Is RAG Becoming Outdated Now That Long-Running Agents Like Claude Cowork Are Emerging?

What this post added

This post analyzes the emergence of long-running agents like Claude Cowork and contrasts their task-driven memory mechanisms with knowledge-driven RAG systems. It details how Claude Cowork likely uses a hybrid approach of persistent long-term memory and on-demand retrieval tools, including semantic and keyword search, to manage state across extended sessions. It also contrasts this with ChatGPT's more structured, layered memory approach (User Memory, Session Metadata, Current Session Messages, and a cross-session summary). The post then outlines the engineering challenges of building writable memory systems for agents, focusing on determining what to remember, how to store and tier memory, and how to implement fast writes without degrading retrieval performance.