I have also stepped back from LLM coding a bit. I still utilize LLMs for API discussions, maybe 1-off things I would stackoverflow before but I have stepped bac

I have also stepped back from LLM coding a bit. I still utilize LLMs for API discussions, maybe 1-off things I would stackoverflow before but I have stepped back a bit from autocomplete everywhere and even agentic flows. And I feel more productive. I recommend that everyone gives it a try. As a tenured SW developer in my company my measurements for success are much more than "how much code can I spit out". There are mentoring, refactoring, code readability/mantainability, and quality that are important to my job. I found that LLM generated code was not hitting the necessary bars for me in these areas (agent or code autocompletion) and so I have stepped back from them. The readability point is extra important to me. Having maintained million lines of code products, I have found that readability is more important than writing a ton of code: and LLMs just don't hit the bar here. When I am playing with side projects that I don't have the same bar on, sure Ill have bolt or lovable generate me some code in combination with cursor or windsurf, but these are low stakes and in some ways I just want to get something on paper.