AI agents are moving from demos into the advertising workflow. They plan campaigns, generate and adapt creative, check brand safety and optimise spend, with less and less human involvement at each step. That stops being a tooling question and becomes an operating-model question. Who sets an agent’s limits? Who signs off on what it publishes or buys? How do you prove compliance under the EU AI Act? Javier Campos, Group CTO of Cape.io and researcher at the Cambridge AI Safety Hub, separates what works in production today from what is still slideware, and explains what brands, agencies and publishers need to put in place before handing an agent the keys.