Why measuring agentic AI in marketing by hours saved is the mistake, according to Treasure AI

By Zendoric · October 3, 2026.

Treasure AI, a marketing platform built on agentic AI (systems that not only generate content, but interpret a signal, decide against a business objective and execute autonomously), is the subject of an interview with its head of product and growth, Rafa Flores, ahead of Agentic World, the event the company itself organizes for marketing and data executives. His central thesis: the underlying mistake is not adopting agentic AI, but measuring its success with the wrong metric. According to Flores, time saved is not what gets reported to a board of directors; what matters is whether revenue, the sales forecast or the brand goes up.

The distinction he draws is important: agentic AI is not synonymous with generative AI. An assistant that drafts a text from a prompt is not an agent; an agent, by the definition Treasure AI uses, needs business context in order to decide well. Flores's example is clear: handing a marketing professional a list of customers with no purchase history, no consent status and no margin target produces a visually perfect campaign that can still target the wrong customer at the wrong moment. The relevant question, he insists, is not what AI can do, but what is delegated to it, under what limits, and how you verify that it worked.

From that comes his second idea, the most interesting in the interview: he redefines what 'human in the loop' means, the principle that supposedly guarantees human oversight of what an AI does. Flores argues that in practice it has been watered down to a simple 'sign-off' on an already finished draft. Real oversight, he says, means exercising judgment before and during the process: what is being approved, what you want the audience to infer, whether the person in charge actually feels what the message is trying to convey. If the answer is no, the entire chain has to be reviewed, not just the final output.

Flores extends the argument to the org chart: he believes the chief marketing officer (CMO) role will tend to merge with other executive functions as AI absorbs pure execution. He himself already holds the title of chief product and growth officer. And he anticipates that the same process will reach other executive positions —sales, operations, finance— which will have to redefine what the person uniquely contributes. He invokes the Industrial Revolution as a precedent: work does not disappear, it is reshaped.

All of this should be read with a necessary caveat: this is sponsored content, a promotional interview published ahead of Treasure AI's own event and carrying its product's sales pitch. Nothing said here is verified against external data on adoption, performance or customer results; it is the story a vendor wants the market to buy. That said, the underlying thesis —that what survives automation is judgment, not execution— is not original to Treasure AI: it matches the pattern we have been documenting sector by sector. In law, the value of expert judgment rises against routine drafting; in banking and insurance, administrative headcount shrinks while risk and compliance profiles grow; in education, the teacher who orchestrates AI wins out over the one who merely delivers content. Marketing fits the same template: agentic AI absorbs segmentation, scheduling and the drafting of variants first; what it does not absorb is deciding which story deserves to be told, and to whom.

In the short term this means, as in the other sectors analyzed, less demand for pure execution in the middle ranks of marketing teams, and real pressure on the function's more administrative profiles. That should not be glossed over. But if the promise holds —agents that execute measurably against a business objective, freeing people for what Flores sums up as empathy, taste, creativity and judgment—, it is a small but concrete example of the thesis we defend: the abundance AI can bring is not about working less, it is about the work that remains being the work that genuinely requires a person. The question no vendor-sponsored interview is going to answer is how many organizations today have the discipline to measure business outcomes instead of settling for the more comfortable figure: the hours they no longer have to pay for.