I used Vercel's ai sdk like everyone else, to try to stay agnostic. However, when I am actually working on a feature that uses an LLM API call, I find that each

I used Vercel's ai sdk like everyone else, to try to stay agnostic. However, when I am actually working on a feature that uses an LLM API call, I find that each prompt might need tweaking specific to that model to get great results. That means if I switched models for that feature, I would likely have to that redo the prompting. Or, am I doing this all wrong?