The default assumption in most LLM developer communities is that you start with raw API calls and graduate to a framework as your project grows.