2 篇博文 含有标签「continuity」

Shutdowns make the argument vividly, but they are the rare case. The common case is quieter and happens constantly: a model you depended on is deprecated, a price tier changes, a product is folded into another product, a rate limit arrives, terms are revised, a feature you built a workflow around is retired.

The interesting question is not whether that happens. It is what it costs you when it does — and that is decided long before, by where your agent's durable parts live.

Memory is the part of an AI agent that is genuinely yours, in the sense that it could not have come from anywhere else. The model is a commodity — impressive, interchangeable, and someone else's. What your agent knows about your projects, your preferences, and the decisions you already made is not.

Which makes it worth asking where that lives, and on whose terms.