AI agent vs chatbot — what's actually different

A chatbot is stateless question-answering

Paste context in, get an answer out. Each message is mostly independent; the model doesn't maintain beliefs about your situation, doesn't track what it's confident about, and doesn't notice when the situation on your screen changed.

That's fine for drafting and one-off questions. It's useless mid-meeting, when the context IS the screen and there's no time to paste anything.

An agent keeps a model of the session

LivePair's agent maintains persistent state across the session: what you're trying to accomplish (goals), what it has concluded so far (beliefs), how sure it is (uncertainty), and how that changed with each turn. A private deliberation pass runs between messages, updating that state before it answers you.

The result: it doesn't start cold every message. It answers inside the context of everything that already happened — which is what a human pair actually does.

Where this matters in practice

Meetings, interviews, debugging sessions — anywhere the question depends on the last twenty minutes. A chatbot can answer 'what does this error mean'. An agent can answer 'is this the same problem we ruled out earlier' — because it remembers ruling it out.

The honest limits

More state means more ways to be subtly wrong — a stale belief can bias a new answer. That's why the uncertainty HUD exists: the agent's internal state is visible, not hidden. If its belief looks wrong, you can see it and correct it instead of trusting a black box.