Post #4307273

@jadehopepunk@hachyderm.io From a criticism standpoint, I would say that the two uses of AI you give, refactoring and builds/installs, don't make sense to me. What I would call refactoring is a very careful task, and anything that is repetitive can and usually has been automated without AI. Same goes for builds/installs, there are all sorts of automated build systems. Specifically what are you automating there? AI is like a train that works on 90% of its track, and you cannot know in advance where the 10% is going to be. LLMs are trained to produce code that looks correct, so they usually do, but will always sometimes produce code that is wrong in ways an inexperienced developer could only dream of. So every line of AI touched code, even something as innocuous as renamed variables, has to be combed through for errors a human wouldn't know how to make. Also the deskilling isn't just not getting enough practice, its a consequence of cognitive offloading and the addictive nature of AI, people thinking it helps when it actually doesn't is well documented. There was a study on both junior and senior devs that found they reported being ~20% faster with AI, but were actually ~20% slower, because they were iterating quickly, but doing less. I don't use AI because I know if I actually use it to do anything without doing the amount of extra work you need to do to check, it will change how I think and wear down my critical thinking.