You're assuming that different models need the same stuff in AGENTS.md
In my experience, to get the best performance out of different models, they need slightly different prompting.