Anthropic's approach to building products has a secret weapon: a small, restless team that operates like a startup inside the AI company. As the firm prepares for its public offering, the Labs group is quietly responsible for some of its most significant developer tools.
Inside the team behind Claude Code
Labs is a rotating cast of roughly 20 employees, led by Anthropic cofounder Ben Mann. The team's mandate is simple: spin up entirely new ideas. Since its creation in 2024, it has produced the definitive AI coding tool of the year, Claude Code, along with the Model Context Protocol for connecting AI agents to external data sources and the newer Claude Design tool.
The group operates like an internal startup incubator. Mann says the team expects most of its bets to fail, citing a success rate between 20% and 30%. Ideas that gain traction "graduate" out of Labs into their own dedicated teams within Anthropic. Projects that don't work out get tossed or merged into other prototypes, and their employees jump to the next task.
The team runs on a tight cycle, reviewing whether to persevere or pivot on each prototype about every two weeks. Because projects typically graduate when they grow beyond four people, the group experiences constant churn as members cycle in and out.
How frontier research awareness shaped Claude Code
Labs' core advantage is proximity to Anthropic's research teams. Developers on the team hear early signals about which new model capabilities are showing promise, giving them a head start on building tools that leverage those strengths.
That's exactly how Claude Code came to be. In late 2024, researchers told Mann's team that new models were showing strength in agentic coding, the ability for AI to write, edit, and run code with minimal human intervention. Mann handed the task to Boris Cherny, a relatively new employee at the time. Cherny suggested building a code analysis tool instead, but Mann told him it wasn't ambitious enough.
"Especially when people are new to the company, they need to change their level of ambition to much higher," Mann said. "Because a lot of what the agents can do, nobody knows."
Cherny built a prototype that became popular internally, and Anthropic released it as a preview in February 2025. The company's model releases over the following year turned Claude Code into a widely used product, driving Anthropic's growth. Cherny now leads the Claude Code team.
Pushing AI's "action space" into new domains
Mann frames Labs' work in terms of growing AI's "action space," the set of things that strong models can actually do in the world. Claude Code expanded that space significantly for developers, but the team is thinking bigger.
Labs has pushed Anthropic researchers to improve audio models for understanding Amharic, an Ethiopian language, and building Claude Design has driven research interest in improving visual outputs. The team draws inspiration from Bell Labs and Mann's own experience on Google's Area 120, a similar internal incubator. Google's moonshot factory, now called X, also helped build Waymo and Google Brain before those projects scaled into larger divisions.
Labs' ethos has spread beyond its own walls. Other product engineering teams at Anthropic now run their own "bets" groups, adopting the same rapid-iteration model.
The post-IPO balancing act
As Anthropic prepares to go public, Labs will play a crucial role in converting frontier AI models into products that drive revenue. That creates a delicate dynamic. Mike Gualtieri, an AI analyst at Forrester, noted that Anthropic faces a high-wire act as it builds tools that compete directly with software used by its own business customers. Claude Design, for example, competes with design tools from companies like Adobe and Figma.
"As they build the platform, they're going to likely create features that make it more attractive to use their platform than someone else's," Gualtieri said. "It's going to be a little dance."
Mann says he wants Labs to eventually contribute to breakthroughs in biological research and clean energy storage, applications that would connect the team's work more directly to Anthropic's mission of helping humanity through stronger AI systems. "The ability to accelerate the hard sciences is extremely exciting and very, very important for sort of starting to deliver more benefits to people," he said.
For now, the Labs team remains a small engine of constant experimentation inside a company racing toward the public markets, where its ability to turn research advantages into products will be closely watched.