The economics of AI coding assistants

This is my first substack post. It's about the internals of Cursor and how it makes a tradeoff between token usage and code sent to the LLM. Feedback is welcome.