AI agents in marketing: the challenge isn't the technology, it's deciding what to delegate and how far

By Marketing4eCommerce México · September 24, 2026.

María Laura Barreto, a member of the Artificial Intelligence Committee at Argentina's Data & Marketing Association (DMA) and part of the Doppler team, argues in that committee's second quarterly report that "the challenge is not technological, it's organizational: it's about learning to work with agents, defining what can be safely delegated and building the controls to make that delegation predictable." The interview, published by Marketing4eCommerce México ahead of EMMS 2026—organized by Doppler and Hostinger—uses an everyday example to distinguish the assistant from the agent: asking someone to "write me an email for the people who didn't open the last campaign" produces a task the human reviews and approves; asking them to get "the people who signed up this week to actually start using the platform" sets off a process, with chained decisions about who to contact first, what to say and when to follow up, that no one dictates step by step.

To decide what to delegate to an agent, Barreto proposes a five-question filter: whether the task is repetitive, whether it has volume, whether the rules can be written down, what happens if it goes wrong and whether human judgment is needed to resolve it. When the first three are a yes and the risk in the last two is low, there is a clear candidate: qualifying leads, making first contact with someone who has left their details, auditing sales calls. On the other side are negotiating with a large account, calming down a furious customer or deciding on an exception; there the agent can gather information and propose a response, but, in Barreto's words, "the final word belongs to someone with a first and last name."

Autonomy, she insists, is not switched on all at once: she compares it to a dial rather than a switch, running from suggesting, to executing with prior approval, and finally to executing on its own within certain limits. The question that matters, she says, is not what the technology can do—"it can do a lot"—but how far each company wants to turn that dial in each specific process. As industry context, the outlet itself places this conversation within an accelerating adoption of agents in companies, with retail among the most active sectors, and links it to two fronts also gaining ground in digital marketing: agentic commerce (purchases an agent executes on the user's behalf) and the fight to be cited in the answers given by AI assistants themselves.

What makes this interview valuable is not marketing itself so much as the pattern it describes, already visible in other trades we have analyzed: automation advances first over what is repetitive and measurable—qualifying, contacting, auditing—and stops, rightly, where judgment comes in: negotiating, defusing, making an exception. It is the same map that separated the administrative from the relational in banking, law or education, now applied to a marketing department. And Barreto's five-question filter is, at bottom, a small governance manual: the same "dial, not switch" logic that AI labs should apply to their own systems shows up here at the scale of a campaign team.

The short-term honesty is clear: teams that today spend hours segmenting, contacting and manually following up on leads will see those tasks absorbed by agents, with less room for junior profiles devoted solely to execution. Barreto's own framework acknowledges that buying the tool is not enough; processes have to be redesigned and a decision made, process by process, about how much autonomy to grant, and companies that fail to do that internal governance work will fail for lack of judgment, not for lack of technology.

But the underlying horizon remains the one we defend: if an agent can qualify leads and audit calls without constant supervision, it frees up the people on the team for what really sets a brand apart—strategy, creativity, the conversation with the furious customer who really does need someone on the other end. The abundance AI promises in marketing is not the disappearance of the profession, but the mechanical part no longer taking up the agenda, leaving only the part that is, on top of everything, the part professionals in the sector enjoy doing.