I think whoever is responsible for the correct operation of a software system will still need to read something that precisely describes the behaviours that they care about.
Today, for most devs, thats the code. We usually don't need to look at compiled output, because the code is enough. We can't just look at the prompts, because they aren't precise enough.
An important difference is the code precisely expresses many behaviours of the program, and a correct compiler guarantees to respect those behaviours.
An AI prompt is not so precise, and an AI offers no such guarantees to respect the behaviours expressed.
This is why the primary artifact of the development process, which we review and version control, is still the code, and not the prompt.
That said, I do think there's a lot of value to be gained by recording and analysing the prompt/response loop behind the code that ends up in a codebase
Unfortunately the tests can't (usually) cover all behaviours.
And perhaps more importantly, they don't capture more abstract properties of the code like maintainability.
AI works best in well maintained code, but unless care is taken, the AI will (today, anyway) make the code less well maintained as it goes.
If the AI is allowed to introduce mess into the codebase faster than its ability to deal with the mess increases, the codebase will eventually run into a problem that it's hard to recover from.
Today, for most devs, thats the code. We usually don't need to look at compiled output, because the code is enough. We can't just look at the prompts, because they aren't precise enough.