
The Pragmatic Engineer Building Claude Code with Boris Cherny
1090 snips
Mar 4, 2026 Boris Cherny, creator and Head of Claude Code at Anthropic and author of Programming TypeScript, joins to talk about building agentic code tooling. He describes Claude Code’s evolution, parallel agents and deterministic review patterns. They explore retrieval from large codebases, sandboxing and safety layers, and how engineering roles shift as models write more code.
AI Snips
Chapters
Books
Transcript
Episode notes
Terminal Hack Grew Into Agentic Code Writer
- Early Claude Code began as a terminal chat bot that Boris gave a single bash tool and file-edit capability.
- The model surprised him by writing AppleScript to check his music and then evolved to use file and bash tools agentically.
Tool Adoption Grew Organically By Being Useful
- Claude Code reached near-universal adoption inside Anthropic by being useful and optional, not forced.
- Internal adoption climbed to ~100% of technical staff; Claude writes roughly 80% of Anthropic's code on average.
Run Multiple Agents With Parallel Worktrees
- Use parallel checkouts or Git worktrees to run multiple Claude Code agents concurrently.
- Boris keeps five terminal tabs (or the desktop app with worktree support) and round-robins agents to maximize throughput.







