As agentic AI moves from pilot projects to core infrastructure, the debate for marketers has shifted from whether to adopt it to how much of the job to hand over. For Rafa Flores, chief product and growth officer at Treasure AI, the answer isn’t about delegating more; it’s about knowing what has to stay human. Judgment, he argues, was never a rubber stamp on a finished campaign; it’s the one thing agentic AI can’t replicate.
Ahead of Agentic World, Treasure AI’s gathering of enterprise marketing and data leaders, we sat down with Flores to discuss what marketers still misunderstand about agentic AI and what real human-in-the-loop oversight looks like.
Agentic AI has moved quickly from concept to something marketing teams are actually using. What do you think marketers still misunderstand about what agentic AI can and should be doing?
Flores: There are a lot of misconceptions around AI for marketers. The biggest one is that people think agentic AI simply means generative AI: you click, you prompt, there you go. That’s too narrow. A useful agent needs to understand a signal in context, make a decision against a clear business objective, take that action, and learn from it.
Picture giving a marketer a customer list with no purchase history, no consent status, no margin targets, no market context. They’ll create a beautiful campaign, because that’s what marketers do. But it could be the wrong message to the wrong customer at exactly the wrong time.
The real question isn’t what AI can do. It’s what we should delegate, under what guardrails, and how we know it actually worked. It’s not about giving machines unlimited authority. It’s giving them a tight responsibility that impacts the business.
Can you paint a picture of what real human-in-the-loop oversight looks like, and how that’s different from what most marketers think it means?
Flores: Human-in-the-loop, by a general psychology definition, means you approve or acknowledge something, something as simple as agreeing that the weather is hot outside. That’s not what we mean by human in the loop when it comes to AI.
Real human-in-the-loop is about judgment, not approval. You have to ask: What am I approving here? What am I trying to get the audience to infer? What do I want them to feel, and do I feel it myself when I look at this? If the answer to any of those is no, you go back and look at the whole path, from ideation through to the final output.
Here’s the quick pitch for the webinar, The Human Half of Agentic AI, that Sam Kwapong, one of our senior product managers, and I hosted recently. We gave tangible examples of human-in-the-loop, and we got a little vulnerable. We wanted the audience to think, “I can relate to this. I’ve made the same mistake.” It doesn’t matter if you’re a marketer or a product person. You use AI in very similar ways, and that is to be creative and to build something.
For marketing leaders who are past the experimentation phase and getting ready to launch agentic AI, what’s one of the first things they should aim to get right?
Flores: Most marketing leaders are making the mistake of being hyper-focused on operational efficiency: Is it saving us time? Can I cut costs? I’ve never once promoted someone because they saved time.
Think about hiring a house cleaner. You pay a flat fee whether they take one hour or four. You’re paying for the job done well, not time saved. That’s the lens marketing leaders need. You’re not going to report to the board on time saved; you’re going to report on business outcomes. Did your brand level up? Did you generate revenue? Is your forecast going up? If you can answer those questions, deploy the AI. If you can’t, you’re still at proof of concept.
As agents take on more of the execution work, how do you see the role of marketers, and specifically CMOs, changing?
Flores: We’ve heard this fearmongering before. Look at the Industrial Revolution: machines came in, some labor was displaced, but new jobs were created. The marketer’s job isn’t disappearing; it’s being recreated. Your job may look different. You have to learn to adapt and get comfortable with that thought, because that’s only going to make you comfortable with AI.
Here’s a hot take: I think the CMO role is going to evolve and potentially streamline into more than just a CMO role. I’m a chief product officer by trade. I like to build products; that’s what I’ve always done. But with AI tools supporting me, plus everything I know on the creative side, I now hold the growth officer title alongside CMO.
I don’t think CMO is the only C-level role that streamlines with AI. CRO, COO, CFO: they’re all going to have to ask themselves tough questions and take on more than they think.
What’s the one thing you hope attendees are thinking differently about once Agentic World is over?
Flores: When you hear “agentic world,” you’re probably thinking, another AI billboard. I want people to walk out thinking about the human element instead: seeing their own face in the mirror.
AI brings speed and scale. People bring empathy, taste, creativity, accountability, and, most importantly, judgment. That’s the new story you want to be part of when you go to work every day.
This interview has been edited and condensed for clarity.