It’s really not difficult to work with the LLM to tweak behavior in eg speckit for higher or lower levels of this behavior. It is setting a baseline of how the LLM should engage with a project which is pretty critical to getting the highest quality output with lower token overhead and API costs. It mirrors the way LLMs are used productively in countless tasks in other industries to structure their output according to needs, be it something like content analysis and summary or tasks that require more structure inherently like data analysis.