An agent's goal isn't to save time. It's to gain performance.
"I've never once promoted someone because they saved time." In Adweek, Treasure AI's chief product officer puts agentic AI back in its place: an outcome tool, not a productivity tool. That is exactly the principle I built mediagentic on.
On 2 October, Adweek ran an interview with Rafa Flores, Chief Product and Growth Officer at Treasure AI. Sponsored content, from an agent vendor: I read it knowing who is speaking. There isn't a single statistic in it. But there is a sentence I wish I had written: "Agentic AI isn't a time-saving tool. It's a business-outcome tool."
His argument comes in three parts. First, the wrong question. When you talk to marketing leaders about agents, they ask: is it saving us time, can I cut costs? But marketing's job is to contribute to the business by growing the brand and sales, not to infantilise, sideline or slash teams. "I've never once promoted someone because they saved time."
Second, what a useful agent is. Not a low-cost copywriter, but a system that "understands a signal in context, makes a decision against a clear business objective, takes that action, and learns from it". Give it a customer list with no purchase history, no consent status, no margin target, and it will produce "a beautiful campaign" that is "the wrong message to the wrong customer at exactly the wrong time".
Third, the human in the loop. Not the one who signs off, but the one who judges: what am I approving, what do I want the audience to infer, do I believe it myself? "Real human-in-the-loop is about judgment, not approval."
This article talks about mediagentic without knowing me. On measurement: I don't bill hours, so I have no reason to sell you hours saved. A set-up fee, then a fixed monthly fee, zero commission; what makes me earn more is you staying, which means results being there. On the useful agent: mine never starts from a bare list. It starts from a written plan, objective, KPIs, channel mix, management rules, historical learnings, and operates the platforms inside that plan. On judgment: agents execute every day, and I, or the expert who follows your account, decide what to test and what to stop, understanding the levers at play rather than the ROAS each platform credits itself with.
The one thing I would add is proof. A tech vendor can assert that an agent is an outcome tool. An agency has to measure it, in incrementality, campaign by campaign. That is what I judge my agents on.