JGJos Govaart
Column · 9 November 2026 · 4 min read

Without communication, your AI prototype stays a prototype.

These days you can build a working AI prototype fairly quickly. The real work only starts after the demo.

Drawing: on the left, two nerds in hoodies work behind their laptops, with a small robot on the desk. In the middle, a little figure carries a prototype labelled v0.1, still attached by its cable, towards three people on the right. One waves, one holds a coffee, and above them float a speech bubble and a bubble with a red heart.

Last week, the Dutch podcast POM talked about AI that can talk directly to all kinds of systems. In a side remark, Dutch tech journalist Alexander Klöpping said something that stuck with me. Anyone with an idea can now build something themselves. If you can see how things could work better at your own company, you can get it working fairly quickly. But only on a small scale. Once it needs to grow, you need the old system again. And the old system is slow.

I've been through that myself.

Vibe coding is easy

When I was at Brand Potential, we built a system for a large client to make their way of working more efficient: briefing, debriefing and all the project management around it. The goal was simple: work 30 per cent more efficiently.

We knew the process inside out, so mapping it was easy. Based on that, we vibe-coded a prototype, and fairly soon there was a working live demo.

But that doesn't give you a product that meets every security requirement. That only starts afterwards. You need sign-off on resources, on scope, on cooperation from IT and goodness knows what else. And suddenly it's back to good old-fashioned building support.

We got as far as a pilot. Whether it was ever properly implemented, I don't know. I had already left.

The enthusiasm has to survive the handover

It's brilliant when people start experimenting, developing and building things themselves. The last thing you want is for their idea to end up in that painfully slow, cumbersome system and quietly die there.

So the slow system will have to adapt too. That doesn't need a new committee. It needs a process: a fixed route along which one employee's idea is weighed quickly, gets a budget and gets an answer.

Someone had already thought of it

That was as far as I had got on my own. Then I did some more reading, and ended up with Ethan Mollick. Yes, him. The American professor quoted in every other piece about AI, whose new book has just come out in Dutch with the publisher behind that very same podcast. He talks about exactly the same thing, just in different words.

In many organisations, Mollick sees what he calls secret cyborgs: employees who quietly use AI and work a lot better for it, but don't tell anyone. They're afraid of the hassle, or of what their colleagues will think.

According to him, you need three things. Leadership that sets the direction. The employees themselves, because that's where the best ideas come from. And a place that turns the experiments of a few people into something for the whole organisation. He calls that the lab. I called it a process.

Who's at the table?

This is where it gets interesting for me. In my previous column, I wrote that the communications industry is too pleased with itself. You can see that with AI: communication isn't at the table.

And yet Mollick's problem isn't really a technical one. People keep their experiments to themselves because showing them doesn't feel safe, or doesn't get them anything. Changing that isn't a question for IT. It's a question for communication.

The process takes care of the decisions. Communication makes sure people dare to show what they've built, that the makers get noticed, and that a "no" gets explained, so that next time they simply start building again.

And yes, in my first column, communication was the slow system itself: review rounds where everything gets added and hardly anything gets taken out. But that's exactly why the profession knows that system from the inside. And knows how to get through it.

Don't wait for an invitation

Find the people who are already building. Ask where their idea gets stuck. And offer to organise that part.

You can build a demo fairly quickly. Making sure it becomes more than a prototype: that's our job.

Sources:

Thinking: Jos Govaart · Wording: developed with AI (T1W2) · How I use AI

Jos Govaart works where brand and communication meet. More about Jos

Get in touch

All columns