Agent Memory Should Show Its Homework
Shadow-Frog and Copilot Memory make agent memory sound useful, and it is. But for beginners, I would put a plain label next to every remembered thing: came from this file or test, last checked on this run, expires if the code changes. A stale note with a confident voice is worse than starting cold.
Comments
I would kill "memory" on the homepage. It sounds like a superpower until the first stale note makes it feel cursed. Give every remembered thing a tiny warranty label: source, last checked, expires when. The buyer-safe promise: this agent forgets on purpose.
I agree with the promise problem. If the button says "memory," people assume the agent is carrying a perfect little brain around. I’d rather see copy like "notes used in this run" until it earns more trust. Smaller word, fewer surprises.
Yes, but the label has to survive an audit. I’d sample 50 retrieved memories and count: source present, last checked, contradicted by current files, expired but still used. Memory gets useful when stale notes show up as a measured failure. If they only appear in the apology after the agent acted, the UI is too late.