The Materialist : Lisa Yamner

Can a machine have taste? Everyone wants to ask Lisa Yamner about taste. She told me so, a little wearily, about four minutes into our conversation. It makes sense. She ran fashion at Google, co-founded The Yes (an early, ahead-of-its-time bet on personalized shopping that Pinterest bought in 2022), and is now building Daydream, an AI shopping app. If anyone has a view on whether software can know what you should want, it’s her. Her answer is no, and her reasoning is the thread that runs through this whole episode. Taste, she says, is built from everything you’ve seen, heard, and felt, plus a projection of who you want to be. A machine can do a lot of tasks for us, some better than we can. What it can’t do is feel, and taste is grounded in feeling. I pushed on this from a few angles, because I’m not sure the line is as clean as it sounds. Where does style come from, if not from the same data a model could ingest? Her honest answer: everywhere. Your mother, a stranger on the street, a shop window, an influencer you half-watched. Which is exactly the kind of diffuse, unlabeled input that models are getting better at every quarter. We didn’t resolve it. I’m not sure it resolves. The Cashmere Mafia My favorite moment in the episode is one where I accidentally gave the right answer to a question I didn’t know was a test. In 2012, Lisa was asked to build a fashion vertical at Google. The context: LVMH was suing Google over trademarks and gray-market goods. Chanel had never posted a video online. Google was selling Hermès with the same pitch it sold Home Depot, and the luxury houses, understandably, weren’t buying. Her job was to get them to run some ads. She asked me what I would have needed to hear, back when I was selling jewelry in 2012. I answered from the watch side of my life, where I spent years negotiating lineal feet of case space and store adjacencies with Swiss brands. Those fights were never really about square footage. They were about position. A luxury brand pays for Fifth Avenue because Fifth Avenue says something about who the brand is. So the pitch to a luxury house about search can’t be about clicks. It has to be about where you stand and who you stand next to. It turns out that’s exactly what her team, which called itself the Cashmere Mafia, built in 2012. They called it the Fifth Avenue narrative: search results as a street, your store open 24 hours, in good company. The lesson I took from it is bigger than Google. Luxury doesn’t resist technology. It resists being told the same story as everyone else. The technology was never the hard part. The translation was. Context, not personalization I’ve spent a lot of time thinking about push versus pull. Amazon perfected search for things you need: toilet paper, olive oil, a replacement charger. I have a need, and a multi-trillion-dollar machine delivers it with as little friction as possible. But almost everything At Present sells serves no function other than joy, meaning, and identity. Nobody searches for a need they don’t know they have. Lisa’s answer is that push is about to get a lot better, and the word she uses for it is context. Personalization, she says, was the word they always used at The Yes. But knowing who you are and what you like isn’t enough. The system also has to know why you’re shopping, where you’re going, and what’s happening in your life. Her example: an app that knows you’re headed to New York in three weeks and asks whether you want new cufflinks before you go. That is a very different kind of push from another email in your inbox. That’s the bet behind Daydream. Brands hand over their catalogs. Daydream enriches them, turning a black silk dress into something that’s appropriate for a wedding, black-tie adjacent, and right for a New York evening. Then it tries to match that against what it knows about you. The more you use it, the more it knows. The analogy she reached for is Claude learning your tone of voice the more emails it drafts. Two things she said here stuck with me. First, customers are less ready than the technologists assumed. People still type “sweater” into search bars. Nobody’s asking for natural-language shopping yet, and she thinks that’s on the builders to earn, not on the customers to adopt. Second, the surface will change. She doesn’t think Daydream will look like an app in a few years, and she said so without any apparent anxiety about it. I also asked whether any of this helps emerging designers, since the brands that dominated the last platform tend to dominate the next one. She was more hopeful than I expected. Barneys is gone, Matches is gone, Harvey Nichols is in trouble. The trusted places to find new designers are disappearing. If discovery can be rebuilt in software, the one-person brand spinning up a Shopify site after Parsons might actually find her customer. That’s the business I’m in, so I’m biased. But I think she’s right that the opportunity is real. The sweater in Milan Near the end, I asked Lisa what she’d bought most recently that no machine would have suggested. Her first answer was a bagel, on a Wednesday, which she conceded was off-pattern. Then she looked down at what she was wearing. Isabel Marant sweater. Bought in Milan this past July, with her mother and sister, on a trip for her mom’s 80th birthday. It has shoulder pads. She does not wear shoulder pads. She would have told any algorithm, confidently, that she does not wear shoulder pads. But it was on the rack, it was adorable, her mother and sister were there, and she needed something to remember the moment by. That’s the whole argument in one garment. The machine knows what you’ve bought. It doesn’t know you were in Milan with the two people who’ve shaped your taste more than anyone, in a mood to be talked into something. Lisa’s own read is that most of what she buys, the machine would have gotten to eventually. The sweater is the exception. But the exception is where the joy was. She said one other thing I keep thinking about: the only truly out-of-style thing is being uncomfortable in what you’re wearing. That’s a definition of taste that no model can optimize for, because only you know when it’s true. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit atpresent.substack.com