Boris Cherny, creator and Head of Claude Code at Anthropic, ships 20 to 30 pull requests per day by running 5 parallel Claude instances across 5 separate terminal checkouts. He starts each in plan mode, iterates on the plan, then lets Claude one-shot the implementation. His prior work at Meta as a Principal Engineer included causal analysis proving clean codebases have a measurable, double-digit-percent impact on engineering productivity. That background is not incidental. It shaped how Claude Code was designed.
The episode, hosted on The Pragmatic Engineer, covers how Claude Code went from internal side project to core Anthropic tooling. Cherny details the mechanics: parallel agents, PR structure, deterministic review patterns, and context retrieval from large codebases. He also explains how Claude Cowork was built. The workflow details here are specific enough to change how you work tomorrow, not just how you think about AI coding tools.
The deeper argument in this conversation is about role compression. As code generation becomes accessible, the boundaries between engineering, product, and design are collapsing. Cherny argues the engineer's role shifts rather than shrinks, but the skills that matter are changing. The full episode on YouTube, Spotify, and Apple Podcasts is worth the time if you want to understand what that shift looks like from someone who is building the tool and using it at production scale every day.
[READ ORIGINAL →]