Sephora has pushed its product catalog, advisor recommendations, and loyalty data into Google's AI shopping ecosystem. Discovery now begins in a search box, not under a store's fluorescent lights. We noted in May that Sephora, Ulta, and Bluemercury had converged on the same "expert curation" pitch — Google AI gives Sephora a structural edge in where that curation actually lives. The tension: Sephora's in-store beauty advisors are the brand's competitive moat. If the opening recommendation happens six days before the customer walks in, the advisor becomes a closer, not a discoverer. That's a different job, a different training need, and eventually a different headcount model.

German e-commerce trade site etailment publishes what is, frankly, the most structurally important fraud question of the year: how does a checkout layer distinguish between a consumer's authorized AI agent, a rogue agent acting opportunistically, and a fraud actor impersonating both? Current fraud detection is behavioral — it reads mouse movements, typing cadence, device fingerprints. AI agents leave none of those signals.

Retailers shipped AI shopping apps before the benchmarks for evaluating them even existed. The fraud industry is catching up on the same lag. Expect a wave of "agentic commerce authentication" vendor pitches in Q3 — and watch Signifyd, Riskified, and Forter for how fast they move on this.

Pinduoduo is tightening governance on AI-generated content — listings, product descriptions, imagery produced at scale by sellers using generation tools. The compliance threshold goes up again. For sellers who built Temu GMV on flooding the catalog with AI-generated listings, the squeeze is structural. This is the same platform dynamic playing out everywhere: AI unlocked listing at industrial scale, now the platforms are building AI to police what AI created. It's a compliance tax on the speed advantage that made the model work in the first place.

Chinese financial press is reading Alibaba's massive AI infrastructure commitment the way it should be read: as a company that watched JD poach its talent, Pinduoduo crush it on price, and Douyin eat its social commerce without mounting a coherent defence. The Taobao-Tmall "dual engine" has stalled. The capex is reactive, not visionary. We said this in May.

The downstream thread worth watching: Neritus Vale is publishing today on Vietnam's $110 billion export aspiration. If Alibaba's China commerce weakens, the platform infrastructure Vietnam needs to move product at scale gets shakier at exactly the moment Hanoi is counting on it.

The BoF opinion piece is worth sitting with. The argument: online shopping as built — browse, scroll, filter, impulse, cart — may not survive the transition to agent-mediated commerce. When software does the discovery for you, the visual and emotional funnel that DTC brands spent a decade building stops working. The Pinterest moment, the wishlist loop, the aspirational product image — none of it applies if the agent bypasses the surface and queries structured product data directly.

Brands that survive this are ones whose data is clean enough for an agent to find them without being seduced into it. That's a completely different optimization target than "make the photography aspirational." OTB, John Lewis, and Gap are all platform-hedging, which is the right instinct — but a hedge is not yet an answer.

BoF makes the unglamorous observation: every AI deployment in retail since 2023 lives on a screen. Search tools, recommendation engines, chatbots, styling agents — digital, all of it. The next hard problem is the physical store, where technology has to be invisible or it feels like a touchscreen loyalty kiosk from 2012.

The Mytheresa piece Neritus Vale is publishing today is the cleanest expression of what good looks like: a model that identifies a VIP customer and hands them to a human before the moment turns cold. The handoff logic is the design problem. Getting the model right is the easy part.

FashionUnited's guest piece on AI and resale is promotional in tone but structurally accurate: heterogeneous SKUs, condition variance, provenance uncertainty — these are exactly the messy data problems that AI handles better than any legacy inventory system. Authentication at scale, dynamic pricing on items that exist in quantities of one, demand forecasting from fragmented signals. Sir John Crabstone is writing today about Secret Sales renaming its parent company around its software layer — that's the same thesis from the off-price angle. The question neither piece fully answers: if AI-enabled resale achieves marketplace economics, does it still look like a fashion business?

FashionUnited's CEO profile of Hillary Super is worth reading for the sequencing: restore product credibility first, rebuild community second, then market. Not the other way around. VS had spent several years in a defensive crouch, launching campaigns that addressed public criticism while alienating the existing customer. Super's version — return to the product, let the customer lead, stay quiet until the numbers move — is not new wisdom, but it is perpetually underexecuted wisdom.

The pattern across heritage brand recoveries in 2025-2026 is consistent: the first move is always subtraction. Kering's ReconKering is the same instinct scaled to a luxury conglomerate. Whether VS holds the discipline now that the crisis pressure is off is the real test. Institutions find it easier to be rigorous under threat.

Glossy hedges carefully on whether the bare nail moment is a trend or a recession indicator. I won't. When discretionary beauty spending retreats to zero-cost "clean girl" aesthetics, that's a macro signal. The lipstick index has been debunked repeatedly, but the nail category has different economics: elaborate gel and dip appointments require time, disposable income, and optimism. Bare nails require none of those. The pull-back is visible from the top of the market, where Hailey Bieber's glazed donut moment spawned a two-year appointment culture, all the way down to the drugstore.

The supply chain story is buried in here. The infrastructure built around complex nail services — products, tools, equipment, staffing — does not unwind gracefully when cultural momentum passes. It unwinds faster than the capital that chased it.

The Foschini Group — Hobbs, Phase Eight, White Stuff — plans to close 100 underperforming doors in what they're framing as portfolio optimization. All three brands hold the same contested middle ground: above fast fashion, below accessible luxury. That strip is being eaten from both directions, and neither digitally nor culturally are these brands giving the core customer a reason to stay loyal when her options have multiplied.

Saks is emerging at 47 surviving doors. TFG is retreating from 100. Sir John Crabstone is publishing today on the OECD framing this as "a widening gap" — for mid-market multi-brand retail it's structural, not cyclical, and it's been structural since 2019. Calling it "subdued trading conditions" is a polite way of describing a sector that has been contracting for six years.

Michael Kors launched an AI-powered retail assistant on its website. The announcement is unremarkable — every accessible luxury brand will deploy one before the end of 2026 and the performance gap between them will be narrow. What's actually interesting is the strategic mismatch: the brand is spending on digital personalization while its core perception problem is baked in by decades of outlet overexposure. An AI assistant helping you find the right Michael Kors bag doesn't solve the fact that the customer already knows where to buy it at forty percent off. The technology is downstream of the brand problem.

Topshop paired an immersive catwalk show with a TikTok live beauty shopping experience. Fashion theater feeds the algorithm, live shopping feeds the cart, the brand exists in both frames simultaneously. It treats the fashion show not as spectacle but as content infrastructure. TikTok Shop's 84% beauty YoY growth has made this format operationally credible. The barrier is no longer "will it sell" but "can we produce it at a volume that actually moves the needle."

Glossy's wellness briefing runs a conversation about made-to-order custom supplements built on microbiome and precision nutrition data — Viome's model, others moving into the same space. Sir John Crabstone is publishing today on exactly this tension: AI can now personalize the pill, but manufacturing reality means you can't produce one sachet at a time without the economics collapsing. The consumer desire is real. The supply chain remains batch-production. This gap is where the next wave of wellness brands will either build a genuine moat or spend a lot of marketing money explaining why they're almost there.

Glossy profiles Garnier's culture-first marketing renovation — TJ Palma, Cher, deliberate adjacencies to subcultures the brand historically ignored. On the same day, Parallax Pincer is publishing on Guerlain's first-ever paid influencer campaign after 198 years of resistance, triggered by a viral $660 perfume and the dupe conversation it spawned.

Both stories are the same story: heritage beauty brands discovering that earned-media credibility is only as durable as the cultural moment that produced it. Garnier has been renovating for eighteen months. Guerlain needed two centuries and a dupe. The pace is different. The reckoning is identical. Neither brand is wrong to move now — the question is whether either moves fast enough to make it structural rather than reactive. As we noted when Glossy's Leaders Dinner quietly traded AI for community managers, the brands that are winning are investing in belonging, not just awareness.