The great compounding
Three months after ChatGPT launched, I watched a room full of Google executives quietly reorganise their lives around it. That, not any demo, is what convinced me. Two and a half years later I'm all-in — building AI-native systems, and thinking hard about what compounding intelligence means for science, for society, and for business. This is where I've landed, and why everything else I'm publishing starts here.
A room that changed my mind
In February 2023, about three months after ChatGPT first went public, I was on a work trip across the US and Canada. One of the stops was Google.
I arrived sceptical. I half-expected another hype cycle — a shiny distraction pulling attention away from the real, unglamorous work we needed to do in retail. I'd seen technology waves oversold before.
What changed my mind wasn't a demo. It was the people. We met members of Google's executive team, and they weren't pitching. They were rattled, in the best sense. One of them said, more or less: I'm deleting meetings out of my calendar. This is a once-in-a-lifetime turning point. I need to hunker down and learn everything I can.
Watching serious operators quietly reorganise their own behaviour around something — that lands harder than any product launch. If the people closest to the engine were rearranging their lives, I should pay attention. I've been paying attention ever since.
What two and a half years taught me
The thing I keep coming back to is not any single capability. It's the compounding.
It starts small. You automate a fiddly task. Then that automation becomes a building block for the next one. Then the blocks start combining in ways you didn't plan, and the whole thing accelerates — bigger, faster, stranger than the month before. I've felt this directly, building my own tools and businesses, and the slope only steepens.
Here's the part that's easy to miss. The technology doesn't just improve itself. It improves its own inputs. AI is already reshaping how we think about energy and energy storage, about silicon and compute, about language and mathematics — and about AI models themselves. A revolution that accelerates the very conditions for the next revolution.
I have zero doubt we're living through a once-in-a-lifetime change. The only joke in that phrase is that compounding probably makes it an understatement — this turning point should pull the next one forward, too.
What it's actually for
It would be a small imagination that reduced all of this to better productivity tools and a slightly higher standard of living.
I think Demis Hassabis has the right frame. The real prize is the chance to understand deeper truth. Physics still has no unifying theory. Our most basic conceptual tools — mathematics, probability, statistics — may simply not be good enough to explain even the reality we can observe, let alone what sits beyond it. The most exciting possibility isn't that machines do our work. It's that they help us see what we currently can't.
That's the horizon. Most days, though, I'm thinking about three things closer to the ground.
One: science
This is the deeper-truth opportunity, and it's the one we talk about least. If intelligence really does compound the way it appears to, the frontier isn't a better chatbot. It's new physics, new biology, new mathematics — a genuine expansion of what humans can know. I find that more thrilling than anything on a product roadmap.
Two: society — the question that worries me
This is the hard one, and I don't have it solved.
I've argued about universal basic income with the same circle of friends for the better part of two decades. One of them actually received a basic income through an experiment. He went in a believer and came out a sceptic — and his reason has stuck with me. He decided that purpose comes from being useful to society. “Jobs”, “showing up”, “earning a living” are all abstractions over something more primal: a real contribution to make, something to grind at, somewhere to prove yourself.
In a world where productivity skyrockets and scarcity recedes, I genuinely cannot yet see how that's straightforwardly good for most people. Abundance is not the same as meaning. That gap is the thing I worry about most.
I worry about two other things alongside it. The first is alignment and control — whether we can actually steer what we're building. The second is how short-sighted so much of business and geopolitics is being. A lot of very powerful people are convinced they're the good guys while behaving recklessly. That's true in the US, and it's true in China.
None of this makes me a pessimist. It makes me think the people who are optimistic and clear-eyed should be in the room, not on the sidelines.
Three: business — where I actually build
I'm most useful here, so this is where I spend my hands.
My conviction is simple: AI-native companies will pull away from everyone else. The advantage is structural — speed, cost, growth — and it comes from architecture, not from bolting a chatbot onto an old operating model. I've been testing this with Best Mate Meals, a fresh dog-food business I run as a near single-operator, AI-native company. What one person can now design, operate and grow is, honestly, incredible.
Two things temper my excitement with realism.
First, adoption will be slower than the technology deserves. I lived through e-commerce. It never went as far or as fast as the believers predicted, even after COVID gave it a once-in-a-generation shove. My honest guess is that in five years, maybe 20 to 40 percent of commerce is genuinely AI-native — not the 80 or 90 percent the capability would justify. The wave should be bigger, deeper and quicker than it will actually be. There's a long window here for the people who move early.
Second, the nature of work changes more than the org chart admits. A huge amount of what large organisations do is administrative scaffolding — meeting protocols, coordination, status management, the machinery of keeping people roughly aligned. Much of that simply dissolves. To make it concrete: I built a tool that takes my meeting recordings, separates the actual decisions from the action points, drops them into my productivity dashboard with zero touch, and wires up an AI session by webhook to start working the agreed actions — before I've even walked back to my desk. That used to be three people and a week. It's now ambient.
Where I've landed
So that's where I am. All-in.
I want to be part of this change all day, every day — building it, and helping nudge it in the best direction I can manage. I've had a meandering, fortunate career: a Swiss village, a decathlon runway, international affairs, consulting, founding, a decade inside big retail. I used to think the through-line was strategy. I now think it was always this: trying to understand how a system really works, and then building the version that works better.
This essay is the starting point for everything else I'm publishing — the podcast, the writing, the tools I'll put out into the world. If you only take one thing from it, take the compounding. It's the part most people are still underestimating, and it's the part that changes everything.