I'm arguing that the people doing this THINK it's saving tokens. I don't think it is. I've been seeing faster and more efficient use of the LLM by removing a lot of the documentation.