$ man rag-retrieval-augmented-generation

RAG (Retrieval-Augmented Generation)

A pattern where AI retrieves real documents before generating a response, grounding output in actual data instead of training data alone.

by Shawn Tenam

This is the "RAG bottle" concept. Without RAG, an AI model answers from its training data - frozen in time, generic, and sometimes wrong. With RAG, you feed it your actual documents, CRM records, or research before it responds. The output is grounded in reality. For GTM engineers, RAG is how you build due diligence research agents that pull real company data instead of hallucinating revenue numbers. It connects directly to context engineering - the better your retrieval pipeline, the better the AI output. Every enrichment cron that feeds data to Claude before generating outreach is doing RAG whether you call it that or not.

The ABM pipeline does RAG implicitly. Exa searches pull real-time company data. Apollo enrichment feeds verified contact info. That data goes into Claude context before generating personalized landing pages or outreach. The context-wiki entries you load in CLAUDE.md are a form of RAG - retrieved knowledge that grounds every session. Building explicit RAG pipelines means setting up vector databases or search indexes that agents query automatically before responding.