Stripe is building the rails. Your product data decides who wins on them.

I spent a day at Stripe Sessions in San Francisco. This was one of the busiest conferences I've ever attended.

Always a treat to hear Sam Altman in conversation with Patrick Collison. You get the clearest signal on where AI is actually heading.

This time, the signal was unmistakable. Agentic commerce has moved from “interesting future” to “build now.”

Here's what stood out:

• Stripe announced 288 new products and features. The scale of what they're working on to build economic infrastructure for AI is staggering.
• They sized the B2C agentic commerce opportunity at $5T by 2030 and named the brands already moving: Etsy, Coach, Kate Spade, Quince, Fanatics, Best Buy, Halara, Ashley, JD Sports, Wix, BigCommerce.
• They introduced a five-level framework for agentic commerce maturity: no more forms, descriptive search, persistent memory, delegating decisions, invisible anticipation.
• The Agentic Commerce Suite expanded with partnerships on Google's UCP and Meta's native ad checkout, plus Link's new agent wallet and Shared Payment Tokens, the primitive that lets agents securely pass payment credentials between buyer and seller.

But one slide stopped me.

It had a product image of a black linen dress on the left. On the right, the JSON object the agent sees, with seven fields: product_id, name, description, color, color_hex, material, price. That's what the agent has to work with.

Now think about how a real shopper actually asks her agent to find this dress.

• “Find me a sleeveless midi dress for a summer wedding”: length and occasion aren't in the feed.
• “Something with tie-shoulder details, A-line, V-neckline”: the actual design isn't there.
• “A vacation dress under $200, machine washable”: no use case or care instructions.
• “Unlined linen, quiet luxury, true to size”: no lining, aesthetic, or fit guidance.
• “A black dress that pairs with tan sandals and gold jewelry”: no compatible accessories, no styling.

The agent can't answer any of these. Not because the dress doesn't have these qualities, but because none of it is structured into the data the agent reads.

This is the entire thesis behind what we're building at Lily AI.

When the buyer is an agent, your product page is no longer the unit of commerce. Your structured product data is. Length, silhouette, neckline, closure, occasion, care, fabric weight, lining, aesthetic, compatible accessories. Every one is a typed field an agent can rank, filter, and transact against. Or it isn't, and you're invisible.

The discovery surface keeps shifting. Search, social, marketplaces, and now agents. The product intelligence layer compounds under it.

Stripe is building the payment rails for agentic commerce. The brands that win on those rails will already have their product data ready.

It was great catching up with Tanvi Narain of Counterpart Ventures and Mandeep Bhatia of Tapestry. Thanks, Mandeep, genuinely lucky to have partners thinking this far ahead about where commerce is headed.