Yesterday in Berlin and Paris, the agentic commerce map of the world got its European coordinate, and it arrived in the strangest packaging possible: a margin miss, a Wall Street warning about AI, and a retailer volunteering to buy its competitors’ products for you.

Zalando, Europe’s largest online fashion and lifestyle platform, reported third-quarter results and used the occasion to announce Zalando Extended, a new platform layer where Zalando’s own AI agent will buy products from other online shops on the customer’s behalf, inside the Zalando app. Westwing, the premium home and living retailer, is the first pilot partner, with agent-driven checkout going live from 2027. It is the clearest example yet of a strategy we have been tracking all year: rather than walling the store off from AI agents, own the agent, keep the session, and monetize everything the agent touches.

The decision only makes sense against the quarter behind it, and the quarter only makes sense against the year around it. Both are worth walking through, because what Zalando just did is what half of retail will try to copy by next holiday season, and because the announcement contains a line about advertising that deserves far more attention than it got.

The quarter that forced the move

Start with the numbers from Zalando’s Q3 2026 release and Reuters’ report on the announcement:

  • GMV: 4.54 billion euros, up 7.5 percent, with growth across all three consumer apps
  • Revenue: 3.17 billion euros (about 3.55 billion dollars), up 5.1 percent, beating the 3.15 billion euro LSEG consensus
  • Gross margin: 38.7 percent, down 0.9 percentage points, a miss against the expectations modeled by Jefferies and JPMorgan
  • Active customers: 63.2 million, up 2.9 percent, a record, with average annual spend up 2.9 percent to 308 euros
  • B2B revenue: 354 million euros, up almost 28 percent, with adjusted EBIT doubling to 42 million euros

The margin line is the one that moved the stock. Shares fell more than 5 percent in early trading before recovering to up 0.5 percent by mid-morning after the CFO’s explanation. Finance chief Anna Dimitrova told analysts the decline “was primarily driven by targeted investments in attractive offers for our customers in our retail business, especially in sneakers,” described the pressure as temporary, and forecast a broadly flat gross margin for 2026.

Then came the sentence that explains the entire announcement. JPMorgan, in a note to clients, said the stock could drift lower given “the lack of any incremental positives” and “ongoing structural risk from agentic commerce development.”

Read that carefully. One of Europe’s flagship analysts did not cite competition from Shein, weak consumer sentiment, or discount cycles as the structural risk. It cited agentic commerce: the shift in which AI agents, from ChatGPT to Meta Muse to Amazon’s Alexa for Shopping, increasingly handle discovery, comparison, and payment on the shopper’s behalf. The fear is disintermediation. If an agent owned by someone else does the discovering and the comparing, the retailer is demoted to an inventory and logistics provider, competing on price alone, invisible to the customer. Every basis point of gross margin Zalando loses to sneaker discounts makes that future more dangerous, because a retailer with thin margins and no customer relationship has nothing left to defend.

What Zalando Extended actually does

The mechanics, per the corporate release, are worth quoting precisely because the strategy hides in the verbs:

“Customers will never have to leave the trusted Zalando app, as its own AI agent will securely buy these products on their behalf.”

Zalando Extended adds products from online shops Zalando does not itself sell to the Zalando platform. The agent does the buying. Partner shops, Zalando argues, benefit too: they reach Zalando’s customer base “without complex technical integration or changes to their logistics.” The first category is deliberately chosen, and it is not fashion: it is bulky home goods, a high-consideration, review-driven category where the shopping experience is genuinely painful. The Westwing cooperation started this year, and from 2027 Zalando’s agent will be able to check out Westwing products directly in the Zalando app.

The backdrop is Zalando’s ongoing platform shift. The partner share, meaning goods sold by partners rather than Zalando’s retail business, rose 2.9 percentage points to 33.6 percent of B2C GMV, on the way to a 2028 target of 40 to 50 percent. Retail media revenue grew nearly 18 percent year over year and now equals 2 percent of B2C GMV. The Zalando Assistant, the conversational shopping tool, has drawn more than 22 million unique users this year. Zalando is steadily becoming a technology platform that other brands and retailers plug into: its B2B arm, which sells logistics, storefront software, and marketplace connections through ZEOS, SCAYLE, and Tradebyte, is the fastest-growing thing in the company.

Zalando Extended is the logical endpoint of that arc. The platform no longer stops at the edge of its own inventory. The agent is the store.

The inversion: from blocking agents to being the agent

Put yesterday’s announcement next to the year’s other agentic commerce moves and a pattern emerges. Amazon has spent the quarter blocking outside agents from its product graph while declaring Alexa for Shopping the primary shopping agent for Amazon, now used by more than 400 million customers. Walmart opened a controlled agent gateway with spending caps. Google is building a universal cart across its commerce graph. The instinct everywhere is the same: agents are coming, so control the door they walk through.

Zalando’s version is the most aggressive inversion yet, because it inverts the direction of the door. Amazon’s message is: agents may shop for me, never against me. Zalando’s message is: my agent will shop anywhere, including my competitors, and the customer will experience all of it as Zalando.

There is real strategic logic here. The analyst fear is structural: lose the discovery session, lose the customer relationship, become fulfillment commodity. Owning an agent that shops beyond your own inventory converts that threat into an expansion. Zalando keeps the session even when the product is not Zalando’s. The app becomes the front door to the entire internet’s inventory, with Zalando’s brand as the trust wrapper. And in the long run, the company that owns the agent session also owns the data about what customers want across the whole market, not just one catalog.

But the move creates three problems the press release does not address.

The referee sells the ads on the rivals’ inventory

The first problem is the conflict of interest, and this time it comes in a triple stack. Buried in the corporate release is this sentence:

“In addition, they will be able to advertise their products on Zalando through its retail media business, creating an additional revenue stream for Zalando.”

So the shape of the new system is: Zalando’s agent selects among products, some of which Zalando sells itself, some of which belong to third-party shops, and those third-party shops can pay Zalando for visibility through retail media. The referee owns one of the teams, sells the advertising that influences the calls, and increasingly monetizes the inventory of the shops it referees. Amazon’s Rufus faced the referee question when Sponsored Prompts became billable placements inside shopping conversations. Zalando Extended raises the stakes, because the agent is explicitly shopping across retailers, and paying for placement may determine which retailer it shops.

When the shopper asks for a sideboard and the agent can buy from Zalando’s own home category, from Westwing organically, or from whichever partner is sponsoring, whose ranking logic decides? Zalando’s answer will presumably be some mix of relevance and commercial rating. Every platform that has ever answered that question has discovered that the commercial rating wins more often than users expect. In an agent, subtlety is cheap: nudging a recommendation by one position, one phrasing, one “best overall” label costs nothing visible and moves real revenue.

Whose reviews does the agent read?

The second problem is the one this publication exists to track: product evidence. When Zalando’s agent checks out a Westwing sideboard, what does the shopper see about that sideboard’s quality, and where did it come from? Westwing’s product reviews are not Zalando’s review corpus. There is no shared evidence layer across retailers, no standard for cross-store product verification, and no disclosure of which signals the agent weighed. The shopper sees the Zalando interface and extends Zalando’s trust to a product Zalando has never sold, shipped from a logistics network Zalando does not operate.

This matters more, not less, because the pilot category is furniture and home goods, a category where reviews are the primary risk-reduction tool. And the review layer itself is exactly where the market’s integrity is weakest:

  • Detection estimates for inauthentic Amazon reviews range from roughly 7 to 42 percent depending on method, and Amazon itself reported blocking more than 275 million suspected fake reviews in a single year.
  • Pangram’s analysis of front-page Amazon reviews found that 93 percent of first-page AI-generated reviews carried the “Verified Purchase” badge, which means the badge shoppers and agents lean on most no longer certifies anything about authenticity.
  • The FTC’s Consumer Review Rule now carries civil penalties above 53,000 dollars per violation, and the agency sent warning letters to ten companies last December, but enforcement catches sellers, not the review data an agent ingests in real time.

An agent that summarizes unverified review data does not fix the problem. It launders it. The summary looks authoritative because the model wrote it well, and the shopper cannot see which inputs were fake. Multi-retailer agents make this strictly worse, because now the agent is mixing evidence from multiple unverifiable corpora into a single recommendation.

The evidence economy is the constraint, not the checkout

The third problem is the one the data says shoppers care about most. This week’s FirstPageSage update showed agent-completed orders hitting 1.52 percent of US ecommerce in October, a 5.6x jump in one quarter, while only 1.5 percent of shoppers let an agent buy fully unsupervised against 7.9 percent who approve each checkout. Shoppers overwhelmingly want a final say. And the Harris Poll line aggregated by SearchPanel remains the sector’s most useful number: 73 percent of shoppers trust AI recommendations more when the recommendation shows evidence, reviews, specs, and sources.

That is the actual adoption curve. Autonomy grows at the speed of trust, and trust grows at the speed of shown evidence. A retailer agent whose selection is sponsored and whose evidence is borrowed from third-party review corpora is optimizing the two things shoppers said they would punish. The retailers that win the agent era will be the ones that can answer one question convincingly, for every product the agent touches: how do I know this product is good, and can I see why?

That question has an answer, but it cannot come from the party selling the product or the party selling the ads. It has to come from a neutral layer. That is the design brief behind GoBuy: fake reviews filtered out before scores are computed, a Smart Score from 0 to 100 built on review quality rather than quantity, a curated top seven rather than an infinite shelf, and a GoBuy Verified badge reserved for products that hold a score above 80 across 90 days. Exposed through MCP at gobuy.ai/api/mcp, it is the kind of independent evidence endpoint an agent like Zalando’s could call before any purchase, from any retailer, without the shopper having to trust the referee’s own inventory incentives.

What to watch now

Four things will tell you whether Zalando Extended is a landmark or a footnote:

  • The 2027 Westwing checkout go-live: does the agent show its work when it recommends a third-party product, or only a price and a buy button?
  • Retail media disclosure: when a partner pays, does the shopper or the agent’s output say so, and will EU regulators, who have opinions about platform self-preferencing, treat an agent that shops its owner’s rivals while selling them ads as a gatekeeper question?
  • The copycats: expect every major multi-brand retailer to announce an owned agent within twelve months. The differentiator will not be the checkout, everyone can build that. It will be whether the agent’s product claims survive scrutiny.
  • The metrics: whether Zalando ever discloses what share of Extended orders were influenced by sponsorship. Amazon reports Alexa interactions up 570 percent and no one outside the company knows what fraction of answers were paid. Watch whether Zalando repeats that omission.

Retailers spent 2026 deciding between walls and doors. Zalando just demonstrated the third option: become the shopper’s agent, shop everyone including yourself, and monetize the whole map. It is the most intellectually honest response to the structural risk JPMorgan named, because it accepts the agent future instead of litigating against it. Honesty about the future, though, is not the same as honesty about products. The first retailer whose agent can prove what it recommends, with evidence nobody paid for, is the one that keeps the session after the novelty wears off.

GoBuy is the trust layer before you buy on Amazon: fake reviews filtered out, authentic ones weighted up, and a Smart Score from 0 to 100 built on review quality, not quantity. Developers can plug the same evidence into any agent through MCP at gobuy.ai/api/mcp, and agent integration docs are at gobuy.ai/agent-docs. Try it at gobuy.ai.