JD.com (京东) has publicly revealed its Oxygen AI architecture for e-commerce for the first time. The framework covers the full stack: recommendation, search, supply chain, and customer interaction layers.

The act of publishing matters as much as the content. When a platform makes its AI architecture legible to the outside world, it is recruiting developers and ecosystem partners — not just claiming turf. This drops the same week Alibaba's quarterly results show it doubling down on AI and instant retail. Two platforms making their bets visible simultaneously is not a coincidence. The third platform — Pinduoduo — has said nothing. Either they are behind or very far ahead.

极睿科技 (Jirui Tech) closes a 100M+ RMB Series A led by Sequoia China. Jirui provides content intelligence and visual AI for fashion and e-commerce brands — the pitch is monthly sales lift in the tens of millions of RMB per client.

Sequoia China does not lead rounds on vibes. What they are backing is infrastructure: the AI layer below the brand, providing content production and visual recommendation rails that Taobao and JD.com merchants depend on. This is the China-side of the fashion AI buildout that is running at least two years ahead of equivalent Western plays. The US comparable would be something between an AI creative production tool and a visual merchandising platform — but neither maps cleanly. China is building integrated stacks. The West is still stitching point solutions together and calling it a strategy.

36Kr asks whether Mode.ai signals the end of human buyer-led fashion recommendation. The platform uses AI to surface picks and guide purchase decisions — not just predicting next clicks, but attempting to replace the editorial voice entirely.

The "is the buyer dead?" framing is overheated. But the underlying shift is real: algorithmic "you might like" has been in market for 15 years and customers barely notice it. What Mode.ai is reaching for is different — a recommendation that feels like taste rather than collaborative filtering. That is the gap wholesale platforms are also moving toward. Nobody has convincingly closed it yet. The brands that try will find out quickly whether the gap is a product problem or a trust problem.

Chinese financial media reports Gap Group is building checkout functionality directly inside Google Gemini — among the first mainstream Western fashion brands to embed a purchase flow inside an LLM. The story is getting more traction in Asia than in the US, possibly because Asian markets already understand operationally what this means.

We covered Gap's dual AI bets when the deals were announced. The Gemini checkout deserves more scrutiny than it has received: brands that build inside Gemini's purchase flow are ceding the transaction relationship to Google. Every conversion that happens inside Gemini is a data point Google owns and Gap does not. That may be the right trade — distribution beats ownership when you are fighting for discovery — but brands should be entering it clear-eyed about the exchange.

哈啰顺风车 (Haro Ridesharing) has launched a full MCP (Model Context Protocol) service, packaging its ride-matching, pricing, and checkout flow as standardized interfaces callable by any LLM or AI agent. Three service tiers, from a link-redirect mode up to full agentic booking.

Neritus Vale is writing today about a ride-hailing app publishing agent-callable endpoints — Haro is the confirmation it is becoming a pattern, not an experiment. A ridesharing company just handed any AI agent the ability to book a car without opening an app. The fashion checkout analogy is obvious. The execution gap is equally obvious: hailing a car has two variables. Fashion checkout has hundreds — size, color, fit, material, availability, return policy, brand relationship. MCP standardizes the interface; it does not simplify the product. The first fashion retailer to publish a clean agent endpoint for purchase will have done something genuinely hard.

Connect this to Gap's Gemini checkout two items up. Both moves describe the same migration: the transaction is leaving brand-owned surfaces and entering AI interfaces. One brand is choosing to go; Haro built the road. Fashion is watching both and largely has not picked a lane yet.

Two bankruptcy survivors find each other: BBB acquires The Container Store for $150M. The rebrand — "The Container Store / Bed Bath and Beyond" — is possibly the most graceless name in contemporary retail. The logic underneath it is more interesting than it looks.

The Elfa and Closet Works units anchoring a new home services division is the actual bet. Physical services create recurring customer relationships that AI recommendation engines actually learn from. Every installation appointment is a home data point — room dimensions, material preferences, organizational behavior — that a digital-only brand cannot collect. The brands that survive the next phase of retail AI will be the ones with regular physical touchpoints, not because stores are magic, but because physical presence feeds AI systems in ways digital-only brands cannot replicate.

Franziska Bell joins Home Depot as CTO with an explicit mandate to build AI-driven customer experience and operational tools. The announcement is spare on details and long on intent.

We wrote about Home Depot's AI strategy as a store amplification play — the thesis that AI there makes the physical footprint legible to the algorithm rather than replacing it. Bell's hire signals this is no longer a pilot program. Fashion retailers with smaller physical footprints, weaker product data infrastructure, and no dedicated AI CTO should be uncomfortable about where this trajectory leads.

QVC Group — owner of QVC and HSN — says there is "substantial doubt" about its ability to continue as a going concern. The company has not yet filed its 10-K.

QVC is the original live commerce platform. Twenty-five years before TikTok Shop, QVC proved that hosted video selling works at scale. The model was never wrong — it got outflanked. The difference between QVC and TikTok Shop is not the video format; it is who the host is. QVC built a trained presenter class. TikTok distributed that role to hundreds of millions of people simultaneously. Sir John is writing today about TikTok Shop's real return trajectory — QVC's going concern warning is the footnote that explains exactly where the live commerce energy and revenue went.

Notice the connection to Hollister's music video campaign two items down. QVC could not survive the disaggregation of the presenter role; Hollister is betting that human-made creative becomes scarce and therefore valuable precisely because algorithmic content has flooded the rest of the field. Two very different businesses, one underlying argument about what happens when production cost approaches zero.

Elliott Hill says Nike is not satisfied but confident in the direction. Q3 numbers are not where they need to be, and the timeline for recovery has stretched again.

The honest read: Nike's DTC pivot cost more wholesale relationships than AI recommendation can rebuild. They bet that owning the customer relationship directly would outperform shelf space next to competitors. It turns out discovery is hard — Amazon, Google, and TikTok are better at it than Nike's own app, and they always were. AI recommendation engines can optimize conversion. They cannot replace the impulse moment that proximity to competing brands on a retail floor used to generate. Nike is learning this the expensive way, and the lesson matters for every fashion brand that drew the same DTC conclusion from the same pandemic data.

Hollister partnered with Gigi Perez to cover a Green Day track for its largest summer campaign ever. No generative content. No AI. A music video.

In a week when every brand is announcing AI creative tools, Hollister bets on human craft — or at least analog affect. Their core customer is 13-17, a cohort that grew up with algorithmic content and may genuinely be hungry for something that feels made rather than generated. The counterargument: Green Day nostalgia is for their customer's parents, not them. Either way, in a market drowning in AI-produced content at $0.14 per second, scarcity of evident human effort might be the differentiator nobody modeled. The brands selling AI creative tools need customers who still care about the difference.

Zalando is launching its AI shopping assistant across all 25 of its markets simultaneously. That is not a test. That is a rollout — and the fact that it is getting coverage in Chinese media first suggests the announcement landed harder in Asia than in Europe, possibly because Asian markets are further along in evaluating what AI shopping assistants actually do in production.

Zalando's real asset here is not the LLM. It is years of European fashion preferences, return signals, sizing behavior, and brand affinities across 25 distinct markets — training material that a standalone AI shopping app cannot buy. The risk is trust asymmetry: Zalando is a utility, not a brand customers love. If the assistant makes one confidently wrong recommendation, the reputational damage lands on Zalando, not on the model. Nobody pricing AI features at rollout scale has fully accounted for that cost.

Listen Labs raises $69M for AI-powered customer interviews at scale — the bet that qualitative research can run at quantitative sample sizes without focus groups or quarterly survey forms.

The retail implication is underrated. Brands have always had two types of customer data: what people do (clicks, conversions, returns) and why they do it (interviews, surveys, occasional NPS pulses). AI recommendation engines are structurally excellent at the first and blind to the second. If Listen Labs works, it creates a continuous voice-of-customer stream that informs assortment, messaging, and sizing decisions in near real time. That closes a loop that behavioral data alone cannot close — knowing someone bought does not tell you they were anxious about it, or that they returned it mentally before it even arrived.

Korean fashion e-commerce platform Musinsa closes a $190M Series C and is now operating in 13 countries. Western fashion media has mostly missed this.

Musinsa is a curation-first marketplace built for a generation that finds Zara and H&M aesthetically flat — strong on Korean streetwear and emerging designer labels, expanding aggressively into markets where Zalando and ASOS have not locked up loyalty yet. The $190M is a bet that editorial curation scales internationally, which is exactly the same thesis behind Poshmark's recent redesign. Two very different platforms, same conviction, same quarter. When curation becomes the competitive differentiator, the platform that can make human taste AI-legible without flattening it wins. Musinsa's edge is that their curation is grounded in actual subcultures, not inferred from purchase history. That distinction matters more as competitors scale.