Here’s What You Should Know About Launching an AI Startup
Julie Bornstein thought it would be a breeze to implement his AI startup idea. Her digital commerce resume is impeccable: vice president of e-commerce at Nordstrom, COO of startup Stitch Fix and founder of a personalized shopping platform acquired by Pinterest. Fashion has been her obsession since she was a high school student in Syracuse, inhaling spreads in Seventeen and hanging out at local malls. So she felt well-positioned to create a company that helps customers discover the perfect clothes through AI.
The reality was much harsher than she expected. I recently had breakfast with Bornstein and his CTO, Maria Belousova, to learn more about his startup, Daydream, funded by $50 million from venture capital firms like Google Ventures. The conversation took an unexpected turn when the women explained to me the surprising difficulty of translating the magic of AI systems into something people actually find useful.
His story helps explain something. My first newsletter of 2025 announced that it would be the year of the AI application. Although there are indeed many such applications, they have not transformed the world in the way I had anticipated. Since ChatGPT launched in late 2022, people have been blown away by the tricks AI performs, but study after study has shown that the technology has yet to deliver a significant increase in productivity. (One exception: coding.) A study published in August found that 19 of 20 enterprise AI pilots delivered no measurable value. I think an increase in productivity is on the horizon, but it’s taking longer than expected. Listening to the stories of startups like Daydream striving to break through, one can hope that perseverance and patience might indeed enable these breakthroughs.
Fashionista Fail
Bornstein’s initial pitch to venture capitalists seemed obvious: use AI to solve tricky fashion problems by providing customers with the perfect clothes they’d be happy to pay for. (Daydream would take a cut.) You’d think the setup would be simple: just plug into an API for a model like ChatGPT and you’re good to go, right? Um, no. Signing up over 265 partners, with access to over 2 million products, from boutiques to retail giants, was the easy part. It turns out that responding to even a simple request like “I need a dress for a wedding in Paris” is incredibly complex. Are you the bride, the mother-in-law or a guest? What season is it? How formal is a wedding? What statement do you want to make? Even when these questions are answered, different AI models have different views on these topics. “What we discovered was that due to the model’s lack of consistency and reliability (and hallucinations), the model would sometimes drop one or two elements of the queries,” says Bornstein. A user participating in Daydream’s lengthy beta test would say something like: “I’m a rectangle, but I need a dress to make me look like an hourglass.” The model would respond by showing dresses with geometric patterns.
Ultimately, Bornstein realized she had to do two things: postpone the app’s planned launch to fall 2024 (although it’s available now, Daydream is still technically in beta until 2026) and upgrade her tech team. In December 2024, it hired Belousova, the former CTO of Grubhub, who in turn brought in a team of top engineers. Daydream’s secret weapon in this fierce war for talent is the ability to work on a fascinating problem. “Fashion is such a juicy space because it has taste, personalization and visual data,” says Belousova. “It’s an interesting problem that hasn’t been resolved.”
Also, Daydream needs to fix this twice— first by interpreting what the customer says, then by matching their sometimes original criteria with the products appearing in the catalog. With entries like I need a revenge dress for a bat mitzvah my ex is attending with his new wife, this understanding is essential. “At Daydream, we have this notion of a buyer’s vocabulary and a merchant’s vocabulary, right? Bornstein said. “Marketers talk in categories and attributes, and buyers say things like, ‘I’m going to this event, it’s going to be on the rooftop, and I’m going to be with my boyfriend.’ How do you actually merge these two vocabularies into something at runtime? And sometimes a conversation requires multiple iterations. Daydream has learned that language is not enough. “We use visual models, which allows us to understand products in a much more nuanced way,” she explains. A customer can share a specific color or show off a necklace they will wear.
Bornstein says Daydream’s subsequent redesign produced better results. (But when I tried it on, a request for black tuxedo pants showed me beige athletic-cut pants in addition to what I requested. Hey, it’s a beta.) “We ended up deciding to go from a single call to a set of multiple styles,” Bornstein says. “Everyone makes a specialty call. We have one for color, one for fabric, one for season, one for location.” For example, Daydream found that, for its purposes, OpenAI models are very effective at understanding the world from a clothing perspective. Google’s Gemini is less so, but it is fast and accurate.
