RAG is dead… except it’s not. In this video I break down why that “RAG is dead” narrative is everywhere (especially on LinkedIn/X) and why it’s only true for a very specific slice of problems: AI coding agents working over structured codebases. When people say RAG is dead, they usually mean traditional semantic search: chunk documents, embed them, store in a vector DB, retrieve by similarity. That pipeline still matters a ton for unstructured business knowledge—Google Drive, SharePoint, internal docs—where language is ambiguous and you need synonyms and conceptual matching (my favorite silly example: “Star Wars spaceships” should find X-wings and TIE fighters even if those words never appear in your query). For code, the game is different. Code is structured, identifiers are exact, file paths give you built-in organization, and agents live in the terminal with tools like ripgrep, glob, cat, and friends. That’s why top tools like Claude Code moved away from vector DB indexing in favor of agentic search, and why others (like Aider) use a repo map/tree-sitter-style overview instead of embeddings. My main takeaway: traditional RAG is basically dead for code, but it’s still critical for most other agentic apps—and the real future is bridging both retrieval strategies so the agent can decide how to search based on the query.