AI in Marketing

Stan Berteloot on AI, Expertise, and the Risk of Letting Your Brain Go Quiet
Stan Berteloot is the Chief Innovation Officer at Nytro Marketing and the host of two podcasts: Back in America, a long-form interview show, and AI in Marketing, a daily news briefing produced almost entirely with AI tools he built himself. He studied journalism at the University of Maryland, starting in photojournalism before moving to print, and began his career at Reuters in London. The journalism never fully left. It just found different outlets.
What makes Stan worth listening to right now isn't his enthusiasm for AI. It's that he uses it harder than almost anyone and is increasingly preoccupied with a question most people in his position don't say out loud: what happens to your own thinking when you hand more and more of it to a machine?

You've been in marketing for a long time, but you started as a journalist. Does that still matter?
It matters more than people might think. When I was a student, marketing was not something I was interested in. But you need to make a living, and you make more money as a marketer. That's the sad truth.
What I kept from journalism is the ability to find an angle. Sometimes you've got clients with very complex technology, and they don't even know how to talk about their product. Being trained as a journalist, I could find the human interest story inside the product and make it marketable.
I also never stopped doing journalism entirely. I had a blog in France covering local issues, and then after moving to the US, I started podcasting. I didn't realize at the time how much that would matter. Opportunities came out of it that I never planned for.
Most people in your position talk about AI as a productivity tool. You seem to think about it differently.
AI should be used as a collaborator. Not a vending machine where you put in a question and get a finished answer.
One thing I do constantly is ask AI to criticize itself. I'll work back and forth with it until it produces something that looks like a final document. And that's exactly when I'll say, "Now step into the shoes of my reader. How would you criticize this? What would you do differently?" Most people skip that step because they assume AI is all-powerful and all-knowledgeable. It's not. It's very good. But it's significantly better if you treat it like a smart colleague who still needs direction.
And that's the part people underestimate. AI is only as good as you are at directing it. If you don't understand value propositions, or segmentation, or what signals in the market actually mean, you can't give AI useful instructions. You'll get something obvious and flat. The human expertise is what makes the AI output worth anything. That's true whether it's me or anyone else with deep knowledge of their field.
You built VoiceStream, an AI-powered podcasting tool, without being a developer. How did that happen?
I had a very concrete problem. I wanted to produce a daily AI news briefing but didn't have the time to do it manually. I started by stitching together about nine different tools on Make.com. It was incredibly complex, very buggy, and broke constantly. But it worked for about a year.
Then what people are now calling vibe coding came along, building software by describing what you want to an AI rather than writing code yourself. I picked up Cursor, and what changed is that you can go from idea to feature in a single session. I wanted search functionality inside the app, so I connected specialized APIs. I wanted newsletter management built in, so people could subscribe and receive content I'd already produced for the podcast. These are things that would have taken months of negotiation with a dev team.
I'm not a coder. That someone like me can build and ship a working product says more about where the tools are now than it does about me.
You've been unusually open about the psychological cost of all this. What does that actually feel like?
Every day I use Claude and see the results, part of me goes, "Oh my god." The other day I needed to scrape a list of contacts from a website. In the old days, that's half a day of copy-pasting. I used Claude Code, described what I needed, and had everything in fifteen minutes.
That's exhilarating. But there's a cost. The more you use AI, the lazier your brain becomes. You stop thinking. You stop writing. You just prompt. "Correct that. Make it better." And at some point you have to ask yourself: how do I keep functioning as a thinking person when I outsource this much of my cognitive work?
I don't have a clean answer. But I've been thinking about it a lot, and I recently published a list of 20 books on my newsletter that I think matter in this moment. My belief is that reading, literature, art, cinema, theater: these force you to think in ways your daily work doesn't. When you read a serious book, you're confronted with ideas and dilemmas you'd never encounter otherwise. Your brain makes connections it wouldn't make on its own. And when you come back to your day job, those new neural pathways are there, quietly working.
It's not a solution. But it's the best counterweight I've found.
Has AI made you afraid of losing your job?
Not yet. But I know people who are. Developers, especially. They see what tools like Cursor and Claude Code can do, and they're rattled.
What's happened is that the bar has gone up for everybody. You need to be excellent at your actual skill, and you need to be excellent at using AI. That combination is now the baseline. It's harder than before, not easier, even though the tools are more powerful.
What are you still figuring out?
Agents. Everyone's talking about them, and I've been experimenting. I tried OpenClaw on a virtual machine, burned through API costs too fast, switched to DeepSeek, which is cheaper but has security issues. I've pulled back for now, especially since Claude has added scheduling features that cover some of what I was trying to do.
But the real issue isn't the tools. It's knowing when to use them. You can play with agents all day, but if you want a real business case, you have to stop and map out your processes first. Where does automation actually make sense? Not everything needs AI. Zapier and Make.com have been around for years, they're much cheaper, and for certain workflows, they're perfectly fine.
The expertise is in knowing which tool fits which problem. That hasn't changed.