Skip to main content

On This Page

Inside Claude Code Creator Boris Cherny’s Development Workflow

1 min read
Share

These articles are AI-generated summaries. Please check the original sources for full details.

Claude Code Workflow at Anthropic

Anthropic’s Boris Cherny outlined his highly productive workflow with Claude Code, running up to 15 sessions in parallel on his MacBook and Anthropic’s servers. He avoids customization, finding the out-of-the-box performance excellent, and emphasizes iterative refinement over direct code editing.

Cherny’s approach focuses on compounding productivity through parallelization, shared learnings, automated prompting, and rigorous verification, addressing the reality that even powerful LLMs require careful oversight and quality control to avoid costly errors and integration issues.

Key Insights

  • Parallel Sessions: Cherny runs up to 15 Claude Code sessions concurrently to maximize throughput.
  • CLAUDE.md: Anthropic teams maintain a shared CLAUDE.md file in git, currently 2.5k tokens, to document learnings and best practices.
  • PostToolUse Hook: Automated code formatting via a PostToolUse hook prevents CI failures.

Working Example

"PostToolUse" : [
"matcher": "WritelEdit",
"hooks": [
{
"type": "command",
"command": "bun run format || true"
}
]
]

Practical Applications

  • Use Case: Anthropic engineers use /commit-push-pr slash commands dozens of times daily, automating common git workflows.
  • Pitfall: Relying solely on auto-editing without a clear plan can lead to suboptimal code and increased review time.

References:

Continue reading

Next article

Kubernetes Is Not a Container Platform (And That Changes Everything)

Related Content