Raw LLM Responses

I mainly use LLMs to write isolated functions or scripts. I've needed to give pretty detailed, lengthy prompts to get the code to be organized how I want (especially for more involved scripts), but this is still way faster than actually writing the code myself. Common, tedious access patterns (like s3 access, local path manipulation, spawning async web requests, etc) are especially where AI shines, these things are so common online that you don't need to be very explicit in the prompt about them. I haven't had much luck with modifying existing code, or with anything that requires design or higher-level thinking. I think it would require multiple step-level improvements to "replace" a competent junior dev, with the current tech at least that's just nonsense.