Over the last two years I have taught about 500 people to build software by talking to it. Most of them had never written a line of code. Some had never opened ChatGPT. They were city workers in Amsterdam, agency directors in Campinas, business school alumni, and a full room of people with laptops on a Sunday afternoon at a festival in Rio. At the Embassy of the Free Mind in Amsterdam I ran half a dozen sessions on AI and the humanities, and more sessions ran online.
I am not a software engineer. I taught myself HTML as a teenager and then spent twenty years imagining technology instead of writing it. But when people ask me where language models pay off most, my answer is code. English is now a programming language, and almost nobody has been told.
The skill takes a minute
The technical part is trivial. You describe what you want, the model writes it, you look at it and say what is wrong. Anyone who can write an email can do it.
What I am really teaching is the confidence to ask. Most people do not know they can build software by asking their computer for it. When they find out, you see it in their eyes.
Here is part of what one round of sessions at the City of Amsterdam produced:
- A cycling route planner that follows public street lighting, so you avoid riding in the dark.
- A minutes tracker for a youth football coach, because parents complained their kids did not get enough playing time.
- A management app for someone who had just bought a horse: feeding schedule, vet notes, genetics, exercise.
- A tool for a new homeowner to track building materials and renovation phases.
- A house hunting check that uses public data to find noise pollution and other nuisances before you buy.
- A generator that turns online hate speech into abstract artwork.
- A chatbot that walks colleagues through procurement requirements.
- A bedtime story app that introduces new words to a five-year-old.
- An app to teach Javanese to grandchildren.
- An app that turns photos of clothes into listings for Vinted.
- A table football tournament manager.
- A garden planner for what to plant where, and when.
As far as I know, everyone left with a working app, and some published theirs to the web the same day. These are toy apps and early prototypes. But they show how fast, and how widely, people can move once they know they can.
I call it the click moment: I can just build things. I can talk to it, and it makes software for me. For some people it is close to a psychedelic experience, because it changes what they expect a computer to do.
The real bottleneck is imagination
After about 500 participants, I think the limit is not technical ability. It is a poverty of imagination. We have spent decades learning to ask which product to buy. Very few of us have practiced asking what tool we would make if making it were free.
That is why I start with personal projects, not corporate ones. When people build for work, they get stuck on security rules and data access, and the click moment never comes. When they build for themselves, they have the context, the motivation and the data already.
The method
Everywhere I teach now, I use two steps. First, people answer a set of guided questions with ChatGPT or Claude. That gives them a short brief of a few hundred words. Then they paste the brief into a builder like v0, Lovable or Replit and get a working web app in one go. Dropping beginners straight into a builder works much worse. The brief is where they do the thinking.
One afternoon is not enough
The first session is shocking in a good way. But for a lot of people it stays a nice memory unless they come back. The value compounds. By your fiftieth build you have prompts, components and decisions you can reuse, and the fifty-first is much easier. So I am moving to recurring sessions and a more advanced group, so people build a habit and not only a story about the day they made an app.
The model does the typing. Deciding what is worth building is still up to us.
