A funny story about a compiler transpiles C to assembly. Nobody on the team reads the assembly. We trust the translation completely, ship the output, and think entirely in C.
Assembly to me is still something junior folks need to learn e.g., read, to understand how their code actually executes as instructions pre-defined by the CPU architecture. Trusting the translation is not the same as ignoring the layer below.
AI coding tools make a similar move. They transpile plain natual language into a programming language. You describe what you want and the tool produces the code
So is a prompt the new C? Not yet. The comparison is useful because of where it breaks.
A compiler is deterministic. The same C produces the same assembly every time, and decades of testing stand behind the translation. That is why nobody reviews compiler output.
AI is probabilistic. The same prompt can produce different code on different days. There is no guarantee behind the translation, so we still read every LOC it produces.
That difference decides where the source of truth lives. In the compiler world, the C is the source of truth and the assembly is an artifact. In the AI world, the prompt is not the source of truth. The generated code is. It is what you review, test, own, and maintain.
Maybe that changes one day, if prompt-to-code translation becomes reliable enough that nobody reads the output. Until then, I treat generated code the way I treat any code I am responsible for: as source, not as artifact.