Non-English scoop, and it is the most important thing in today's feed. Etailment's Pascal Bayer asks what nobody in the American retail trade press is asking: when an AI agent completes a checkout, how does fraud prevention distinguish the customer's authorized bot from an attacker who just cloned one? The answer is that it mostly doesn't. Fraud stacks were built around human signals — device fingerprint, behavioral velocity, timing patterns. A well-trained shopping agent mimics a power user by design and passes those checks without trying.

This matters now because Amazon's Alexa for Shopping is live, Shopify's agentic checkout integrations are shipping, and retailers shipped apps into ChatGPT and Claude faster than the evaluation frameworks exist. Every agentic purchase that completes is a transaction the existing fraud infrastructure wasn't designed to verify. OAuth tokens, saved payment methods, agent authorization chains — these are a new attack surface that the payment fraud industry has not yet priced in.

The German trade press gets this because Germany has stricter payment liability rules and its fraud vendors are closer to the regulatory edge. American retail AI coverage is debating whether agents will kill product discovery. The Germans are asking who eats the chargeback when the agent goes rogue. One of these conversations is more urgent.

Bain's China trajectory note puts the market heading toward RMB 1.5 trillion in e-commerce volume. The direction is up, which is what you'd expect from a note Bain publishes to sell China strategy engagements. What the note cannot say — because it would undermine the pitch — is that Alibaba is spending ¥380 billion defensively in that same market because it's losing share to Pinduoduo, JD, and Douyin. A 1.5 trillion RMB market where the dominant platform is bleeding position is not a stable opportunity. It's an arms race. Know which side you're entering on before you book the strategy engagement.

Non-English scoop. 证券时报 — China's Securities Times, not a trade blog — is reporting that Taobao Tianmao's twin-engine growth model is in deceleration: JD is poaching sellers, Pinduoduo is grinding prices, and Douyin is taking session time. Read against the ¥380 billion AI infrastructure pledge at Alibaba Pledged ¥380 Billion, the commitment looks different. A company investing aggressively from a position of strength and a company investing aggressively to arrest decline can announce the same number. The Securities Times is saying it's the latter. That matters for how Western brands should read Alibaba's pitch at their next partnership meeting.

BoF's opinion piece argues that AI shopping agents will comparison-shop so efficiently that brand loyalty evaporates and margins collapse — every purchase becomes a commodity decision. The threat is real. The half it misses: the brands that survive agentic shopping aren't the ones with the best price, they're the ones that can't be compared at all. If your product is legible to a catalog metadata layer, an agent can substitute it. If what you're selling is ritual, status, or experience that doesn't compress into a spec sheet, the agent can't evaluate you and the consumer finds you themselves.

The counterintuitive implication is that the most AI-resistant brands may be the ones that are deliberately hard for bots to discover — thin data, no affiliate integrations, no API hooks. That is a strange competitive moat to construct in 2026, but the data from Pinterest's luxury pitch points the same direction: the buyer who hasn't formed a question yet cannot be served by an agent. Taste is the last thing that scales slowly, and that's exactly why it's valuable.

McKinsey's live commerce is-transforming-shopping piece is competent and roughly three years behind the market it's describing. The actual news is in our piece today on Mangano and cable commerce: the format McKinsey calls transformative was a multi-billion-dollar industry before TikTok existed. The comment thread already closed 84% of TikTok Shop's beauty purchases. Taobao Live has been running since 2016. QVC predates the internet. The question Western brands should be asking is not whether live commerce works. It's why they keep needing a management consultant to confirm it.

Snap is expanding AR try-on tools for fashion retailers and the demos look good, because they always look good. The real news is that we are still having this conversation. The Taobao virtual try-on paper in May was unambiguous: model quality is a solved problem. The bottleneck is catalog metadata — color variants, fabric weight, sizing attributes — that most retailers' product databases cannot provide at the quality the rendering requires. Snap can make the jacket look photorealistic on your face. Whether the jacket's actual measurements are clean enough for the fit to be meaningful is a different department, and that department is usually a spreadsheet from 2019 that nobody owns.

Snap's commerce push is real and the AR layer matters for discovery. But every partnership announcement in this space will run into the same wall until retailers treat catalog data as infrastructure. Right now it's an afterthought with a job title nobody keeps for more than eighteen months.

This week's preprint proposes cascading multiple independent recommendation agents — retrieval, ranking, layout — instead of a single monolithic ranker for personalized e-commerce storefronts. The claimed benefit is modularity. What it actually describes is structurally identical to what Google's AgenticRecTune showed last week: the recsys gain is moving out of the model and into the orchestration layer between models. Two independent research groups converging on the same architecture in the same seven days is not a coincidence. The next generation of retail recommendation infrastructure is an orchestration problem, and the model vendors are about to discover that their moat is narrower than they thought.

16% projected e-commerce growth for Southeast Asia, powered by video commerce and AI-driven discovery. This is the demand-side infrastructure our piece today on Hanoi's $110 billion export forecast is wagering on from the supply side. The platform pipes are real. Shopee is real. The volume is real. What is not yet real is whether Vietnam's apparel and furniture manufacturing can scale at platform speed. That is not a Shopee problem or an AI problem. It is a roads-and-warehouses problem, and a growth forecast does not solve it.

Modern Retail names three structural shifts: mass drugstore closures, luxury consolidation (Saks, Neiman Marcus), and retail media network dominance. The retail media point is real and consistently underweighted in brand budgets. Selling through Amazon, Walmart, or Target now means renting the shelf twice — once with margin, once with advertising spend — to reach the customer who already chose your retailer. Calling this a new rule is charitable. The toll arrived in the same mail as the opportunity and has been rising since 2020.

The detail Modern Retail skips: the media spend brands are pouring into retail networks is funding the data infrastructure those same retailers will use to launch private-label products with better targeting than the brands have. You are paying to train the model that will eventually compete with you. This is worth knowing before the next media plan is approved.

Guerlain has created more than 1,100 perfumes across 198 years. It has never paid an influencer — until a $660 fragrance went viral on TikTok, the dupe conversation exploded around it, and the brand apparently decided that 198 years of prestige positioning could no longer hold the line alone. The campaign exists because the dupe content arrived first and the brand is responding, not leading.

Compare this to Refy handing twelve creators full control with no approval round — a brand that designed the campaign around virality from the start. Guerlain is catching up. What's interesting is that the $660 price is surviving the dupe conversation at all: the customer who knows a $40 alternative exists and buys the original anyway is not a customer who needs influencer validation. She needs acknowledgment. The campaign Guerlain actually needed was probably something closer to "you already know" than a product push. Whether they ran that is behind Glossy's paywall. The instinct that kept them off TikTok for 198 years was not entirely wrong — it just finally ran out of runway.

Alana Pallister sold her house to fund 300,000 units of the Blare tracksuit. Less than five months later the brand had moved one million of them. Every startup advisor in the room would have said test with 3,000 units, learn, iterate. Pallister bought 100 times that number and was right. The Coachella bet is next, and the real question is not whether the brand can produce a second moment — it's whether anyone at Coachella will be surprised by it. The first advantage was that nobody saw it coming. That advantage is gone now.

The Pheloung sisters named their charter jet "Acquired Air" before takeoff and used the trip to build pre-launch heat for Phe Phe. Planned, disciplined, effective. Compare it to Swan Beauty's @acquiredstyle moment — an AI mirror startup that was four months old when the trip broke the internet. Swan's moment worked because it felt accidental. Phe Phe's worked because it was meticulous. The lesson is not that one approach beats the other. The lesson is that bachelorette trips are now a product launch format, and the ones that look spontaneous are almost certainly not.

Algolia's sixth annual ecommerce search report finds that AI investment is resilient and search is retail's top digital priority. I note this not to dismiss the data — the sample is large and the direction is consistent with non-vendor sources — but because the framing requires disclosure. Every vendor's annual report finds that the thing they sell is the most important thing in retail. The search finding is directionally correct: search ROI is measurable and more defensible than other AI spend, which is why the trust gap we covered in May tends to be narrower in search than in generative creative. But "AI investment resilient" in a search vendor's report means something narrower than it sounds in a headline, and the headline is what gets quoted in the board deck.