2026: My current Vibe Coding process After a long time, I once again chose Claude Code as the main force for my Vibe Coding development, and updated the entire development process, recording it for everyone's reference. one ️⃣ Choose the main model and tools If you have the ability, please try to choose the highest quality model currently available, which can significantly reduce the hidden time and effort required for rework or fixing bugs in the future. Due to the suspension of my Claude account, I reluctantly chose GLM 4.7 as the main programming model. I chose Claude Code as the IDE tool. two ️⃣ Dual architecture mode of architect and programmer Choose an AI model as a system architect, product manager, or think tank, and discuss your product requirements and ideas with it. I have chosen Gemini 3.0 Pro. After communicating the product architecture, you tell it that you also have an AI IDE as the core development assistant, and ask the AI architect to tell you how to develop the software step by step. You don't need to communicate too much with this AI architect about the implementation details, keep the context window compact because you need it to help you complete the context awareness throughout the entire system development process. The task of this AI architect is: 1. Conceptualize the entire system 2. Split development process 3. Assign specific AI IDE PROMPT for each development process 4. Acceptance results, provide PROMPT for the next step three ️⃣ AI programmer This is a programmer who truly does dirty and tiring work (coding, coding). Send the PROMPT given to you by the AI architect to this AI programmer and supervise it to complete this step. Conduct code testing, identify issues, and fix them. Specific code implementation, document reading, test result feedback, etc. are all communicated and improved with this AI Coder. Send the test results to the AI architect for you to evaluate the completion status. If everything goes smoothly, move on to the next step. four ️⃣ Don't be afraid to start over from scratch Last December, I attempted to develop a MEME Auto LP system. Unfortunately, the entire system is increasingly deviating from my initial vision. The accumulation is difficult to reverse, the code of Shishan is layered. In the end, I chose to start anew, create a new folder, and start writing the entire system from scratch (following the dual architecture process mentioned earlier). Switching the development process and selecting Claude Code+GLM 4.7, the development went exceptionally smoothly this time. Instead of spending time trying to fix the Shishan code, it's better to have a more intelligent model help you start writing from scratch. five ️⃣ Testing, acceptance, and submission Be sure to configure Git management tools for your code repository. The GitHub private repository I have chosen, along with Antigravity/GitHub Desktop, automatically generates a commit message. After each step or function is completed, conduct testing. Test is correct, make sure to submit to Git. This way, you have the right to roll back and overwrite at any time if AI writes, deletes, or acts randomly. This is the most effective way to control AI from acting recklessly. Based on the above steps and methods, basically if you want to develop a slightly more complex project, it won't be too difficult. But if it's just a simple program, then any AI IDE tool can directly complete it. The above.