Anthropic Labs: discontinuous bets

Why AI is going vertical (again) | Dianne Penn (Anthropic)

Topic 4 of 18

Published Jul 26, 2026 · Clip 23:34–27:30 (3:56)

Diane explains Labs' thesis: identify and pull the thread on discontinuous, large bets that might not be on the core roadmap, then determine whether there's 'a there there' and what its 10x, 100x, or 1000x could be — yielding Claude Code, Skills, Claude Design, and MCP. The operating approach is being strongly opinionated about the theme or area while staying loosely held about the exact prototype, with a culture of experimentation where engineers are self-driven. Theses that don't work yet get revisited one to two model generations later, and prototypes are valuable learning even when nothing ships. Labs works because of culture: Ben Mann sets an incredible vision pushing people to think about the 10x-100x of ideas, pods are small (ideas sometimes start with one engineer), and the team deliberately selects for people passionate about zero-to-one work who accept that many bets get turned down or off.