Model weights are a lossy, frozen compression of the world, so every serious AI system needs a way to reach the rest of it. This course teaches that path, from the inverted index of the 1970s to the research agents of 2026. Half the assigned papers were published in the last eighteen months, and one of them has no venue yet. The framing is information retrieval for AI, not AI for information retrieval. Classical retrieval is the toolkit rather than the subject, and BM25 earns its place not as history but because dense retrievers still lose to it out of domain. Monday is a lecture on the method. Wednesday is a structured discussion of a research paper that uses it. Alongside both, the class builds its own document collection, writes its own queries, judges relevance by hand, and measures a retrieval system against the judgements it produced itself.