What we’ve learned at IdeaTech
The great technological revolutions of the past — steam, electricity, computing, the internet — share a property we take for granted: they were always fast for markets, but slow for people.
The first, steam, lasted a century. Those who were coachmen died as coachmen. The next generation dealt with steam horses instead of four-legged ones. Steam changed generations of workers, not individuals within their own lifetimes. The same was true for electricity, for computers, for the internet. Each faster than the last, yes — but still within a rhythm the human brain could absorb.
AI has broken that property.
It’s the first time in history that a technology evolves at a speed the human brain, on its own, cannot keep up with. It doesn’t ask for a new generation of workers. It asks the same individual, within the same working life, to shift paradigms more than once. It’s not just faster than what came before — it operates on an entirely different unit of measurement.
Everything we’ve learned at IdeaTech over these two years is an attempt to manage that fracture.
In a previous article I wrote about the two AI coordinates: individual and organisational. Here’s what happened while we tried to build both — and what we understood that we didn’t know before.
The personal coordinate: those whose eyes light up
The personal coordinate is about the individual learning to think with AI — not to use it the way you use software. It’s a permanent skill, like Excel was, with one difference that changes everything: you learned Excel once. AI changes under your feet while you’re learning it.
The first mistake on this coordinate came quickly.
A meeting room like any other. About twenty people, a working lunch together, a few laptops open. Me at the front with the slides. I’d called it “Lunch with AI” — trendy name, simple concept: explaining to colleagues what I’d discovered about artificial intelligence. Demos, use cases, possibilities. Genuine enthusiasm on my part.
Polite attention on theirs. Zero contagion.
In that moment I sensed something that took me months to fully understand. I thought you could transfer AI with a couple of slides recounting a personal experience. You can’t. Against a pace like AI’s, at the individual level, you can’t train people. You can only find those who are already switched on.
I stopped looking for people I needed to convince and started looking for people who didn’t need convincing. Not the most talented. Not the most senior. The ones already exploring outside working hours, with tools that offered no measurable advantage — purely to understand. The ones whose eyes lit up when it came up in conversation.
I brought them together.
In a recent articIe wrote that from the outside it looked like a cult — and maybe at the start it was a little. What I didn’t say in that post was that the choice ran counter to instinct compared to how you normally build a group inside a company. Normally you build a group with an objective, a metric, a deadline, and names that make sense on paper. This was none of that. No objective, no metric, no deadline. The only thing in common: sharing what each person had discovered. Much more listening group than operational team.
It seemed like a contradiction. For months I wondered whether I was just taking time away from people who could have been producing elsewhere. The answer came slowly — not from a metric, but from a phenomenon: the converts from the opposition. People who had been firmly against AI, who at some point sat down, tried it, and became the most convinced supporters. Not because we’d convinced them. Because they’d found their own reason.
That moment, every time it repeats, is the signal that the culture is real. It can’t be taught — it spreads. And it only spreads if there are people with the fire already lit, who just need a space to let others see it.
The organisational coordinate: the chaos and the ERP
On the second coordinate we made the biggest mistake.
We already had Slick — structured processes that worked. And when you have a solid foundation, the temptation is natural: take every block and layer AI on top. Opportunity management, sales management, software analysis, architecture, project management, task assignment for developers, coding. For each one, one or more agents.
The plan looked solid. It was chaos.
Every agent produced output. The outputs became documents. The documents multiplied. People were spending more time validating what AI produced — checking whether the agent’s reasoning held up, refining its outputs, making the steps add up — than doing the actual work. We’d built a text machine, not a work machine.
We went back to the drawing board. It wasn’t easy — when you’ve already invested, when you’ve already communicated the direction, stepping back feels like defeat. It isn’t. It was the hardest — and most necessary — decision we made on the second coordinate.
In stepping back, I recognised an old mistake dressed up as something new. It’s the same mistake struggling companies made when they decided to adopt a new ERP — convinced the software would “force” them to follow processes. As if software came bundled with culture and operational discipline. It didn’t work then. It doesn’t work with AI now. AI doesn’t create processes that aren’t there — it amplifies what exists. For better and for worse.
And here the rhythm fracture returns. Changing the process isn’t enough: you have to change the way people work. And that requires the same cognitive effort as always, regardless of how fast AI runs. For every block. Multiplied by every person. Go too fast and you get chaos. Wait for perfection and you miss the train.
We started again, one block at a time. Understanding where AI genuinely adds value — not where it seems obvious that it should. Stabilise it. Wait for it to become normal. Then move to the next.
The compass
From these two experiences — the cult that lit itself up and the agent chaos — came the only strategy that seems to hold. I call it a compass because it doesn’t say where to go. It says what to ignore.
Nobody can run at the speed of AI. Nobody. Anyone who says they can is selling something. So “AI first” can’t mean adopting everything that comes out. It has to mean the opposite: having principles that hold regardless of the tool of the moment, and a compass for choosing what to integrate when the tool changes. And it always changes.
Culture, for us, isn’t there to help us adopt faster. It’s there to help us judge better. Structured processes give us the place to put what passes the filter. Together, they mean we don’t chase the market — we choose what to integrate, when, and why.
AI first isn’t adopting everything. It’s knowing what to ignore.