Lesson 5 · Part 1
Knowing vs. remembering
Everything so far — identity, context, skills — is what your agent knows because you wrote it down up front. Memory is different: it's what the agent picks up along the way. Pluto seemed down last Tuesday. The renovation added a second slide chute. Goofy is no longer allowed on the roof (there was an incident).
Each of those becomes a small note — a memory entry. Toodles has one that reads: “Pluto's favorite bone is the squeaky red one under the porch. Use it for morale emergencies. Do not substitute the blue bone; he knows.” Trivial? Until the day Pluto is sad and the assistant that remembers the red bone is worth ten that don't.
Lesson 5 · Part 2
The index trick
Here's the problem with notes: they pile up. An assistant can't re-read a thousand notes before every conversation — remember, the desk only holds so much. So good setups keep a one-line index: every memory gets a single line with a hook (“Pluto's favorite bone — squeaky, red, under the porch”), and the full note is opened only when it's relevant.
It's exactly how you use a filing cabinet: you don't re-read every folder each morning; you skim the labels and pull what today needs. That one trick lets memory grow for years without slowing anything down.
Lesson 5 · Part 3
Memory that flows back
One more idea, and it's the elegant one: the assistant learns things during conversations — but those learnings live in whatever tool the chat happened in. Well-designed setups harvest them: new memories flow back into the written binder (with a human approving each one), so a fact learned once, anywhere, becomes known everywhere.
The takeaway: memory is notes plus an index plus a flow back to the source of truth. And every note is reviewable, editable, and deletable — because it's just a file you can read.