Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Julie Bornstein thought Her idea of would be easy to implement Artificial intelligence startup. Her biography is in Digital commerce Impeccable: VP of e-commerce at Nordstrom, COO of startup Stitch Fix, and founder of the personalized shopping platform acquired by Pinterest. Fashion has been her obsession since she was a Syracuse high schooler sniffing out products at Seventeen and hanging out at local malls. So she felt well-placed to create a company for clients to discover the perfect clothes using AI.
The reality was much more difficult than I expected. I recently had breakfast with Bornstein and her CTO, Maria Belousova, to learn about her startup. DaydreamingWith $50 million in funding from venture capital firms such as Google Ventures. The conversation took an unexpected turn when the women taught me about the amazing difficulty of translating the magic of AI systems into something people will actually find useful.
Her story helps explain something. My first newsletter of 2025 declared it would be so General application of artificial intelligence. Although many of these applications already exist, they have not changed the world as much as I expected. Since ChatGPT launched in late 2022, people have been amazed by the tricks AI performs, but study after study has shown that the technology has yet to achieve a significant increase in productivity. (One exception: encoding.) a A study published in August It found that 19 out of 20 enterprise AI pilot projects did not deliver any measurable value. I think increased productivity is on the horizon, but it’s taking longer than people expected. Hearing the stories of startups like Daydream seeking breakthroughs gives some hope that persistence and patience may actually lead to these breakthroughs.
Fashion designer fails
Bornstein’s original pitch to venture capitalists seemed straightforward: Use artificial intelligence to solve difficult fashion problems by matching customers with the perfect clothes, which they would be happy to pay for. (Daydream may take a while to implement.) You might think setup would be simple, just call the API of a form like ChatGPT and you’re good to go, right? Or not. Signing up over 265 partners, with access to over 2 million products from boutique stores to retail giants, was the easy part. It turns out that fulfilling a simple request like “I need a dress for a wedding in Paris” is very complicated. Are you the bride, mother-in-law or guest? What is the season? How formal is a wedding? What statement do you want to make? Even when these questions are resolved, different AI models have different views on such matters. “What we found is that due to a lack of consistency and reliability in the model — and hallucinations — sometimes the model might drop one or two items from the queries,” says Bornstein. A user in the extended beta of Daydream might say something like: “I’m a rectangle, but I need a dress that makes me look like an hourglass.” The model will respond by showing off dresses with geometric patterns.
Ultimately, Bornstein realized she had to do two things: postpone the app’s planned fall 2024 launch (although it’s available now, Daydream is still technically in beta until sometime in 2026) and upgrade her technical team. In December 2024, it hired Belousova, former CTO of Grubhub, who in turn hired a team of top engineers. Daydream’s secret weapon in the fierce talent war is the opportunity to work on a great problem. “Fashion is an exciting space because it contains flair, customization and visual data,” says Belousova. “It’s an interesting and unsolved problem.”
Moreover, Daydream has to solve this problem twice– First by interpreting what the customer says and then by matching his sometimes strange criteria with the goods on the side of the catalogue. With inputs like I need a revenge dress for a coming-of-age party my ex-husband is attending with his new wife. This understanding is crucial. “At Daydream we have the idea of shopper vocabulary and merchant vocabulary, right?” says Bornstein. “Merchants talk in terms of categories and adjectives, and shoppers say things like: ‘I’m going to this event, and it’s going to be on the roof, and I’m going to be with my friend.’” How do you actually combine those two idioms into something at runtime? And sometimes it takes multiple iterations in conversation. Daydreaming has learned that language isn’t enough. “We use visual models, so we actually understand the products in a more precise way,” she says. A customer might share a specific color or show what necklace they’ll be wearing.
Daydream’s subsequent restructuring has yielded better results, Bornstein says. (Although when I tried it, the order for black tuxedo pants and beige sweatpants showed me in addition to what I ordered. Hey, it’s a beta.) “We ended up deciding to go from one call to a group of multiple models,” says Bornstein. “Each one makes a specialty call. We have one for color, one for fabric, one for season, and one for location.” For example, Daydream has found that OpenAI models, for their own purposes, are really good at understanding the world from a clothing perspective. Google’s Gemini is less so, but it’s fast and accurate.