This is probably not going to shock anyone, but I generated some more Rust code with GPT-6-Luna today. I had a Rust function (well 2 functions) that was around 80 lines of code. I decided to write a spec for it, and then sent it to GPT-6-Luna for code generation.
GPT-6-Luna responded with a new function, and one that was only 30 lines of code.
• I reviewed the function; it seems GPT used early returns, which reduced at least one branch, and combined both functions into one.
• I tested it and it worked correctly.
• I would not be surprised if it was faster than my old code too, due to having one fewer branch.
It seems to me that spec-driven development may become a more prominent part of software development. It is likely the spec will become more important than the code. In any case, Luna did the code gen in 21 seconds for only $0.0011—it seems like a good model to do simpler code generations with.