Back in the previous century, I was working (mostly coding) for a small tech startup. The language we were using had been designed and implemented by one of the company’s principals, so when code refused to run properly, if we suspected a compiler bug, we could go straight to the creator. More often than not, as we were explaining the problem to Bob, our mistake (as was often the case) became obvious. Bob used to say that programmers need a monkey on their shoulder to whom they might explain problems, because the act of re-casting the problem for other eyes often revealed the flaws in our thinking.
I have no monkey and I have no Bob. I do still sometimes write buggy code. A few times, in desperation, I have asked an LLM chatbot to solve the coding problem. The chatbot code has always been wrong. And unusable. However, in a couple of the cases, just looking at the chatbot code has opened my eyes to what was wrong with my own code.
I think that LLM chatbots that write code will always write buggy code. In my direct experience, the current chatbots I have used have written abysmal code. Since Kernighan’s Law says that it will take longer to debug bad code than it took to write it, I suspect that LLMs will never replace programmers. Still, if all you need is a monkey on your shoulder, LLMs might just fill the need and thus become a valuable tool.
—2p