Gemini is my favourite Saas coder, but they lie. They claim 1 million context but the models become braindead beyond 200,000.
GPT 20B is my go-to local coder. It's definitely on par with gemini flash, but not as good as pro. Probably better coder than I am. I have max 120k context configured, but up around 60-80k it gets dumb.
What's also quite interesting, they are particularly bad at ursina and panda3d.
Now I realize it was asking whether or not you use AI to write code. English is fun.
(My answer: I don't use AI to help with writing code, even (ironically?) when I'm working on deep learning systems.)