My rule with Cursor is that unless I know exactly what I want it to do, I have it stay out of my way. If I have a general idea, I may turn on auto-complete. I reserve the agent for things like tests, UX, and rote programming that saves me time.
When I do use the agent, I inspect its output ruthlessly. The idea that pages of code can be written before being inspected is horrifying to me.
When I do use the agent, I inspect its output ruthlessly. The idea that pages of code can be written before being inspected is horrifying to me.