I find this vibecoding idea just to risky in a form it is presented now. Even if an app is finally working you still do not know if it is working as intended.
I am no coder. My approach is to work as a BI: discuss ideas, specify requirements, ask questions, review the code and ask more questions, ask agent to write tests and execute and then test yourself. In the process of that you will learn a lot, have fun and do stuff.
I started with Cursor but #goose is on my list.