I like the idea of saving the work done into files - helps to prevent the llm from redoing the same work. Maybe one day instead of code in a repo it will just be a list of prompts.