RAG vs Agent Memory: Key Differences Every AI Developer Should Know

RAG (Retrieval-Augmented Generation), formalized by Lewis et al. in 2020, retrieves relevant passages from a static document corpus at query time and feeds them into a model's prompt without retaining anything learned. Agent memory, by contrast, is a read-write system that stores corrections, preferences, and decisions gathered through interactions, accumulating and refining knowledge across sessions. While RAG is well-suited for question answering over fixed document sets like manuals or wikis, it cannot update based on user feedback, meaning the same errors can repeat across sessions. Agent memory addresses this by supporting mechanisms such as reinforcement, decay, and forgetting, drawing on cognitive science concepts like episodic and procedural memory. The distinction is critical for developers building AI agents that need to learn and adapt over time rather than simply retrieve static information.