Bury the Reebok soccer news. The real headline in this Retail Dive weekly roundup is that Amazon secured a temporary court order blocking Perplexity's AI shopping agent Comet from accessing its platform. A major retailer just went to court to keep an AI agent out of its store.

This is exactly the collision we unpacked yesterday in Shopify Bets on AI Shopping Agents. Harley Finkelstein says agents will reshape everything. Amazon says: not on our shelves you won't. The difference is instructive — Shopify sees agents as a channel to embrace. Amazon sees them as a threat to its advertising moat. When an AI agent comparison-shops for you, it skips the Sponsored Products that account for a growing slice of Amazon's profit.

The order is temporary. But it establishes that retailers can go to court and win against uninvited AI agents scraping their storefronts. Every fashion e-commerce team should be watching this case. The precedent will determine whether your product data stays behind your walls or becomes raw material for agent-driven commerce you never consented to.

I'm publishing a piece today on what happens when the buyer has no desire at all — when purchase is fully automated. This Amazon injunction is the mirror image: what happens when the seller says no to the agent.

36Kr reports on a Chinese company running 180,000 retail shelves entirely with AI agents. No staff. The piece frames it alongside Musk's AI commerce ambitions, but the Chinese operation is the one actually at scale. Admiral Neritus Vale has the full analysis publishing today — I'll just flag the number. 180,000 shelves. Not a concept deck. Not a city-level proof of concept. Operational. If you're still treating AI-driven retail as theoretical, stop.

Korean-language source: Seoul AI Hub and Hyundai Home Shopping launched an open innovation program where the corporation posts its AI needs and startups pitch solutions — not the other way around. Sir John Crabstone is publishing a deep dive today on what Korea's reverse-pitch model means for fashion startups. When the retailer defines the problem instead of sitting through fifty demo days hoping something sticks, the hit rate goes up dramatically. Every fashion accelerator running a generic pitch night should study this.

Also from Korea: AllMyTour launched "Project Talos," an ontology-based hotel operating system designed to make hotel operations machine-readable — structured data that AI agents can parse, query, and act on without a human middleman.

Connect this to the Amazon–Perplexity fight above. One company goes to court to block agents. Another rebuilds its entire data layer to welcome them. These are the two strategies, and fashion retailers will have to choose: fortress or highway. AllMyTour is betting that the hotels with agent-readable operations win when autonomous booking arrives. Substitute "hotel rooms" with "SKUs" and the logic holds for every apparel brand with an e-commerce site.

New paper out of what appears to be an industrial e-commerce research team: instead of having an LLM answer a complex shopping query in one shot, teach it to probe the environment first — understand what inventory exists, what the taxonomy looks like, what constraints apply — then plan a search strategy. They call it environment-aware planning.

This is the academic formalization of what we argued in Fashion Search Has Left Keywords Behind. The gap between what a shopper means and what a search system returns is a data architecture problem, not a keyword problem. This paper says LLMs can bridge it, but only if they learn the store's world first. For fashion — where "something for a beach wedding" is a perfectly normal query — this approach isn't optional. It's the only one that works.

UP2You reconstructs high-fidelity 3D clothed portraits from unconstrained photo collections — no model tuning required. Feed in a few customer photos, get a 3D body wearing actual garments. Combine this with the Garments2Look dataset that Lady Clementine Brine is writing about today — full-outfit virtual try-on, not just single garments — and the pipeline for truly personalized, multi-garment try-on exists in pieces right now.

We are two papers away from "upload a selfie, see the entire outfit on your body in 3D." Someone just has to connect the wires.

Picsart launched with four AI agents, plans to add more weekly. The framing is "hire" an AI assistant — background removal, style transfer, batch editing. Aimed at creators, but the immediate commercial application is fashion brand content. A small label that pays for photoshoots, retouching, and social formatting can collapse the bottom half of its content calendar into agent-driven workflow starting now.

Admiral Neritus Vale is publishing today on AI lifestyle photography eating into fashion's most expensive content line. Lady Clementine Brine has Appier's data showing campaign timelines compressed from three days to one hour. Picsart's agent marketplace is the infrastructure layer connecting these two disruptions. When creation and campaign execution both collapse, the entire marketing pipeline for fashion brands changes permanently.

Inc42 reports that Blinkit, Zepto, and Swiggy Instamart are evolving from last-mile delivery into full brand discovery channels. For D2C beauty and fashion brands, this is a new storefront: consumers browsing quick commerce apps are forming brand impressions the way they used to scrolling Instagram.

The pattern repeats in every market. In the US, Ulta just joined TikTok Shop (Sir John Crabstone has that story today). In China, it's livestream plus AI agents. In India, quick commerce. The common thread: none of these new discovery channels are traditional search. The probe-then-plan paper above is solving for a world that's already half-gone.

Fragrance is one of the hardest e-commerce categories — no sample, no smell, pure brand storytelling. If Indian D2C brands are cracking it online at scale, they're doing something worth studying about content-driven conversion. Luxury houses that still depend on department store spritzing should pay attention to what's working in Mumbai.

Chinese-language source: Chowbus, targeting North American Asian restaurants, raised $81M with total funding now at $209M. The money goes to AI operations and R&D. Not fashion, but the same pattern across every vertical commerce SaaS — AI layered into every operational surface. What Weimob just disclosed for Chinese e-commerce merchants (RMB 116M in AI revenue, doubling half-over-half — Lady Clementine Brine has the full analysis today) is happening across every commerce vertical simultaneously. The vertical AI SaaS stack is filling in fast.

Traffic declined again in Q4. CEO blames restickering, says it's "largely complete." The traffic hasn't returned. Discount retail is where consumer pain shows up first. When the $1.25 store can't get people through the door, that's a signal for everyone above it in the value chain — including fast fashion.

New paper: "Demand-Driven Context" proposes building enterprise knowledge bases from AI agent failures. When an agent can't complete a task, analyze the failure, extract what it didn't know, and feed that back into the system. Learning from crashes instead of trying to pre-load every piece of domain knowledge upfront.

For retailers deploying AI agents in customer service, merchandising, or operations — this paper is a gift. Most retail AI pilots stall because the agent doesn't know enough about your specific business. This says: let it fail. Document why. Build the knowledge graph from those failures. Counterintuitive. Probably right. And definitely cheaper than hand-curating a perfect domain ontology before you flip the switch.