Ask HN: Anyone struggling to get value out of coding LLMs?
While some engineers can't seem to stop raving about how much of a superpower large language models are for coding, others find it to be an extremely finicky process that requires an extreme amount of coddling to get average results at best. This could be because of the type of work the engineers are doing - AI seems to be productive for greenfield work, but it is pretty useless when it comes to maintaining existing systems, adding more complicated features, or needing to know business domain details. Some engineers find it much quicker and easier to just write code themselves as they always do a better job in the end.