If you want to do some non-eugenics, fascist-free AI coding, try Zed with GitHub Copilot. I’ve been using it this week with better results than I ever had with Cursor. There’s even a low token “MAI-Code-1-Flash” model which has been giving me better results than any Composer model I used to use and the tokens used seem to be way less.

I asked it today to fix a non-simple bug and MAI fixed it in one shot with less than 70k tokens (Cursor would have used probably half a million tokens based on my previous usage). Orgs need to start getting more visibility into why Cursor burns so many tokens.