380 billion AI pledge is defensive capex, not a bold bet. Today's update is the talent angle: JD is specifically targeting Alibaba's recommendation and AI engineering teams.

You can spend ¥380 billion on infrastructure and still lose if the people who know how to run it have already left. Watch for retention packages or an Alibaba org announcement in the next two quarters. And yes, a JD that's talent-rich and reportedly also pricing Very Group in the UK (above) is a company that has found its second wind.

Meta and Reliance Industries have announced an AI-enabled data center partnership in India. The press release is about compute capacity. What it's actually about: Meta buying proximity to Reliance's commerce and telecoms infrastructure. Reliance runs JioMart, one of India's largest e-commerce operations, and Jio's network has around 450 million subscribers. Instagram Shopping's next large-scale market is India, and Reliance has the merchant relationships and last-mile logistics that Meta cannot build from scratch.

The counter-read: this is defensive compute, not commerce strategy — Meta needs Indian infrastructure for its own AI models regardless of shopping ambitions. But Reliance doesn't sign data center deals for compute alone. There's a commerce layer coming in the fine print. Watch Chinese platform infrastructure spending in parallel: if Meta and Reliance build out AI commerce in India before Pinduoduo's international expansion reaches the subcontinent, that is a meaningful competitive wall.

Modern Retail reports that every retail media network, agency, and tech platform is showing up in force for the World Cup. The framing: this is the moment retail media proves it can monetize live sports attention the way linear TV did for decades. My skepticism is specific. Retail media's actual edge is purchase-intent proximity — it works because the customer is already in a shopping mindset. World Cup viewers are in a sports mindset. You can absolutely convert kit and jersey impulse buys. But if retail media is selling brand awareness CPMs against live sports, it is competing on linear TV's terms, in linear TV's arena, without linear TV's reach. That is a bad structural bet.

The brands that will learn something useful are the ones who wire a direct commerce path underneath the activation and measure the conversion gap honestly. Everyone else is buying a logo in the corner of a screen and calling it performance.

Glossy reported from its E-Commerce Summit, where brands including Ulta Beauty, Supergoop!, SharkNinja, and Tarte discussed how they're training employees on AI tools. This is the less glamorous, more important story in retail AI: not the vendor announcements, not the agentic demos — but whether the person running digital merchandising or sitting at the beauty counter actually knows how to use the tools the company purchased. The gap between "we bought AI" and "our team uses AI" is where most enterprise retail transformation quietly dies.

The May data on marketers buying AI but not trusting it is still the right frame. Training programs are the attempt to close that gap. Whether they work depends entirely on whether the training is about the tools or about the actual job the tools are supposed to help with. Those are very different curricula.

Pinterest has named its first-ever Chief Shopping Officer with an explicit mandate to move from "beautiful playground" to "commerce destination." We covered Pinterest's luxury pitch in May — the platform's value is appearing when taste is still forming, before the buyer has a question. That is genuinely defensible and probably undermonetized. The risk of a hard commerce mandate is that it collapses the very tension that makes Pinterest useful. The moment Pinterest optimizes aggressively for conversion, it starts looking like every other shopping surface, and the people who came to look at kitchens start leaving for TikTok.

The hire suggests Pinterest has decided the ambient-discovery model isn't growing fast enough. That's probably true. The question is whether "commerce destination" is the right alternative or just trades one platform's audience for another platform's playbook.

Michael Kors has added an AI-powered retail assistant to its website. No vendor named, no performance data, no specifics about what it can actually do. This is the fifth or sixth "AI retail assistant" launch from a mid-tier fashion brand I've logged in two months and they are starting to blur together. The question worth asking every single time: is this assistant trained on the brand's actual inventory, sizing logic, and styling logic — or is it a generic fashion chatbot wearing the brand's logo? The difference between those two things is everything, and the FashionUnited piece does not answer it.

Ace Hardware scoped Hey ARMA to 2,300 of its 5,200+ stores because it understood that advice quality degrades without domain specificity. Michael Kors is a global brand with a coherent aesthetic and a contained SKU universe. It has no excuse for not going deep on its own catalog.

Topshop — now a brand without stores, operating as wholesale and DTC IP under ASOS's ownership — paired an immersive catwalk event with a simultaneous TikTok live beauty shopping experience. This is the right architecture for a brand in Topshop's position: you cannot afford bricks-and-mortar, but you can engineer a cultural event that converts instantly. TikTok Shop's beauty surge was 84% YoY and the closing happens in the comment thread, not the video — Topshop is building on the correct mechanism.

The open question is whether "Topshop" carries enough cultural weight with the TikTok audience that never saw the original Oxford Circus flagship. Nostalgia is a Millennial currency. TikTok is not.

A viral dupe conversation around a $660 Guerlain fragrance finally pushed the maison into its first-ever paid influencer campaign. Two things to hold simultaneously: (1) the dupe drove the campaign — meaning the threat of imitation is now the primary marketing trigger for heritage perfumers, and (2) the campaign was reactive, not designed. For a house that has been making perfume since 1828, that should sting a little.

This connects structurally to today's piece from Sir John Crabstone on Pentland — old capital that answers to nobody has its own timeline for marketing experiments. Guerlain isn't family capital (it's LVMH), but the 198-year hesitation on paid influencer spend had the same logic: we don't need to do this until we need to do this. The dupe economy just told them they need to.

Also: our May piece on luxury and the customer who can read the bag applies directly here. Guerlain's customer can distinguish the original from a $30 dupe — but she also apparently needs to be reminded that the original exists. That is a new kind of problem for a 198-year-old brand, and "first ever paid influencer campaign" is a strange sentence to write about a house that invented L'Heure Bleue.

Marks & Spencer has signed a franchise agreement with PT Mitra Adiperkasa Tbk to expand into the Philippines. Mitra Adiperkasa is Southeast Asia's dominant luxury retail franchise operator — it runs Zara, H&M, Sephora, and others across the region. M&S is licensing its brand to the infrastructure operator who already knows these markets. The Gulf ran this playbook years ago: franchise agreements let British and European brands appear global without building global logistics. It limits upside and hands margin to the operator, but it eliminates the execution risk that has burned overseas retail before. Smart, if unambitious.