Tipping into AI

A training session with me runs for two days, four three-and-a-half-hour blocks. The first holds no practice. The next three are entirely practice, with rising difficulty. Here's how it works.

A hundred training days last year, across sixty sessions, with around three hundred and fifty people. The repetition has let me sharpen this format, block by block.

The mechanism it triggers, I've laid out in When you can, you want to. Perceived ability comes before desire. As long as the person doesn't know they can, they don't want to. And the move from "I don't know how to do that" to "I can do that" isn't decided. It's witnessed. The trainer doesn't teach the shift. They engineer the conditions for it, both material and emotional.

The wall falls before the hand does, or it doesn't fall at all.

First block, dismantling

No keyboard. No hand on the machine. Three and a half hours of theory and demonstrations.

The theory holds in a few strokes. A large language model is a system that statistically predicts the next word in a sentence, drawing from billions of texts it has read. No internal knowledge. No intent. No memory of its own. A probability that comes out of a calculation. That single line is enough to shift the locus of control. The machine stops being an oracle. It becomes a tool again.

Then come the demonstrations. Deliberately ambitious, deliberately broad, deliberately beyond what the participants would have come for. A deep search on their sector, sourced, compared, cross-checked. A two-voice podcast assembled in minutes from one document. An infographic taking shape in front of them. A song, a simulator, a visual report. I show them things they would never have imagined being able to do.

I also show them the cross-prompt, my method for producing a new press release from an old one and a document describing the substance of the new one, without rewriting the form by hand. The output is sound, the tone holds, the house style is respected. Someone in the room always says, at that moment, "no way."

The target isn't performance. The target is astonishment. Wonder. The raw emotion of being able to do what you'd never imagined being able to do. That emotion is what brings the wall down. Practice comes when the wall is gone.

Second block, the first hand

The afternoon of day one, hands hit the keyboard. Three and a half hours of practice. Not for what amazed them in the morning. For something more modest.

I work the posture, the "AI collaborator" against the oracle. I work the CONTE method, which structures the prompt as a contract with a brilliant amnesic intern. I have them rewrite a tricky email they had to send last week, or summarize a twelve-page document they've been dragging for two weeks. The target is the first experience of success by their own hands, the moment when the thing they'd been putting off resolves itself in five minutes.

And I slip in, regularly, small surprises. A tool that produces a diagram from their text. Another that turns a paragraph into a short video. The room laughs. Someone keeps saying "no way" and a neighbor answers "I swear." The morning's wonder plays out again, this time it's their hand triggering it.

By five-thirty, they're packing up. Many of them say it: "I got some of it, but I'm not sure I know how to use it yet." That's exactly the right state of mind for what comes next.

The night

What happens between the two days is invisible and decisive.

I know because the participants tell me about it the next morning, almost all of them, without anyone asking. Someone tried it on a personal text the night before. Someone else got the machine to write a message to their union rep. A project manager showed the tool to her teenager, prepping for an oral exam. A director generated a clean registration form for his tennis club, something he'd been putting off for three months. No one told them to do that. They did it on their own, in the blind spot of the organization that sent them to be trained. That's what I call BYOA, Bring Your Own AI, and it's also what I describe at the macro scale in The revolution that never happened.

The night carries no pedagogy. It carries what pedagogy cannot manufacture.

The first time you try it on your own, because you want to. Desire has just been born. It won't sit still. It overflows the training frame and it overflows the work frame. That's exactly what had to be engineered.

Third and fourth blocks, playing to remember

On day two, seven more hours of practice, with difficulty rising in steps. On the surface, a classic program: real cases, complex professional writing, multimodal output, custom agent creation. Inside, we play.

We write a sketch about a steering committee they hate. We compose a Suno song about a reporting task they've been dodging for six months. We have the machine write a rap about their employee handbook. We laugh a lot. The room is louder than the day before.

It's not a flaw. It's a device. The review by Tyng and colleagues, published in 2017 in Frontiers in Psychology, synthesizes decades of research: emotion modulates memory consolidation to the point of being a central factor in any durable learning.^[1] Content acquired in joy, surprise or laughter is remembered longer than content acquired in neutrality. Laughter isn't decoration. It's the most effective pedagogical tool I know of.

In parallel, I sharpen their critical thinking. I teach them to read what the machine produces. To name what doesn't work. To put words on why. The cross-prompt method they saw demonstrated on the morning of day one, they now practice on their own files. They discover that the machine is at their service, that it follows, word for word, what you know how to ask. The skill isn't to prompt better. It's to know how to put words on what you want, and on what you don't want.

I also weave together the personal and the professional. The same tool that drafts a meeting summary drafts a song for a school play, a note of apology to a neighbor, an administrative letter no one likes writing. And I do it on purpose. Because personal use reinforces professional use, and the other way around. The participant who masters the tool at home in the evening, helping their kid prep a presentation, is the same one who'll dare to use it on Monday morning on a hard file.

And I tell them one last thing, in closing, that they weren't expecting. AI isn't for everything. It's for shutting off the painful, repetitive, time-eating tasks, the ones that wear you down. Not for shutting off the ones you love, the ones that bring well-being to your work. Using AI also means knowing when not to use it. And the best way to use AI is still to spend less time on a computer and on AI.

And that's where the job changes.

Not in the learning of new techniques. In the discovery that you can build the tool that didn't exist. A project manager walks out with an assistant that drafts her meeting notes in the exact format her department uses. A director walks out with an agent that summarizes his weekly meetings according to his own grid. A field worker walks out with a GPT that prepares his rounds based on criteria no software vendor would have thought to code.

They came in to avoid being replaced. They walk out understanding that the tool lets them go further on what they already do well, better on the dull stuff, and keep for themselves what they love. Well-being at work goes up. It doesn't get lost.