There are a lot of personal AI assistants now, and most comparisons list features. Features are not what makes one stick. Four properties are.
1. Where it lives
An assistant in its own app competes with every other app for your attention, and loses. An assistant in the messaging app you already have open all day does not have to win that fight.
2. How much context it starts with
Zero-context assistants make you the integration layer: you paste, they answer. The ones that keep earning their place start each request already knowing the six months of conversation it refers to.
3. Whether it can act unprompted
A tool that only responds is a better search box. A tool that can run at 7am on a Friday, notice nobody replied, and tell you — that is an assistant. This requires scheduling, durable state, and bounded autonomy.
4. Whether you can read its mind
Every assistant builds a model of you. The good ones write it down in files you can open, edit and version, so a persistent misunderstanding is a one-line fix rather than five more corrections in five more chats.
The checklist to score any of them
- Does it see the context I already have, or do I have to supply it?
- Can it act on a schedule, and can I see every scheduled thing in one place?
- Can I read and edit what it believes about me?
- Can I state its limits from memory — what it will never do, in any mode?
- When it fails, does it tell me where it got to, or fail silently?
Score five out of five and the category label stops mattering.