Anthropic Team's Claude Code Use Cases

Anthropic's internal teams broadly use Claude Code for debugging, understanding codebases, writing tests, prototyping, documentation, and automation.

• New data scientists onboard quickly by feeding in the entire codebase and CLAUDE.md, and the product engineering team uses it as the first context-gathering tool before bug fixes and feature work.
• For testing and review, the product design team creates tests for new features and automates PR comments and format fixes with GitHub Actions. The security engineering team produces more reliable code with pseudocode and TDD, and the inference team even ports tests to unfamiliar languages like Rust using Claude.
• For debugging, the security engineering team feeds in stack traces and documentation to trace incidents, processing tasks that used to take 10-15 minutes of manual scanning 3x faster. The data infrastructure team identified Kubernetes pod IP exhaustion and was guided through adding a new IP pool, saving 20 minutes.
• The product design team runs autonomous loops based on Figma files to write feature code and iterate on tests, and data scientists build React visualization apps without knowing TypeScript.
• Documentation uses MCP and CLAUDE.md to turn scattered knowledge into runbooks and troubleshooting guides, reducing search time for the inference team, who have no ML background, from 1 hour to 10-20 minutes.
• The growth marketing team analyzes hundreds of ad CSVs to replace underperforming creative, and generates hundreds of variants in minutes using two sub-agents. A Figma plugin generates up to 100 ad variants at roughly 0.5 seconds per batch. The legal team also built a phone tree prototype for internal connections.

The key takeaway is that Claude Code is most effective when used as a thinking partner that augments human workflows rather than just a code generator.