Agent memory
The persistence layer through which an autonomous agent carries state across invocations — working context, retrieved documents, prior decisions, and learned preferences — distinct from the model weights themselves.
The position
Memory is the part of an agent that governance frameworks have not noticed. Weights are versioned and reviewed; memory accumulates silently and changes behaviour without any model change at all. Two identical deployments with different memory states are different systems, and no current model-risk process would catch that. If you cannot snapshot and diff an agent's memory, you cannot reconstruct its decisions, whatever your audit log says.
This section is a claim, not a definition. It is argued rather than asserted, and it is the part of this page you are invited to disagree with. Corrections and counter-arguments to [email protected] are published.