Bookmark this

80% of the time I'm just going back and forth with Cursor. So me, the developer, and Cursor, the LLM native IDE. Back and forth, prompting the IDE, it writes code, I review the code, accept or reject it, run it locally, test it, grab errors or feedback, send it back to Cursor, back and forth. That's the feedback loop. 15% of the time I'm enlisting GPT 4.0. This is for when I want to do more complex things, like I want to do some re-architecting in my product, or I want to develop kind of a major new feature. I will enlist GPT 4.0 to be my consultant, to think through how to approach the development with me, and to generate a series of prompts that I then send over into Cursor, at which point I go back into this feedback loop between me and Cursor. 5% of the time I'm enlisting GPT 4.0 to write a mega prompt that I then send over to O1, your super powerful reasoning model, to then generate a gigantic list of tickets that I can individually send down into Cursor. Usually this is when I'm starting from scratch a new project, or I'm doing some very big update onto a product, and then I send all of those tickets individually into Cursor, and after each one I enter this feedback loop between me and Cursor to refine, and then one at a time I'm sending the tickets from O1 into Cursor. That's generally my workflow. Hope that helps.