Put Each Fact in the Right Kind of Context
Context is temporary attention. Knowledge is maintained evidence. Memory is continuity. Caching is reuse. Mixing them creates confident stale answers. Distinguish chat context, Project instructions, Project knowledge, memory, connectors, retrieval, and API prompt caching. Choose what to persist, retrieve, summarize, refresh, or discard. Build a source registry with authority, ownership, sensitivity, and freshness metadata. Reduce context overload without deleting required evidence. Explain which Claude product behaviors are changeable and must be verified in current documentation. A team creates a Claude Project for quarterly planning. They upload policy files, meeting notes, sales exports, and an old product roadmap. They also add Project instructions that say, "Use the latest approved plan." Three months later, Claude recommends a launch date from the old roadmap. The date is present in Project knowledge, appears in several historical meeting notes, and conflicts with a newer decision stored in a connector. The response sounds certain because the context contains repeated evidence for the wrong answer. The team calls this a hallucination. It is mostly a knowledge-management failure. They treated a Project as a document warehouse, memory as an authority source, and retrieval as a guarantee of truth. More context is not the same as better context. A trustworthy system knows which facts are temporary, which are authoritative, and who is responsible for keeping them current. Claude can…
Put Each Fact in the Right Kind of Context: Context is temporary attention. Knowledge is maintained evidence. Memory is continuity. Caching is reuse. Mixing…
This free lesson is part of the AI Engineering from Scratch curriculum. Read the full explanation, run the lesson code, and verify the result in the interactive reader or from the repository source.
Browse the complete course catalog or open this lesson on GitHub.