Julie Bornstein thought implementing her AI startup idea would be very basic. Her digital commerce resume is impeccable: VP 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 at Seventeen and hanging out at local malls. So she felt well-prepared to create a company where customers could discover the perfect clothing using artificial intelligence.
The reality turned out to be much more challenging than she expected. I recently had breakfast with Bornstein and her CTO, Maria Belousova, to learn about her startup, Dreamfinanced with $50 million from VC funds such as Google Ventures. The conversation took an unexpected turn when the women made me realize how surprisingly challenging it is to translate the magic of artificial intelligence systems into something that people actually find useful.T
Her story helps explain something. In my first newsletter of 2025, I announced that this would be the year of AI applications. While there are many such apps out there, they haven’t changed the world like I predicted. Since ChatGPT launched in tardy 2022, people have been impressed by AI’s tricks, but study after study shows that the technology has yet to deliver significant productivity gains. (One exception: encoding.) A study published in August found that 19 out of 20 enterprise AI pilots produced no measurable value. I really think that productivity gains are on the horizon, but it will take longer than expected. Hearing the stories of startups like Daydream striving to break through gives hope that persistence and patience can actually lead to these breakthroughs.
Fashionista failure
Bornstein’s original pitch to VCs seemed obvious: exploit artificial intelligence to solve challenging fashion problems by matching customers with the perfect clothing they’re willing to pay for. (Daydream would require cutting). You’d think setup would be plain – just connect to the API for a model like ChatGPT and you’re good to go, right? NO. Signing up over 265 partners, with access to over 2 million products, from boutiques to retail giants, was the basic part. It turns out that fulfilling even such a plain request as “I need a dress for a wedding in Paris” is extremely complicated. Are you a bride, mother-in-law or guest? What season is it? Like a formal wedding? What statement do you want to make? Even after these issues are resolved, different AI models have different views on such issues. “We found that due to the model’s lack of consistency and reliability—as well as hallucinations—sometimes the model missed one or two query items,” Bornstein says. A user taking part in Daydream’s long-term beta test would say something like, “I’m a rectangle, but I need a dress that makes 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 fall 2024 launch (though the app is already available, Daydream is technically still in beta until around 2026) and modernize its tech team. In December 2024, it hired Belousova, former CTO of Grubhub, who in turn recruited a team of top-class engineers. Daydream’s secret weapon in the fierce talent war is the chance to work on a fascinating problem. “Fashion is an extremely juicy space because it has taste, personalization and visual data,” says Belousova. “It’s an interesting problem that hasn’t been solved.”
Moreover, Daydream needs to solve this problem twice— first by interpreting what the customer says, and then by matching their sometimes quirky criteria to the items in the catalog. With inputs like I need a revenge dress for the bat mitzvah my ex is attending with his fresh wife. this understanding is crucial. “At Daydream we have this concept of a buyer’s vocabulary and a seller’s vocabulary, right?” says Bornstein. “Sellers use categories and attributes, and buyers say things like, ‘I’m going to this event that’s on the roof and I’ll be with my boyfriend.’ How do you actually combine those two vocabularies into something at runtime? And sometimes the conversation takes a few iterations. Daydream has learned that language isn’t enough. “We exploit visual models so we understand products in a much more detailed way,” he says. The customer can share a specific color or show the necklace they will be wearing.
Bornstein says the next Daydream upgrade produced better results. (Although when I tried it out, the request for black tuxedo pants showed me a beige sporty cut pants in addition to what I asked for. Hey, it’s a beta.) “We ended up deciding to go from a single call to a multi-style set,” Bornstein says. “Everyone makes a specialized call. We have one for color, one for material, one for season and one for location.” For example, Daydream found that OpenAI models were really good at understanding the world from a clothing perspective. Google’s Gemini is smaller, but it is swift and precise.