One thing that helped shift my perspective was to use it for its intended purpose. I have it enabled on my code editor to use for auto-complete instead of traditional code parser or snippet library, it’s honestly very good at that, it still makes a few mistakes and suggests shitty code, but overall I think it mostly works and it’s easier to hit tab and have the full for loop or small function written and correct the variable access it got wrong when it does.

Another thing that has made it very useful to me was in situations where I need to write code using libraries or languages I’m not used to. Having a copilot or Claude tab opened and asking it how to do certain stuff is a lot faster than reading the documentation to figure out the API or syntax. If something doesn’t work you feed it the error and it usually spots the problem. This has made me a lot more productive with for example Jenkins, since it’s a different language from what I use for everything else, and to properly test it you have to commit the code and let the pipeline run, before LLMs this was a very tedious work of reading docs, stack overflow, extrapolating responses, etc. Now it’s still tedious work, but at least I have my first draft much quicker and can then deal with the hallucinations or obsolete APIs it told me to use.