What is an AI marketing agent?

The distinction worth holding onto is between a tool and an agent. A tool waits to be asked — you supply a prompt, it returns copy. An agent is given a goal, such as launch a campaign for this product to this audience, and works out the intermediate steps: research the market, position the offer, choose channels, write for each one, then measure. In practice most useful systems are several narrow agents with an orchestrator, rather than one model asked to do everything, because the skills involved genuinely differ.

Two things break first, and neither is writing quality. The first is voice: a system generating at volume drifts away from how the brand actually sounds, and the drift is gradual enough that nobody notices until the output reads generically. The second is truth. Marketing copy makes claims, and a model will produce a confident statistic or a customer benefit it has no basis for — which is a legal problem as much as an editorial one.

So the questions to ask are about supervision rather than capability. Does the system grade its own output, and is the grader independent of the writer? Can you see what a claim was based on? Is there an approval step before anything is published, and is that step meaningful or a formality? And can it tell you whether the work made any money — an agent that produces plausible assets and cannot connect them to revenue has moved the labour, not the outcome.