Rethinking RAG from first principles for agents with @beyang and @Sourcegraph Rather than requiring a general retrieval engine that infers an optimal context ranking for a broad set of user tasks, agents can effectively take advantage of a much simpler but broader suite of retrieval tools.