Ask HN: Preferred Dev Workflow with LLMs?

Hi HN - so my current dev workflow using AI assistance is basically: Give a spec to chatgpt, copy generated code into my IDE, run code, inspect output, then iterate by asking chatgpt for additions, changes etc, re-pasting into my IDE. The problem with this is that I often want to make my own changes to code directly in the IDE, and as such what's in my IDE gets out of sync with what's in chatgpt. Am sure this is a solved problem, so I was wondering what people's preferred approaches were for integrating LLMs into their dev workflow?