how i actually use AI when i code

i used autocomplete for a while. sometimes it helped. mostly it got in the way.

when i'm trying to think, constant suggestions are noise. i end up checking the AI instead of solving the problem.

what worked better for me was giving it a full small task, not the next line.

i say what i want, with rules. like: change this file to use this interface. don't change behavior. tests must still pass. then i read the diff and decide.

that works. vague prompts don't. "make this better" usually gives me something that looks fine and is slightly wrong.

i still check everything. the model doesn't know my codebase history. it doesn't know what already broke once.

my loop is simple. say the task clearly. get a change. review it. test it. then merge.

same standards. just a faster first draft when i use it right.