Claude Code began as a bet in Anthropic's Labs incubator

Anthropic's Labs team puts its own success rate at 20% to 30%, cofounder Ben Mann said: most of what the roughly 20-person group starts is meant to fail. Mann described the team's cycles, its hits and its post-IPO role in an interview with Business Insider.

At a glance

• Labs works in cycles, bringing prototypes it calls bets to a "persevere or pivot" review about every two weeks, with weak ideas killed or merged and their staff reassigned.
• The team was created in 2024 and has produced Claude Code, the Model Context Protocol and this year's Claude Design; ideas that survive graduate into their own teams once they pass four people.
• Anthropic is preparing to go public, and Labs is the team tasked with turning frontier models into products, work that now puts Claude Design against design software from Adobe and Figma.

The structure reads as an answer to a problem specific to a frontier lab: product teams that learn what a model can do only after it ships are late by months. Labs sits close enough to research to start building on a capability before it is public, which is likely why Claude Code existed as an internal prototype rather than a roadmap item. The cost is constant churn, and a majority of bets written off in advance.

Mann created Labs in 2024 after arguing Anthropic needed to ship products

A few years ago, Mann said, he had to convince Anthropic that it should release products at all rather than ride philanthropic funding toward advanced AI. He created Labs in 2024 as a place to ideate freely on new product ideas, and the group has since become one of the company's growth engines.

The model is not new. Mann cited Bell Labs as a reference point and worked in 2018 on Area 120, Google's own internal incubator; Google's moonshot unit, now called X, helped build Waymo and Google Brain. Anthropic's version counts leaders including Instagram cofounder Mike Krieger, and runs a rotating cast of about 20 employees.

Claude Code began after Mann told Boris Cherny his idea was too small

Labs developers have access to what Mann called the "Anthropic milieu of what's going well in research." Well before Claude Code's development in late 2024, the team heard from researchers that new models were showing strength in agentic coding, and Mann assigned the work to Boris Cherny, then a new hire.

Cherny had originally proposed a code analysis tool, which Mann told him was not a big enough idea. His prototype became popular inside the company, Anthropic released it as a preview in February 2025, and the model releases that followed over the next year turned Claude Code into a widely used product that accelerated Anthropic's growth.

Cherny now leads Claude Code. Mann said the episode reflects a pattern: employees new to the company need to raise their level of ambition much higher, because a lot of what the agents can do is not yet known to anyone.

Labs graduates a project once it grows past four people

The team works in cycles, bringing prototypes it calls bets to a "persevere or pivot" review roughly every two weeks. Weak ideas are dropped or merged into other bets, sometimes surfacing later in other products such as Anthropic's Claude Chrome extension, and their staff move to the next task.

Mann put the hit rate at 20% to 30% and said Labs has had more hits than he expected, with the successful ones going far. Projects usually graduate into their own teams once they pass four people, which keeps the roster churning as members leave and new staffers arrive.

Mann frames Claude Code as part of a goal of expanding AI's "action space," the set of things strong models can do in the world. He said Labs has pushed researchers on other teams to improve audio models' handling of Amharic, and that building Claude Design drove research interest in visual outputs. Other product engineering teams now run their own bets groups.

Biology and clean energy storage next

Mann said he would like Labs to contribute to breakthroughs in biological research and clean energy storage, work he sees as closer to Anthropic's mission. How product development runs after the listing is still open: Forrester analyst Mike Gualtieri told Business Insider that Anthropic faces a high-wire act as its tools push into business software, describing the balance as a little dance. No timeline has been attached to the science projects.