Anthropic reveals Claude Code best practices

The move to formalize "agentic context" through files like CLAUDE.md marks a shift from generic prompting to persistent, repo-specific AI configuration.

• –The CLAUDE.md strategy creates a shared "handbook" for the agent, providing instant access to build commands and style guides without constant re-scanning.
• –A "Two-Phase" workflow improves success rates by forcing the agent to brainstorm and get approval for a plan before it touches a single file.
• –Agentic search (using grep/find) is positioned as a more scalable and cost-effective alternative to full-repo vector indexing for massive enterprise repositories.
• –The "Slot Machine" philosophy encourages a high-velocity, low-stakes approach where frequent git commits act as a safety net for autonomous sessions.