I've been thinking a bit about vibe coding with trust-critical apps. My solution has been hand-code and test the parts where bugs would mislead users, and vibe code the rest. In my case that's been hand-coding backend calculation logic and vibe coding the UI and server (this is also the part I am least expert in). In practice this does wind up including a lot of little judgment calls at the interfaces.
In the end, my feeling is there needs to be transparency in how bulletproof-tested a product is. IMO even a calculator that might be wrong can be useful if it's the most convenient option and the user knows the risks (though to be clear, that is not the philosophy I am employing in my personal project).
In the end, my feeling is there needs to be transparency in how bulletproof-tested a product is. IMO even a calculator that might be wrong can be useful if it's the most convenient option and the user knows the risks (though to be clear, that is not the philosophy I am employing in my personal project).