Point an iPhone at a saved screenshot of a stranger’s outfit and Daydream hands back the coat, the boots, the bag — real products, priced, ready to buy. The results arrive as a clean editorial grid, not a list of blue links, and you refine them by talking instead of filtering. There is no search box to fill and no query to phrase. Julie Bornstein’s wager is that the checkout was never the hard part in apparel; taste was, and no one has automated it.

The app went live in June 2025 with nearly two million products from more than 8,000 brands, Net-a-Porter and Khaite shelved beside Uniqlo and Dôen, on $50 million in seed money from Forerunner, Index Ventures, True Ventures, and Google’s GV. It carries no inventory, sells no ad placement, and takes a commission when it sends you off to the retailer to pay. Bornstein called it “the ChatGPT of shopping” at BoF VOICES 2024. That framing mistakes the interface for the product.

Discovery in apparel was always somebody’s eye before it was anyone’s algorithm. Diana Vreeland’s “Why Don’t You…?” column ran in Harper’s Bazaar, a stream of styling instructions issued before the reader knew she wanted them. A buyer or a personal shopper sells the same thing: judgment, the nerve to say no on your behalf. Search engines swapped that eye for a query and a ranking; the marketplace feed swapped it for whatever moves volume.

The giants are racing, and they are racing toward the transaction. Amazon built Rufus to answer questions inside its shopping app, and is already folding it into an Alexa shopping agent. Google’s AI Mode now reaches across more than fifty billion listings, renders how a fabric folds and drapes on your own body, and will complete the purchase for you. Retrieval, rendering, payment: largely solved. None of it is taste.

Taste is not retrieval; it is refusal.

This is where Daydream’s bet turns interesting, and exposed. Bornstein describes the app as “an intelligent layer over the world of fashion. One that understands taste, context, and individuality,” trained on global catalogs and your own corrections. Feed a model the entire catalog and it drifts toward the center of it: the safely pretty, the broadly liked, the midi dress that offends no one. Pay an engine per click and it has every reason to flatter your last tap, none to call it a mistake. A good buyer’s value was that she could say no and mean it.

Bornstein has run this experiment before. The Yes, her last taste-driven shopping app, launched in 2020 and sold to Pinterest in 2022 for $87.6 million, which promptly switched it off. The wager this time is that large language models finally close the gap between what a shopper can describe and what she means. The interface has changed — whether the eye behind it has is the only question worth asking.