> When it was time to start coding, I asked Claude to give me a big and complicated markdown file outlining my existing infrastructure, my desired new infrastructure, what I’m trying to achieve, why I'm doing it, etc.
Yes, imo, too many people believe current LLMs are capable of doing this well, They aren't. Perhaps soon! But not today, so you shouldn't try to use LLMs to do this for serious projects. Writing that MDN file sounds like a wonderful way to get your own head around the infrastructure and chat with other devs / mentors about it though, that's a great exercise, I'm going to steal it.
Anyway, LLMs are, as we all know, really good text predictors. Asking a text predictor to predict too much will be stretching the predictions too thin: give it one file, one function, and very specific instructions, and an LLM is a great way to increase your productivity and reducing mental overload on menial tasks (e.g., swap out all uses of "Button" in @ExperimentModal.tsx with @components/CustomButton.tsx). I think LLM usage in this sense is basically necessary to stay competitively productive unless you're already a super-genius in which case you probably don't read HN comments anyway so who cares. For the rest of us mortals, I argue that getting good at using LLM co-pilots is as important as learning the key bindings in your IDE and OS of choice. Peel away another layer of mental friction between you and accomplishing your tasks!
Yes, imo, too many people believe current LLMs are capable of doing this well, They aren't. Perhaps soon! But not today, so you shouldn't try to use LLMs to do this for serious projects. Writing that MDN file sounds like a wonderful way to get your own head around the infrastructure and chat with other devs / mentors about it though, that's a great exercise, I'm going to steal it.
Anyway, LLMs are, as we all know, really good text predictors. Asking a text predictor to predict too much will be stretching the predictions too thin: give it one file, one function, and very specific instructions, and an LLM is a great way to increase your productivity and reducing mental overload on menial tasks (e.g., swap out all uses of "Button" in @ExperimentModal.tsx with @components/CustomButton.tsx). I think LLM usage in this sense is basically necessary to stay competitively productive unless you're already a super-genius in which case you probably don't read HN comments anyway so who cares. For the rest of us mortals, I argue that getting good at using LLM co-pilots is as important as learning the key bindings in your IDE and OS of choice. Peel away another layer of mental friction between you and accomplishing your tasks!