Stripe launched the Agentic Commerce Suite at Sessions 2026. The pitch: give Stripe your product catalog, and they’ll let AI agents sell it for you. Wix, BigCommerce, WooCommerce, and Squarespace are already plugged in, alongside Etsy, Coach, and Urban Outfitters.
Connect your product catalog to Stripe. Select which AI agents you’d like to sell through. Stripe helps with discovery, checkout, payments, and fraud detection, and sends you order events so you can continue using your existing commerce stack.
The payments industry has a term for this: a sales channel. In-store. E-commerce. MOTO. Each channel comes with its own liability framework — who’s responsible when a transaction goes wrong, how authentication works, what fraud vectors the issuer and acquirer need to watch for. Card-present has chip and PIN. Card-not-present has 3DS. MOTO has voice authorisation and manual review. The rules differ because the risk profiles differ.
Stripe is building a fourth sales channel but piggy-backing on the liability rules of an existing one. The Agentic Commerce Protocol uses Shared Payment Tokens — the human authenticates through Stripe’s checkout, and the agent gets a scoped credential with no cryptographic identity of its own. From the scheme’s perspective, this looks like a card-not-present transaction initiated by a human. It isn’t. It’s a transaction initiated by software acting with delegated intent — the human said “find me running shoes under €120,” but the agent picked the brand, the size, the colour, and the moment to buy. The human may not have individually approved the specific purchase.
That liability mismatch will matter. Issuers need to defend cardholders against fraud, and an agent making purchases autonomously breaks the behavioural models that fraud systems are trained on. Acquirers need to protect merchants, and when a dispute arrives, “the AI bought it” isn’t a response code anyone’s built rules for. The identification gap I wrote about in The Payment Lasagna — where agents are treated as passive form factors despite having active autonomy — is exactly the gap Stripe is building on top of.
But the real story isn’t the payment protocol. It’s the catalog.
Stripe is asking merchants to upload their products, prices, and availability to Stripe’s infrastructure — and Stripe syndicates that to every connected AI agent. That’s the same pitch marketplace platforms have made for years: give us your catalog, we’ll give you distribution. Merchants who sell on Amazon, on Zalando, on Bol.com already know this game. They’ve spent a decade learning that whoever owns the canonical product feed controls the customer relationship.
The difference is scale. A marketplace is one additional sales channel. Agentic commerce is every AI agent simultaneously — ChatGPT, Gemini, Copilot, and whatever comes next, all querying the same Stripe-hosted catalog. The operational challenge that merchants have been half-solving since the early days of e-commerce — real-time inventory across channels, unified fulfilment, consistent pricing — becomes unavoidable. An agent doesn’t know or care whether it’s selling “in-store stock” or “e-commerce stock.” It just wants to know if the item exists and can be delivered.
The platforms that already solved this — Shopify, commercetools, the headless commerce players with structured product APIs — are the ones that can plug in overnight. The merchants still running on spreadsheets and manual inventory counts? This isn’t their channel.
And there’s a question Stripe isn’t answering yet: who owns the customer? When an agent buys a pair of shoes from Coach through the Agentic Commerce Suite, Coach gets an order event. But the agent chose Coach — the human didn’t browse the site, didn’t see the brand experience, didn’t enter the store. The distribution win comes with a brand control loss. That’s a trade-off every marketplace seller already understands. Now it’s coming for direct-to-consumer brands too.