While this might be true, there’s a big difference in using LLMs for auto-completions, second opinion PR reviews, and maybe mocking up some tests than using it to write actual production code. I don’t see LLMs going away as a completion engine because they’re really good at that, but I suspect companies that are using it to write production code are realizing/will soon realize that they might have security issues and that for a human to work on that codebase it would likely have to be thrown away entirely and redone, so using slop it only costed them time and money without any benefits. But we’ll see how that goes, luckily I work at a company where managers used to be programmers so there’s not much push for us to use it to generate code.
While this might be true, there’s a big difference in using LLMs for auto-completions, second opinion PR reviews, and maybe mocking up some tests than using it to write actual production code. I don’t see LLMs going away as a completion engine because they’re really good at that, but I suspect companies that are using it to write production code are realizing/will soon realize that they might have security issues and that for a human to work on that codebase it would likely have to be thrown away entirely and redone, so using slop it only costed them time and money without any benefits. But we’ll see how that goes, luckily I work at a company where managers used to be programmers so there’s not much push for us to use it to generate code.