I find it kinda a lock-in strategy by OpenAI. You do RAG and function calling, etc. all using a convenient Assistant API (https://platform.openai.com/docs/assistants/overview).
But I'm wondering if anyone has found it better than creating RAG pipelines manually by yourself. I like the idea of being able to switch from one API backend to another one without losing much functionality. But if you use OpenAI's assistants, it seems like there's no easy way to just "switch" to Mistral API for instance.