Anthropic’s Small Labs Team Rapidly Dismisses Bad Ideas and Restarts Innovation

Anthropic’s internal Labs group, a rotating roster of roughly twenty engineers, is tasked with spawning brand‑new AI product concepts.

The unit was launched in 2024 by co‑founder Ben Mann to give the company a dedicated space for rapid ideation separate from its core research stream.

Mann says the team operates like a true startup incubator, accepting that most experiments will not survive.

“We do treat it as a startup incubator in the true sense of the term,” Mann said, “and we expect most of the bets to fail.”

Labs cycles run on a two‑week cadence, ending with a “persevere or pivot” review that decides whether a prototype moves forward or is discarded.

Labs as an Internal Incubator

During each cycle, prototypes—referred to as “bets”—are built, tested, and judged against emerging research breakthroughs.

Bad ideas are either merged into other concepts or abandoned, allowing engineers to immediately shift to the next challenge.

Successful bets “graduate” out of Labs and become standalone products within Anthropic’s broader portfolio.

According to Mann, the team’s success rate hovers between twenty and thirty percent, a figure he finds higher than expected.

“We've had more hits than I would have expected, and the hits that we've had have gone really far,” he added.

Notable alumni of the program include Instagram co‑founder Mike Krieger, who contributed to early prototypes.

From Prototype to Product: Claude Code

One of Labs’ most celebrated outcomes is Claude Code, an AI‑driven coding assistant that writes, edits, and runs software.

The concept originated when researchers reported that new Anthropic models showed strong “agentic coding” capabilities in late 2024.

Ben Mann assigned the idea to new hire Boris Cherny, who initially proposed a modest code‑analysis tool.

Mann urged Cherny to aim higher, noting that newcomers need to elevate their ambition.

Cherny’s prototype quickly gained internal traction and was released as a preview in February 2025.

Subsequent model upgrades throughout 2025 amplified Claude Code’s performance, turning it into an “uber‑popular” product that fuels Anthropic’s growth engine.

Claude Code now exists as a standalone team and has even spawned extensions such as the Claude Chrome add‑on.

Impact on Anthropic’s Growth and IPO

Anthropic’s Labs team not only accelerates product pipelines but also gives the company a strategic edge by aligning development with the latest research breakthroughs.

Because Labs engineers have direct visibility into frontier model improvements, they can prototype tools that capitalize on new capabilities faster than competitors.

This rapid feedback loop helped Anthropic assemble a portfolio of market‑ready offerings ahead of its planned public listing in 2026.

The success of Claude Code and related tools demonstrates how a small, agile unit can drive significant revenue streams for an AI‑focused firm.

As Anthropic approaches its IPO, the Labs model is expected to remain a core engine for turning experimental AI advances into commercial products.

Why This Matters: Anthropic’s Labs team shows that a lean, fast‑moving internal incubator can turn cutting‑edge AI research into market‑ready tools, bolstering the company’s growth ahead of its public debut.