Nobody tinkers in front of a boss while pretending to already know.
It took me an embarrassingly long time to figure out that this was the actual problem I was hired to solve. Not skills. Not tools. Not access. Shame.
Here’s the session I keep replaying.
Standard room, standard afternoon, about thirty people. We were maybe forty minutes in. I’d asked some open question about whether anyone had tried using these tools for anything real, expecting the usual pause, and a guy in the third row said something so quietly I asked him to repeat it.
He’d been using AI for months. Daily. He had a whole system. He had figured out, entirely on his own, with no training and no permission and no one to ask, how to make a genuinely tedious part of his job about eighty percent faster, and he had told exactly zero people at work about it.
He said it the way you’d confess to something. Voice down, shoulders in, a little half-shrug at the end that meant I know, I know. He was braced. He fully expected the room to look at him like he’d been caught cheating, or worse, like he’d been caught being one of those guys.
Instead, the woman next to him said, “wait, you did what?”
And that was it. Within five minutes there were six people around his laptop. Not near it. Around it, chairs dragged, one person kneeling. I stood up front, at the podium, with my slides on the screen, and I stopped talking, because six people were out-teaching me and it wasn’t close.
I wrote down what I heard. The actual lines are better than anything I could paraphrase:
“Here, let me show you.”
“Wait, it can do that?”
“Just talk to it like a friend, it’ll understand.”
That last one is better pedagogy than anything in my deck. I have a whole slide about conversational prompting with a diagram on it. He said it in nine words while pointing at a screen, and everyone got it.
Every one of us, at work, is running a low-grade continuous performance of already knowing. It’s not dishonesty. It’s just the ambient requirement of having a job. You are paid, in part, to be a person who knows things, and so the muscle memory is to nod, to have a take, to be up to speed. Fine. Mostly harmless.
Except that AI arrived and made all of us beginners at the same time, and the performance did not update. So now you have entire floors of people who are quietly experimenting at eleven at night, alone, ashamed, and hiding the exact experiments that everybody else in the building desperately needs to see. The learning is happening. It’s just happening in a hundred sealed rooms with the doors shut.
Curiosity is a permission structure. You cannot install it with a policy memo. But you can grant it, in a room, by being the guy at the front who says out loud that he doesn’t know either.
What comes out the other side, once permission exists, is not abstract.
Graham Dean, an SVP who runs manufacturing ops in the UK, wrote to me after a session. I’ll quote him directly because his sentence structure is doing something my summary couldn’t: “I ‘played’ on the Ops scorecard which was my frustration and very quickly built the comparison tables I normally build manually, I got it to conditional format and then got graphs built. All basic stuff but I could not have done this before I had time with you both.”
Look at the quotation marks he put around played. Look at “all basic stuff,” a senior executive preemptively minimizing his own win before anyone else can. And look at what he actually did, which was take a thing that annoyed him every month and make it stop annoying him.
I forwarded that note to my team with two words on top: “One for the team fridge.”
And then there’s the smaller one, which is the one that actually gets me. Somebody in a session described a recurring task they had genuinely dreaded, week after week, for years, and how it had become simply a done thing. Then they said the part I’ve quoted back to people a dozen times since: it makes a kid’s swim meet a “yes” instead of a “maybe,” instantly.
That’s the whole business case, if I’m honest. Not the productivity math. A parent in a folding chair in a humid natatorium, on a Tuesday, because Tuesday opened up.
A few months ago I was building out an AI fundamentals curriculum, feeling reasonably good about my craft, and I had a thought I did not enjoy: Copilot could probably write this. So could Claude. Maybe better. Certainly faster, and without the three days I spent rearranging module four.
I sat with that for a while, in the way you sit with a thing that has just walked up and taken your lunch.
The content was never the scarce part. It hasn’t been the scarce part for a long time, and I was slow to notice. What I actually bring is the room. The ninety seconds of safety. The moment where a guy in the third row decides that saying the true thing out loud is survivable. No model generates that, because it isn’t information, it’s social weather, and social weather requires a person standing in it with you.
I’m not the instructor. I’m the permission-granter, and the instructor is already sitting in the room, quiet, in the third row, waiting for someone to make it okay.
The best moment in any session I run is the moment I become the least useful person in it.