Retrieval: RAG vs Tools vs Long Context

An agent is only as good as the information it can bring to bear. Getting the right facts in front of the model is the retrieval problem, and there are three broad strategies: stuff everything into a long context, retrieve relevant chunks (RAG), or let the model fetch what it needs through tools. This chapter compares them and explains when each wins, and why a tool-based approach is often the right default for document agents.