Some time circa late 1950s, a coder is given a problem and a compiler to solve it. The coder writes their solution in a high level language and asks the compile

Some time circa late 1950s, a coder is given a problem and a compiler to solve it. The coder writes their solution in a high level language and asks the compiler to generate the assembly code from it. The compiler: I cannot generate the assembly code for you, that would be completing your work ... /sarcasm On a more serious note: LLMs now are an early technology, much like the early compilers who many programmers didn't trust to generate optimized assembly code on par with hand-crafted assembly, and they had to check the compiler's output and tweak it if needed. It took a while until the art of compiler optimization was perfected to the point that we don't question what the compiler is doing, even if it generates sub-optimal machine code. The productivity gained from using a HLL vs. assembly was worth it. I can see LLMs progressing towards the same tradeoff in the near future. It will take time, but it will become the norm once enough trust is established in what they produce.