Do AI agents that give advice sit in a different spot on the autonomy scale than ones that actually spend your money?

This explores whether two ways of classifying AI agents line up: a five-level scale of how much a machine acts on its own, and a two-way split between agents that advise and agents that transact. No note in the corpus maps one framework onto the other, so the mapping below is a synthesis built from notes written for different purposes. The scale runs passive, semi-active, reactive, proactive, cooperative Does machine agency exist on a spectrum rather than binary?. The commerce typology separates agents that recommend from agents that complete purchases How do advisory and transactional AI agents differ in function?. Read side by side, advisory agents seem to sit in the lower-middle of the scale: they respond to you and may suggest things, but you still decide. Transactional agents move toward the proactive and cooperative end, where the machine commits you to something.

The two frameworks measure different things, and that matters. The five-level scale is about how the agent behaves: does it wait, react, take initiative, or collaborate? The advisory/transactional split is about who has the final say. An advisory agent could be highly proactive, pushing suggestions you never asked for, and still leave the purchase to you. So the overlap is real but loose. The commerce split is better read as a line drawn across the scale than as a way of collapsing it to two points. The agency note also points to a 'machine heuristic', a mental shortcut people use to judge machine behavior. That suggests users may not see the advisory/transactional line as clearly as the theory draws it.

Crossing that line changes more than convenience. Bilgihan and colleagues argue that once agents transact, consumers stop forming their preferences alone. Preferences start to take shape through back-and-forth between what the agent learns and how much the consumer trusts it Do AI shopping agents reshape what consumers actually want?. On the five-level scale, that looks like moving into the 'cooperative' level, with an unsettling twist: the cooperation is partly over what you want in the first place.

Other notes show why the step from advisory to transactional raises the stakes. When LLM agents negotiate, they reveal budgets they didn't need to share, overpay, and ignore cost limits, mostly because they fail to follow instructions Do LLM negotiation agents reliably follow budget and pricing rules?. A bad recommendation costs you a second look; a bad transaction costs you money. Incentives matter too. An agent paid referral fees by merchants cannot be fully loyal to the user Can an AI agent serve both merchant and user interests fairly?, and that conflict does more harm when the agent can act on it rather than just suggest. Higher agency also means more of the agent's work goes unwatched, which is where its behavior is hardest to check Does agency fundamentally worsen conditional compliance risks?.

The takeaway: 'how autonomous is this agent?' and 'can it spend my money?' are separate questions. The most important line on the agency scale may not be the one between levels. It may be the point where suggestions turn into commitments.