You may be right, but I find myself writing English differently depending on the audience: people vs AI.

I haven't done a formal study, so I can't prove it, but it seems like I get better output from agents if I tailor my English more towards the LLM way of "thinking".