Playing vibe coding should ideally solve a problem in a work scenario or a life scenario.

First, we need to organize this scenario into an SOP, and then based on this SOP, create a PRD development document. After that, we will continuously improve this document.

Secondly, the B model can be used to check the development documentation generated by the A model, cross-checking it several times, for example, using Claude to check GPT, and then checking it once more. As the teachers say, 70% of the time should be spent on early requirement sorting, while actual development only takes 30%. In the future, I estimate that the development time will be even shorter with the support of AI.

Finally, just throw the PRD document to the IDE to run it (of course, the basic work before running the code development is to configure the computer environment and install the corresponding software).

Basically, this is the idea I have learned so far. In fact, the first step of sorting out the SOP process is crucial because AI only understands logic. The more rigorous the logic, the smoother the development will be, and the fewer bugs and inexplicable issues there will be.