Comment by notepad0x90

maybe a dumb but wise approach is to just code as usual without thinking about "AI", and when you have difficulties or efficiency issues, look for tools to solve that. think in terms of specific tools instead of "ai" or "llm".

Do you need better auto-completion? Do you need code auto-generation? do you need test cases to be generated, and lots of them? maybe llms can are ideal for you, or not.

Personally, the best use i've gotten out of it so far is to replace the old pattern of googling something and clicking through a bunch of sites like stackoverflow to figure things out. and asking llms to generate an example code of how to do something, and using that as a reference to solve problems. sometimes i really just need the damn answer without having a deep debate with someone on the internet, and sometimes I need a holistic solution engineering. AI helps with either, but if I don't know what questions to ask to begin with, it will be forced to make assumptions, and then I can't validate the suggestions or code it generated based on those assumptions. So, it's very important to me that the questions I ask an AI tool are questions whose subject domain I have a good understanding of, and where the answers are things I can independently validate.