The internet has made vintage clothing abundant in the manner of a warehouse after the lights go out. Everything may be in there. Good luck finding the right thing. A shopper can spend an evening opening tabs, translating vague desires into keywords, squinting at photographs, and wondering whether “excellent condition” is an assessment or a prayer. Sandra Lifshits looked at this peculiar form of leisure and saw unpaid labor. Her company, Retrofit, proposes an editor: AI agents that search broad pools of secondhand inventory and return with a smaller, more personal rail of clothes.
The idea did not arrive as a clever application of a fashionable technology. It grew out of a less glamorous sequence: countless weekends at vintage markets, conversations with hundreds of sellers, and volunteer shifts inside several New York vintage shops. Lifshits wanted to understand how stores operated and how shoppers behaved when the merchandise was singular, inconsistent, and full of tiny judgments. Vintage is an unusually human category. A stain can be fatal or forgivable. A label can matter less than a seam. The best piece in the room is often the one nobody thought to describe properly.
“Our goal is to make vintage fashion as accessible, fun, and normal as buying new.”Sandra Lifshits
A field study dressed as shopping
Lifshits brought a product leader's habits to the racks. Before Retrofit, she had led product work at two ecommerce startups. She had also attended Brown University, then entered Stanford Graduate School of Business in 2022. At Stanford, the company moved from an idea into the Venture Studio, a cross-campus program where student teams work through new ventures with mentors, peers, and alumni. Retrofit appeared at the studio's 2024 Demo Day. The progression is tidy in retrospect: ecommerce experience, business school, venture studio, company. The useful part is much messier. She kept returning to shops and sellers until the complaint had texture.
Online secondhand shopping, she and future co-founder Maddy Yip concluded, had failed to keep pace with demand. The supply was scattered. Listing quality varied. Personalization was thin. Shopping asked customers to become taxonomists, authenticators, condition graders, and professional web hunters. The pair met while Lifshits was completing her Stanford MBA. Yip brought computer vision experience from Google and Stanford computer science, including work that led to two patents. Lifshits brought product judgment and the field notes. In 2024 they founded Retrofit, with Lifshits as CEO and Yip as CTO.
The product is the edit
Retrofit's machinery begins where ordinary filtering becomes tiresome. Its agents sift through hundreds of thousands of listings from vintage sellers across the country. They evaluate the signals a practiced shopper might juggle at once: current trends, sales patterns, brand, price, materials, condition, size, and an individual's style. A customer can share a Pinterest board or other inspiration images. The marketplace then tries to present the pieces that deserve attention, not merely the pieces that match a literal string.
This is a subtle change in who carries the burden of expertise. Conventional marketplaces hand the shopper a search bar and a formidable pile of results. Retrofit attempts to absorb more of the interpretive work. When the current inventory does not contain the right item, the shopper can submit a request and ask the agents to keep looking. That feature is almost comically modest. It is also revealing. Instead of pretending every possible desire is already inside the catalog, the company turns absence into an instruction.
Even the name carries the thesis neatly. To retrofit something is to work with what already exists, adding new capability without discarding the useful structure underneath. That is precisely the secondhand proposition: the garment has been made, worn, photographed, and placed somewhere on the web. The problem is not producing another object. It is giving an existing object a better route to its next owner. The marketplace applies new searching and presentation machinery to old clothes, a rare case of startup naming that earns its keep.
There is also a seller-side problem hiding behind every polished shopping grid. Independent vintage stores often hold remarkable inventory but lack the time to standardize every photograph, measurement, description, and tag for online discovery. Lifshits and Yip designed their agents to bring inventory from stores nationwide into detailed, visually consistent listings. The goal is not only to help a shopper find a coat. It is to help the coat become findable in the first place. One side of the marketplace supplies taste and objects; the other supplies desire. The software must translate between them without making every shop look interchangeable.
The distinction matters because style does not speak fluent database. People say “a jacket like the one in that film,” “something severe but not corporate,” or “the boots I should have kept from 2007.” Fashion vocabulary is wonderfully imprecise. A system that merely labels dresses and trousers can organize a closet, but organization is not taste. Yip has described Retrofit's ambition as building AI with fashion taste, trained to understand what makes a vintage piece special rather than simply which category contains it. Lifshits's role is to turn that technical possibility into a shopping experience a person will actually trust.
Trust is the hard currency here. A recommendation can be technically plausible and still feel clueless. A vintage shopper notices when a system confuses costume with character, or trendiness with personal style. Retrofit therefore has to learn from both broad signals and intimate constraints. The sales data may say a silhouette is moving. The shopper's board may say it belongs nowhere near her closet. Product judgment lives in resolving that argument gracefully. Lifshits's background makes this less a machine-learning spectacle than a familiar product task: listen closely, rank competing needs, and make the interface carry the complexity.
Keeping the thrill, losing the trawl
There is a danger in making vintage too efficient. The category runs on surprise. People enjoy finding the improbable coat that had no business waiting in a town three states away. Reduce the experience to a perfectly optimized funnel and the pleasure can evaporate. Retrofit's more interesting wager is that personalization can protect serendipity by removing the wrong surprises: the broken zipper, the hopeless size, the material a shopper dislikes, the ninety pages of near misses. A good edit does not tell someone what to want. It creates enough quiet for wanting to become legible.
That tension gives Lifshits's stated aspiration its bite. She wants vintage to feel as accessible, fun, and normal as buying new. “Normal” does not mean ordinary. It means reliable enough to enter the weekly rhythm of commerce, without requiring a free Saturday and the patience of an archivist. The company wants to preserve what fast fashion cannot easily offer: singularity, history, and the agreeable risk that somebody else once had better taste than you.
Retrofit entered Y Combinator's Winter 2025 batch as a two-person company in New York City. Its public launch placed the founders beside a cohort crowded with AI infrastructure and business software. This one dealt in leather jackets, skirts, bags, and the stubbornly subjective question of what looks good. TechCrunch later included Retrofit among ten startups to watch from the batch's Demo Day. Lifshits celebrated the recognition plainly on LinkedIn, but the more revealing launch note thanked the people who had supported the company and invited both vintage devotees and the “vintage-curious” to try it. The phrase is good product positioning in miniature. It welcomes expertise without requiring it.
The old lesson inside the new machine
For founders, the stealable idea is not to sprinkle AI over a marketplace. It is to study where customers are doing invisible work. Lifshits found people performing hours of sorting because existing products treated abundance as the same thing as usefulness. She did the unscalable work first: shop floors, seller conversations, repeated market visits. Only then did the team ask agents to carry part of the hunt. The order matters. Automating a poorly understood task tends to produce the same confusion at machine speed.
There is an earlier clue in Lifshits's public record. In 2012, she coordinated the first TEDx conference at the Anglo-American School of Moscow. The job involved drafting the plan, leading fellow coordinators, recruiting speakers from the region, and training them in the TEDx presentation style. Long before Retrofit, she was already working on a version of curation: finding people, shaping presentation, and helping an audience see what deserved attention. It would be too neat to call this destiny. It is fair to call it a recurring preference for making unruly material intelligible.
Retrofit is still an early company. Its public facts are compact: two founders, one YC batch, a marketplace, and a large ambition. Yet the founding method is unusually concrete. Go where the behavior happens. Listen until the complaint becomes specific. Pair domain observation with technical range. Then design the edit. Lifshits is betting that the next useful layer of online shopping will not be another ocean of choice. It will be a discerning guide with a long memory, a quick eye, and no objection to spending all night on the web so the customer does not have to.