Start with the client’s question
Choose a recurring question that has a useful answer in your teaching. Identify the relevant lesson or framework, the context needed to apply it, and when a human should take over. This gives you a concrete way to evaluate an AI experience before expanding its scope.
Build a review set before you build a library
Write a small set of representative questions and the principles a good answer should include. Add examples where the correct response is to ask for more context or decline to give a definitive answer. Use this set to discuss quality with anyone building your assistant.