
RAG and Agentic AI Techniques

AI is Smart Until It's Dumb: Why LLM Will Fail When You Least Expect It - Emmanuel Maggiori
Topic 10 of 19
Published Nov 10, 2025 · Clip 30:01–34:12 (4:11)
Emmanuel explains RAG (retrieval augmented generation): a separate algorithm finds relevant documents from private data, inserts their text into a long prompt, and instructs the model to answer using that data—specializing AI without changing the model, as likely done in Google AI results. He then explains agentic AI: using a model autonomously where its outputs trigger automated actions via APIs (e.g., scanning emails, searching flights), chaining agents together—but fundamentally every agent is just an LLM receiving a fancy prompt.