Those are fun thoughts. I've imagined a world where you can create a long, complex prompt that results in a reproducible response of code that does exactly wha

Those are fun thoughts. I've imagined a world where you can create a long, complex prompt that results in a reproducible response of code that does exactly what you want, and does it correctly. In that world, the prompt could be your versioned source and the build process could start from that. Implementing changes could be done by tweaking the prompt. Of course you'd still want to archive the "intermediate" source code, if only to protect yourself from an LLM outage. In that world, you actually would be getting new code from the source prompt regularly, and yes, it hopefully would improve over time. In our world however (at least for me), usage tends to be more conversational, so there really isn't even "a prompt" to save. Not to mention that I never use the output as-is.