Sephora is wiring its product catalog, loyalty data, and beauty advisor logic into Google's AI shopping ecosystem. The play is obvious: be the first prestige answer when someone asks a chatbot what to buy. The irony is also obvious: Sephora's moat for twenty years was the human beauty advisor, the sampling counter, the store as experience. Now it's exporting that expertise to a surface it doesn't control, can't differentiate on, and will share with every other brand that signs the same Google deal.

We said it in May: when everyone pitches curation, nobody has a moat. Sephora is betting the brand name survives the commodification. That bet might be right. Watch what happens when Ulta signs the same agreement in Q3.

Etailment — the German trade publication that covers e-commerce with more rigor than most English-language outlets — asks the question the agentic commerce hype cycle is actively avoiding: when an AI agent completes a checkout, how does fraud prevention distinguish between a legitimate authorized agent, a stolen credential operating autonomously, and an outright attacker? Current fraud stacks were built to verify a human at the keyboard. The trilemma is not hypothetical; it's live anywhere an AI agent has been handed a payment method.

Everyone's publishing AI shopping guides this week — retailers have been shipping these apps faster than the benchmarks can evaluate them — and this is the first piece I've seen that takes the fraud surface seriously. The payment networks have published nothing on agentic authorization. Someone is going to have a very expensive quarter when that gap makes the news.

Pinduoduo is tightening AI content governance across its marketplace, raising the compliance threshold for e-commerce operators. This is not altruism — it's a quality-signaling move against competitors. If your AI-generated product copy doesn't clear the filter, you don't surface in results. The effect is to make AI content compliance a cost of platform entry rather than a differentiator.

The cross-border implication is real. What Pinduoduo tests in Shenzhen, Temu deploys globally. Platform crackdowns in Southeast Asia suggest the enforcement pressure is traveling, and AI-generated listings that pass today's bar may not pass next year's anywhere.

Securities Times breaks down Alibaba's dual-engine slowdown in language considerably more direct than the company's investor relations materials: JD is poaching engineers, Pinduoduo is bulldozing on price, and Douyin has eaten the product-discovery layer that used to belong to Taobao search. The ¥380 billion AI infrastructure pledge is the frame; the Chinese financial press is reading it as defensive capex.

We've been tracking this since May. Today's piece from Sir John Crabstone — Alibaba Owns a Marketplace. Baidu Is Betting the Answer Is Enough. — closes the loop: if AI search replaces browse-and-discover, Alibaba's storefront advantage is precisely the thing that evaporates. The company is spending ¥380 billion to defend a position that may not exist in five years.

FashionUnited profiles Hillary Super, who walked into Victoria's Secret with activists circling the board and turned it around before the shareholder vote, by nine days. The playbook wasn't complicated — pick a lane, execute without pivoting, say no to more things than you say yes to — but the execution clearly was. Super's thesis: less logo-forward, more craft, different rather than louder. Worth reading in full as a counterweight to the many retail turnaround stories that don't turn around. The ones that work usually involve someone who actually stopped doing things.

FashionUnited rounds up brands actually implementing the EU's Digital Product Passport requirement, and the consistent message is: the regulation is live, the timelines are tight, and the data infrastructure needed to make a DPP meaningful doesn't exist at most companies. Several brands are shipping a QR code that links to a PDF and calling it compliance. That is not the point. The point is traceability — fiber origin, dye chemistry, labor provenance — and most supply chains have never had to produce that information because nobody asked.

The OLAF counterfeit ring from May rhymes here: the smugglers filed clean transit documents while the cargo moved elsewhere. A DPP built on unverified self-reporting is the same problem with a better QR code.

Zalando is launching a dedicated UK shopping app. The timing is the point: this follows the ABOUT YOU acquisition, which was about consolidating European fashion e-commerce scale. We called the Zalando strategy "scale over protocol" in May — absorb the competitor, then expand the footprint. A UK mobile app is what that strategy looks like in the next chapter. The UK fashion e-commerce market has lost confidence since ASOS restructured and Boohoo stumbled; Zalando is reading that gap as an opening.

easyGroup — the British conglomerate behind easyJet, easyCar, easyHotel, and a graveyard of brand extensions — is launching easyShop, an online marketplace. The brand logic is that "easy" connotes value and accessibility, and those are desirable things. The problem: easyJet works because air travel is almost perfectly commodified and price is the only real lever. General merchandise e-commerce with Amazon as the baseline is not that market, and brand halo has never been sufficient to build loyalty in categories where someone else already owns fulfillment expectations. File under: we'll see, but the prior base rate for easyGroup horizontal expansion is not encouraging.

Elliott Advisors — the fund that does not acquire assets to leave them as found — has reportedly joined the bidding for The Very Group, operator of Very.co.uk and Littlewoods. The actual moat here is the consumer credit business: Very has survived the UK digital retail shakeout partly because it is also a lender, providing buy-now-pay-later infrastructure that predates the BNPL fintech wave by decades. Elliott will have strong views on how that asset is priced and what it should be doing. UK digital retail has been brutal: ASOS restructured, Boohoo restructured, Missguided failed. Very's survival has depended on not being purely retail. Watch what happens to that when PE money arrives with a "focus on core" mandate.

Michael Kors has deployed an AI retail assistant on its website. The press release offers almost no specifics about what it does, which tells you where we are in the cycle: "AI retail assistant" is still a category claim, not a capability description. Every brand deploys one now. The meaningful data arrives when someone publishes conversion lift, session depth, or return rate against a control group. The evaluation frameworks still don't exist — we said that in May, still true in June, will probably still be true in September.

Topshop — revived as a digital-only brand after the Arcadia collapse — ran an immersive catwalk show paired with a TikTok Live beauty shopping experience, letting viewers buy in real time from the comment layer. The format is legitimately interesting: runway show as live commerce event, the transaction inside the same stream as the spectacle. We noted in May that TikTok's 84% beauty surge is being closed in the comment thread, not the video above it. Topshop built a runway show around exactly that mechanic.

The open question is whether Topshop has enough residual brand equity to make the conversion numbers work, or whether this is a smart format wearing a thin suit. But if it converts, you will see every fashion week presentation chasing this structure within two seasons. Connection worth noting: Sephora goes into Google's AI ecosystem for text-based discovery; Topshop goes into TikTok Live for visual, real-time discovery. Prestige retail is splitting across two radically different models simultaneously, and brands large enough to fund both are playing a different game than everyone else.

BoF runs an opinion piece arguing that AI agents doing your shopping will kill online retail as a behavior — no browsing, no impulse purchase, no discovery, just silent errand-running. The argument is partially right: if every purchase is delegated to an agent optimizing against last week's preferences, the serendipity layer disappears and a lot of retail margin disappears with it.

But today's piece from Parallax Pincer — Daydream Bet Taste Was the Last Thing AI Shopping Hadn't Solved — is the structural counterargument. What Julie Bornstein built is discovery-native AI: an assistant that cultivates taste rather than fulfilling a shopping list. That is a fundamentally different product from "buy my usual stuff automatically." Whether it works is a separate question from whether those are the same thing, and conflating them is the BoF piece's main error.

And then there's Neritus Vale's piece today on Quince hitting $10 billion on no-brand-at-all. Three live experiments — discovery AI, errand AI, no AI — running simultaneously. Whoever tells you they know which one wins is selling something.

Kantar's 2026 fashion report gives a name to what Kearney was describing in May: consumers in a tight economy trading down on staples, up on treats. "Treatonomics" is the coinage. It is a consulting word and it will appear in every earnings call before December. The more interesting argument in the Kantar report is about retail media: Kantar's thesis is that retail media networks win the AI-shopping moment because they sit at the point of purchase. That's a real structural advantage — assuming the point of purchase remains somewhere a brand can buy into, and not something an AI agent handles silently before the retailer's interface ever loads.