Recently, I feel that Cursor/Windsurf is not as user-friendly, so I installed Gemini CLI and Claude Code and found the optimal path for AI programming: 1 Explore solutions using ChatGPT, identify feasible paths, and copy and save them to the product requirements document; 2. Have cursor/Windsurf build the project framework based on the document; 3. Use Gemini to read the open-source code repository and official documentation used in the project, re evaluate and update the product requirements document; 4. Use Claude Code to write code; 5. Conduct code review using Gemini. 3.4.5 It is possible to directly use ZEN MCP to achieve contextual collaboration among multiple large models. Overall, using chatbot tools with unlimited monthly usage for creative and conceptual discussions (primarily to save API credits); Build project framework using AI code editor; Use Gemini's ability to read and audit lengthy documents and code repositories with ultra long context; Programming with Claude Code