Sixty minutes to get someone building for themselves

People ask me fairly often to “come and say something about AI”. It is kindly meant and it does not work. Spend an hour listening to me and you will know what I think. You still will not know what it feels like to describe something in plain language and watch it appear.

So I turned it around. Not a talk with an exercise in it, but an exercise with some talking around the edges. The whole format is below. Take it.

I run sessions like this in various places and in various shapes. For people working on public sector innovation, for founders, for writers, for anyone else who is curious. The audience changes, the format mostly does not. What changes is the example I show of my own work, and where the conversation goes afterwards.

The shape of it

One rule is fixed: at four minutes they are typing. If the opening runs long, I cut the opening, never the building.

Why doing comes before explaining

This is the whole idea, so it is worth slowing down for.

Explain first and people start carefully. They have heard what can go wrong, they know there is an expert in the room, and they write their first instruction as though it will be marked. That carefulness is exactly what you need to remove, because this way of working depends on complaining out loud to the machine when the result is wrong.

Start them cold and they work it out themselves. Within five minutes everyone has noticed that the first answer is never quite right, that you can simply say “no, that is ugly, do it differently”, and that it moves faster than expected. You can also just tell people those three things. They land differently.

The cost is real. The first few minutes are uncomfortable, for the room and for me. No agenda, no introductions, no slide with learning objectives. That discomfort is part of the design, and the urge to talk over it is strong.

Where audiences differ

The exercise is the same everywhere. The conversation after it is not.

In public sector innovation the value is not the one prototype, it is the sequence around it. Make something, put it in front of people, check whether the assumption underneath holds, build the next version, and keep going until you know what actually works. Only at that point does the real question become interesting: do you build this into the systems you already have, or does something new need to exist. And when the answer is “something new”, you design and build that the same way. Speed is not the prize. Being able to doubt cheaply is.

With founders the conversation tends to land on cost. Testing an idea used to mean a quote and a few weeks, which is exactly why most ideas went untested. With writers and makers it goes the other way, towards their own way of working: not building something to sell, but making a tool that only has to work for one person.

What holds across all of them: until you have done it yourself, it stays a story about somebody else’s work.

The assignment

On a card, and this is the full text:

Think of something small you would like to have. Describe it in plain language. Complain when it is wrong. Twenty minutes.

I give no further steering. Choosing your own thing is the point. An assigned task feels like work, and work makes people careful again. I am counting on one person building a mortgage calculator and another a countdown to their retirement, and that is exactly right.

The downside of free choice is that there is no shared result to discuss. The round afterwards covers that, because the patterns are shared even when nobody’s creation is.

What to do when the technology fights back

I assume the worst: managed laptops, no install rights, possibly a network that blocks the whole thing. The format must not collapse under that, so there is a ladder. When level one fails I move to two, and so on.

1. Own laptop, browser, done.
2. Phone on mobile data, if the network blocks the domain.
3. Pairs on one working screen. One types, one directs.
4. The room dictates, I type on the projector.

Level four sounds like failure and is not. Directing without typing is what most people will actually do afterwards. If it comes to that I say so out loud and run the block anyway.

One email a week ahead probably moves the whole group back to level one: bring a laptop, create a free account beforehand, two minutes of work. The worst case is an assumption, not a law of nature.

The caveat, and why I have to raise it myself

Eleven minutes at the end, about the thing they just made. Not in general, because in general everyone agrees and nothing changes.

1. Who maintains this a year from now?
2. Where did your input go?
3. What if everyone here does this without telling anyone?
4. Who is liable when it calculates something wrong?

Question one has a single answer, and that answer is “nobody”. Let the silence sit. Question three matters most: in any organisation of reasonable size this is already happening, and nobody knows how much. Raise it without blame, or the room will start defending itself instead of thinking.

The message of this block is not that AI is dangerous. It is that the prototype was free and the upkeep is not. Anyone who feels that difference asks better questions on the next project.

I put this at the end rather than the start, but I never skip it. Sending people out enthusiastic without those four questions means I sold them half a story. It surfaces eventually, and by then it comes from someone else at a less convenient moment.

What goes wrong

Three things recur, whoever is in the room.

The opening is too quiet. Four minutes reads as short on paper and feels long in a room, and filling it anyway is the mistake I make most often myself.

The round overruns. Everyone wants to show their thing, which is what you want, but it eats into the closing block. I would rather cut my own part than cut those questions.

Someone gets frustrated. That is not a problem, it is the lesson: let them complain at the model instead of at me. The hard part is watching without taking the keyboard, and the urge is strong.

Try it yourself

The whole format fits on half a sheet of paper. What you need:

Card per participant:
  "Think of something small you would like to have.
   Describe it in plain language. Complain when it is wrong.
   Twenty minutes."

Email beforehand:
  bring a laptop, create a free account, nothing else.

For yourself:
  the four questions, the fallback ladder, and the discipline
  to keep quiet until minute four.

Start small. Run it with four colleagues around a table before you use it on a larger group, because keeping quiet is easier to practise with four people than with twenty.