J. Crew's New AI Stylist Runs on Anthropic's Claude Models

J. Crew launched a consumer-facing AI stylist called Style Advisor on Wednesday, trained on thousands of outfit pairings from the retailer's photo shoots and built on a collection of Anthropic's Claude models. The tool debuts on J. Crew's desktop website, with a mobile app version expected later.

For the first time in roughly four decades, the brand famous for cashmere cardigans and tweed jackets — staples worn by Michelle Obama, Demi Moore, and Meghan Markle — is letting software make styling calls that humans have always owned. J. Crew started as a catalog before growing into a physical and omnichannel retailer.

"It's not just about the piece that you're looking for, but how to style it," says Danielle Schmelkin, chief information officer at J. Crew. "What those stylists bring to the table—and really recommend more than just the item that you may have picked up when you walked into the store—we are really looking to create that online."

Why retailers are betting on AI shopping now

The launch lands in a fast-crowding field. More than 50% of global consumers have used AI to buy goods within the past three months, per McKinsey, and new tools have proliferated — from Meta Platforms' Muse agent and Amazon's Alexa for Shopping to retailer-built assistants at Lowe's, Ulta Beauty, and Macy's.

But there's a gap worth watching: while 56% of consumers say they use AI chat or search tools like Claude or OpenAI's ChatGPT, only 27% gravitate to retailer-created shopping assistants, according to McKinsey. That adoption gap is the real competitive question J. Crew is trying to answer.

Schmelkin says the differentiator will be visual. "I think the visual piece is really important," she says of the edge she promises Style Advisor will offer. Down the line, the tool will pull in a customer's past purchase history to sharpen recommendations — replicating what already happens in the omnichannel experience.

The tool also works with context. Tell it you need a wedding outfit — in Miami, but not too formal — and it will factor those nuances into its picks.

How the model stack is wired

J. Crew didn't lean on a single model. Style Advisor runs on a variety of Claude models, each with a job:

• Fable 5.1 handles complex, ambiguous judgment — the part that lets the tool "think like a stylist"
• Sonnet 5.5 and Haiku 4.5 keep the back-and-forth responses fast and efficient
• Opus 5.5 was used by J. Crew's team to build the working prototype

"What is the actual experience that we want the shopper to have, and then, how do we think about making sure that we're mapping the right models to be able to deliver that," says Lauren Collett, Anthropic's enterprise go-to-market leader.

Cost control is built in. Schmelkin says J. Crew added a back-end observation layer to closely monitor spend. "We want to make sure we are constantly refining and using these models in the right way, both in terms of the outputs for our customers, but also the efficiency we want to see to make this scalable," she says.

What does this change for J. Crew's workforce?

Plenty, internally. The retailer runs an AI fluency program that starts with basics and advances to vibe coding. Its internal AI lab, "Second Thread," began with the tech team and now includes all corporate employees discussing AI use cases and proposing tech solutions. A two-day tech summit — running for four years, with workshops, hackathons, guest speakers, and vendor presentations — has been opened beyond tech staff. Employees can use enterprise tools from Anthropic and OpenAI.

Notably, Style Advisor wasn't a C-suite mandate. A J. Crew digital product designer conceptualized it and built a working prototype before engineers joined — a deliberate outcome of Schmelkin's push for employees to chart their own AI path.

"I've been screaming to anyone who will listen for years now to get in there and build a habit," says Schmelkin. "You have to find your a-ha moment and then all of a sudden…this could really change the way you think about your own life, work, and team productivity."

What does this change for practitioners? A template: pair model specialization with cost observability, prototype from the design bench rather than the boardroom — and remember that only about one in four shoppers currently trusts a retailer's AI enough to use it.