AI is creating answers in *.md files, but I need to regenerate couple times before I actually get the file - which burns tokens. Hurts especially in high-cost models :)