Descriptions:
Mike Krieger, co-founder of Instagram and now a member of technical staff at Anthropic, took the stage at the AI Engineer conference to discuss how building with large language models has fundamentally reshaped his engineering practice — and how Anthropic’s internal labs team is structured to move fast without constant reorgs.
Krieger describes shifting from Anthropic’s Chief Product Officer role to hands-on builder after experiencing mounting FOMO watching others code directly with Claude. He outlines the transition from task-delegation prompting to goal-state prompting — describing the desired end state and letting the model, particularly Claude Code and the newly released Fable, surface trade-offs along the way. He also details an unconventional move: porting a Python project to TypeScript mid-development because Claude Code had independently arrived at a better deployment story using Bun, a decision he says would have been unthinkable under traditional engineering norms.
On the organizational side, Krieger explains how Anthropic Labs runs on two-week “persevere or pivot” cycles, where every active project is reviewed and can be wound down without stigma. Rather than aligning the team structure to individual bets — which would require constant reorgs — Anthropic draws cross-functional pods fluidly across product, engineering, and research. He also covers how code review itself is evolving: instead of line-by-line diffs, his team increasingly shares Claude-generated artifacts that document intent and trade-offs, with human review focused on what the code was trying to accomplish rather than how it was written.
📺 Source: AI Engineer · Published August 27, 2026
🏷️ Format: Interview
