We watched an AI agent shop across hundreds of thousands of stores — here's what it actually reads

We gave an AI agent a deliberately vague request — “find me something warm to wear indoors, real wool, not synthetic, Norwegian if you can” — and watched it shop.

It didn't open a browser. It didn't visit a store. In a few seconds it queried Shopify's Global Catalog, weighed hundreds of results from merchants it had never been told about, applied a constraint we only implied — Norwegian-made — and returned a 100% traceable merino half-zip from a Norwegian brand, priced better and matched more precisely than the same search typed into Google. Then it built a cart and generated a checkout link, ready to pay.

The surprising part wasn't that it worked. It was how little of each product the agent actually looked at to get there — and what that means for every merchant whose products it will read next.

The instinct is to picture a smarter Google: the agent “searches the web,” lands on product pages, and reads them like a person would. That's not what happens. Shopify exposes a Global Catalog at a single endpoint. One query reaches across every merchant that participates and comes back ranked by relevance. The agent never crawls store to store — it asks once, and the catalog does the matching on its side. What comes back isn't a page of blue links; it's structured product data the agent reasons over directly.

Every agent interaction over the Universal Commerce Protocol moves through four clean, permissioned stages. Read top to bottom, it's roughly what a careful human shopper does — the difference is a machine doing it at the scale of the whole catalog, reading structured data instead of rendered pages.

Watching the agent discover and choose, it read a narrow, consistent slice of each product. Those few fields carried the entire decision. Everything else in the payload — and there is a lot of it — sat unread while the match was made.

The fields you may treat as an afterthought are the ones the agent leans on hardest. A title padded for a human skimming a collection page (“NEW ✨ Bestseller — Cozy Winter Knit”) reads as noise to an agent matching “warm wool, made in Norway.” A description that leads with brand story instead of material and origin gives it less to match on. Tags that are internal shorthand leave you invisible to the query that should have found you. There's a specific opportunity for fashion and apparel here: aggregating your product data — consistent material and origin fields, structured variants, tags that describe fit and use — is what makes an agent confident enough to surface you over a bigger competitor whose data is messier. Clean, complete, honest product data isn't hygiene; it's distribution.

Today an agent that finds products across many merchants still checks out at each one separately — one cart, one merchant, one payment. That boundary is moving. Shopify's Universal Cart — cross-merchant baskets that let an agent assemble items from different stores and check out once — is in early access now. This isn't a prediction; it's a capability being rolled out. When one basket can span merchants, the friction that keeps an agent loyal to one store disappears, and discovery-by-relevance extends all the way through to payment. When the cart no longer belongs to one store, being the best-matched product for your slice of the request is the whole game.