RAG Isn't an Agent — I Built the Layer Between Retrieval and Action

A distinction is made between Retrieval-Augmented Generation (RAG) systems that find information and agents that take action. To test this separation, three systems were built in Python: a retrieval-only system, a deterministic action-only planner, and a hybrid system connecting both. The architecture uses a TF-IDF retriever on a custom corpus and a simple agent that acts on a ticket store environment. The same nine tasks were run through all three systems to empirically measure how connecting retrieval with action-taking capabilities impacts performance.