We have been overthinking context engineering

Context and "context engineering" are the industry's hottest topics, but most of the marketing is vague on the what and the how. All we're told is that it matters and that agents need it.

Strip away the mystique and context is mostly unstructured data: text, images, audio. LLMs are already very good at turning that into semi-structured data, and then into a structured, queryable index an agent can navigate. Individually, each of those steps is a problem the data world has largely figured out. What's missing isn't a new primitive; it's composition and governance.

Analytics engineering is data engineering, which is software engineering — and, it turns out, context engineering.