Julie Bornstein thought it might be a cinch to implement her concept for an AI startup. Her résumé in digital commerce is impeccable: VP of ecommerce at Nordstrom, COO of the startup Sew Repair, and founding father of a personalised purchasing platform acquired by Pinterest. Style has been her obsession since she was a Syracuse excessive schooler inhaling spreads in Seventeen and hanging out in native malls. So she felt well-positioned to create an organization for patrons to find the right clothes utilizing AI.
The truth was a lot more durable than she anticipated. I had breakfast just lately with Bornstein and her CTO, Maria Belousova, to find out about her startup, Daydream, funded with $50 million from VCs like Google Ventures. The dialog took an sudden flip as the ladies schooled me on the shocking problem of translating the magic of AI techniques into one thing individuals really discover helpful.T
Her story helps clarify one thing. My first publication of 2025 introduced that it might be The 12 months of the AI App. Although there are certainly many such apps, they haven’t remodeled the world as I anticipated. Ever since ChatGPT launched in late 2022, individuals have been blown away by the methods carried out by AI, however research after research has proven that the expertise has not but delivered a major increase in productiveness. (One exception: coding.) A research printed in August discovered that 19 out of 20 AI enterprise pilot tasks delivered no measurable worth. I do suppose that productiveness increase is on the horizon, but it surely’s taking longer than individuals anticipated. Listening to the tales of startups like Daydream which can be pushing to interrupt by way of offers some hope that persistence and persistence would possibly certainly make these breakthroughs occur.
Fashionista Fail
Bornstein’s authentic pitch to VCs appeared apparent: Use AI to resolve difficult trend issues by matching prospects with the right clothes, which they’d be delighted to pay for. (Daydream would take a minimize.) You’d suppose the setup could be easy—simply hook up with an API for a mannequin like ChatGPT and also you’re good to go, proper? Um, no. Signing up over 265 companions, with entry to greater than 2 million merchandise from boutique retailers to retail giants, was the simple half. It seems that fulfilling even a easy request like “I would like a costume for a marriage in Paris” is extremely complicated. Are you the bride, the mother-in-law, or a visitor? What season is it? How formal a marriage? What assertion do you wish to make? Even when these questions are resolved, totally different AI fashions have totally different views on such issues. “What we discovered was, due to the dearth of consistency and reliability of the mannequin—and the hallucinations—generally the mannequin would drop one or two components of the queries,” says Bornstein. A consumer in Daydream’s long-extended beta take a look at would say one thing like, “I’m a rectangle, however I would like a costume to make me appear like an hourglass.” The mannequin would reply by exhibiting attire with geometric patterns.
In the end, Bornstein understood that she needed to do two issues: postpone the app’s deliberate fall 2024 launch (although it’s now obtainable, Daydream remains to be technically in beta till someday in 2026) and improve her technical staff. In December 2024 she employed Belousova, the previous CTO of Grubhub, who in flip introduced in a staff of prime engineers. Daydream’s secret weapon within the fierce expertise struggle is the possibility to work on an enchanting downside. “Style is such a juicy house as a result of it has style and personalization and visible knowledge,” says Belousova. “It’s an fascinating downside that hasn’t been solved.”
What’s extra, Daydream has to resolve this downside twice—first by decoding what the client says after which by matching their generally quirky standards with the wares on the catalog facet. With inputs like I would like a revenge costume for a bat mitzvah the place my ex is attending along with his new spouse, that understanding is essential. “We’ve this notion at Daydream of purchaser vocabulary and a service provider vocabulary, proper?” says Bornstein. “Retailers communicate in classes and attributes, and consumers say issues like, ‘I’m going to this occasion, it’s going to be on the rooftop, and I’ll be with my boyfriend.’ How do you really merge these two vocabularies into one thing at run time? And generally it takes a number of iterations in a dialog.” Daydream discovered that language isn’t sufficient. “We’re utilizing visible fashions, so we really perceive the merchandise in a way more nuanced means,” she says. A buyer would possibly share a selected shade or present a necklace that they’ll be carrying.
Bornstein says Daydream’s subsequent rehaul has produced higher outcomes. (Although once I tried it out, a request for black tuxedo pants confirmed me beige athletic-fit trousers along with what I requested for. Hey, it’s a beta.) “We ended up deciding to maneuver from a single name to an ensemble of many fashions,” says Bornstein. “Every one makes a specialised name. We’ve one for shade, one for cloth, one for season, one for location.” As an illustration, Daydream has discovered that for its functions, OpenAI fashions are actually good at understanding the world from the clothes standpoint. Google’s Gemini is much less so, however it’s quick and exact.
