Julie Bornstein Her idea would be simple to execute for an AI startup. Her résumé in digital commerce It is a perfect fit: Vice President of Ecommerce for Nordstrom, CEO of Stitch Fix startup, and founder & co-founder of personalized shopping platform purchased by Pinterest. She has had a passion for fashion since her Syracuse high-school days, when she would devour spreads of Seventeen Magazine and hang out in the local shopping malls. She felt she was well positioned to start a business that would help customers find the right garments by using AI.
She was not prepared for the reality. Bornstein’s CTO Maria Belousova joined me for breakfast to discuss her startup. DaydreamThe project was funded with 50 million dollars from VCs, such as Google Ventures. As the conversation turned unexpectedly, the women educated me about the difficulty in translating AI magic into something that people find actually useful.
The story she tells helps to explain something. The first email I sent in 2025 said that the newsletter would be announcing a new date. The Year of the AI App. Although there are many of these apps, I don’t think they have transformed the world in the way that I thought. People have been amazed at the AI tricks since ChatGPT debuted in 2022. But study after research has revealed that AI has not led to an increase in productivity. (One exception: coding.) The A study published in August Nineteen of twenty AI enterprise pilots projects failed to deliver measurable benefits. The productivity increase is coming, although it will take longer than expected. The stories of Daydream and other startups that have been working hard to achieve their goals gives hope that patience and persistence can indeed lead to breakthroughs.
Fashionista Fail
Bornstein’s initial pitch to VCs was obvious. Use AI to solve difficult fashion problems, by matching the right garments with customers who would happily pay. Daydream takes a percentage. You’d think the setup would be simple—just connect to an API for a model like ChatGPT and you’re good to go, right? Um, no. The easy part was signing up more than 265 partners with over 2,000,000 products available from retail giants to boutique shops. Even a seemingly simple request, like “I need a dress for a wedding in Paris” This is an incredibly complicated process. What are you? Are you the guest, mother-inlaw or bride? It’s what season? What is the formality of a wedding ceremony? What is the statement you wish to make? Even after the questions have been answered, AI models still differ in their opinions. “What we found was, because of the lack of consistency and reliability of the model—and the hallucinations—sometimes the model would drop one or two elements of the queries,” says Bornstein. Daydream users in the long-term beta testing would have said something along these lines: “I’m a rectangle, but I need a dress to make me look like an hourglass.” A model with geometric prints would be the answer.
Bornstein finally realized she would have to postpone Daydream’s launch in fall 2024 (even though it’s available now, Daydream will technically be in beta mode until sometime in 2026), and upgrade her tech team. Belousova was hired in December of 2024. The former Grubhub Chief Technology Officer brought a group of highly qualified engineers with her. The chance to solve a challenging problem is Daydream’s secret weapon against the talent war. “Fashion is such a juicy space because it has taste and personalization and visual data,” Belousova. “It’s an interesting problem that hasn’t been solved.”
The Daydream team has also got to find a solution for this issue Double-click to learn more—first by interpreting what the customer says and then by matching their sometimes quirky criteria with the wares on the catalog side. Inputs such as Need a dress to get revenge for my ex’s new bride who is going to the bat mitzvah. It is important to understand. “We have this notion at Daydream of shopper vocabulary and a merchant vocabulary, right?” says Bornstein. “Merchants speak in categories and attributes, and shoppers 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 run time? And sometimes it takes several iterations in a conversation.” The Daydream team learned that just knowing the language wasn’t sufficient. “We’re using visual models, so we actually understand the products in a much more nuanced way,” She says. The customer may share the color of a particular necklace or a certain shade.
Bornstein claims that Daydream has improved since its re-design. Although when I used it, my request for black athletic pants was met with beige athletic trousers. It’s beta. “We ended up deciding to move from a single call to an ensemble of many models,” says Bornstein. “Each one makes a specialized call. We have one for color, one for fabric, one for season, one for location.” Daydream for instance has determined that OpenAI is very good at understanding clothing from a user’s perspective. Google Gemini’s accuracy isn’t as good, but its speed is.
