there’s a new kind of coding I call ‘rep coding’, where you fully embrace the inference time, accept that the agent needs 2 minutes, and go do a set of pullups. you don’t watch the terminal. you don’t read the diff until you’ve hit failure. when the agent finishes, you finish your rep first. it’s not that the code writes itself, it’s that by the time it’s written I’m stronger than when I prompted. some PRs take 3 sets. big refactors are leg day
@steeve Ur generalizing too much with the word “tokenmaxxing” but i get your point. As a junior myself i spend tokens for easy tickets and non complex side projects. But i work the math and product for complex problems
there’s a new kind of coding I call ‘rep coding’, where you fully embrace the inference time, accept that the agent needs 2 minutes, and go do a set of pullups. you don’t watch the terminal. you don’t read the diff until you’ve hit failure. when the agent finishes, you finish your rep first. it’s not that the code writes itself, it’s that by the time it’s written I’m stronger than when I prompted. some PRs take 3 sets. big refactors are leg day