Definition
Plain language
A computer trick of keeping only what you're actively using in fast memory and fetching the rest from storage only when needed.
As stated in the literature
An operating-systems memory strategy that keeps the active working set resident and faults in other data on access; reused as the design metaphor for hierarchical agent context, where compact gists unfold to summaries or raw text on demand.
Why it matters: It lets a system handle far more data than fits in fast memory by fetching pieces only when needed, which is what makes large context manageable for agents.
For example, a phone keeps the app you're actively using ready in fast memory and only pulls a background app's data back in the moment you tap it.
Heard on the show
“That's demand paging, and it's why memory cost scales with what you use, not with the total size of everything you might use.”Episode 130 — Why AI Agents Coordinate Better Through a Shared Board Than a Boss