AI agents: write and checkout the plan (www.dein.fr)

In a significant advance for collaborative coding practices, Boris Cherny, the creator of Claude Code, advocates for initiating AI-assisted coding sessions in a "planning mode." This approach emphasizes the importance of developing a well-structured plan before implementation, which should be reviewed and merged into the repository as a separate pull request. This separation aligns with the classical research-design-implement loop, allowing teams to focus on architecture and technical choices before diving into coding, ultimately minimizing costly revisions later in the process. The proposal highlights a comprehensive outline for effective planning, including a clear implementation strategy, success criteria, and risk management. By documenting the planning phase within the code repository, teams not only preserve the rationale behind architectural decisions but also enable AI agents to reference these plans during development. This practice fosters better tracking of lessons learned and enhances future product development. By establishing a template and best practices for planning, coding agents can expedite workflows, paving the way for greater efficiency and improved onboarding processes for new engineers.