Why I Carry the Title Chief Imagination Officer

I think back to the pandemic years. The time that suddenly existed, and what it got spent on. For my part it got spent in the browser, in Google Colab, where anyone with a Google account could borrow a GPU and train language models on goodwill and patience. And in OpenAI's Playground, when they first let us in to the large models. DaVinci, finishing sentences with an uncanny matter-of-factness. Disco Diffusion, taking twenty minutes to paint a picture of a single line of text. It felt like opening a door onto something magical. Far past what had been fun about LSTMs and the early generative models, which mostly resembled interesting digital acid trips.

Now, five years later, I see it was never about play. It was an exercise in exponential thinking.

Those of us sitting there with notebooks and wait times did not understand that we were calibrating our intuition. Every time something impossible became possible, and then trivial, within months, the linear gut feeling got overruled a little more. GPT-2 demanded fine-tuning, a GPU lottery, and patience. GPT-3 demanded a sentence. Disco Diffusion demanded a dictionary of parameters and a night in the queue. Stable Diffusion demanded a second. The distance between these points is not a story about better tools. It is a curve.

Ray Kurzweil has been my companion in this long before it was comme il faut to cite him. What people remember is the singularity, and that is a shame, because the most important thing in Kurzweil is far more down to earth: humans systematically underestimate exponential change because we are built to extrapolate linearly. His image is the chessboard where the grains of rice double on every square. The first thirty-two squares look manageable, and we think we have seen the scale. But it is the second half of the board that counts. The pandemic experiments were the first half. We are on the second half now.

That is why my title at SkyeTec is Chief Imagination Officer. It tends to earn a raised eyebrow, and that is part of the point. The title is not a gimmick. It is a thesis about where value is moving.

For two hundred years, competitive advantage has lived in execution. Knowing how, faster and cheaper than your neighbor, was the game. That game is ending. Autonomous systems absorb how at an exponential pace. Code that took a team a quarter takes an agent an afternoon. Analyses that justified an entire department fit inside a prompt. When how becomes a commodity, it stops being the place you win.

What stays scarce is intention. Knowing what should be done, what is worth doing, and what should never be done. That is not an analytical skill. Analysis optimizes within the possible. Intention requires imagining something outside it, and imagination is precisely the muscle that linear thinking lets wither. If your picture of next year is this year plus ten percent, you do not need imagination. If your picture of next year is shaped by a curve, imagination becomes the most operational capability you have.

That is why I treat imagination as a discipline, not a decoration. At SkyeTec we try to build it into the culture the way others build in quality or safety. In practice it looks like this: every plan is stress-tested against the curve, not the tangent. We ask what the plan assumes about model capability, and what happens to it if capability doubles before the plan is finished, because it usually does. We prototype futures instead of debating them, because a working demo of an absurd idea teaches us more than a workshop about a sensible one. And we hold imagination accountable for business results. Without anchoring, imagination is science fiction. Anchored in innovation and results, it becomes direction: the thing autonomous systems cannot deliver, because they execute intentions but do not create them.

That triangle, imagination, innovation, results, is the real job description. Imagination finds the doors. Innovation opens them. Results prove the room behind was worth walking into. Most organizations have the last two and starve the first, then wonder why their AI strategy looks like everyone else's: same use case, same pilots, same incremental ambitions. They extrapolate the tangent in an age that compounds.

Europe is, slowly, learning the same lesson at continental scale. The Commission's new Cloud and AI Development Act is many things, but at its core it is an act of intention: a decision about what Europe wants to own rather than rent. I read it as a signal that the question we ask internally at SkyeTec, what is worth doing once doing becomes cheap, is on its way to becoming the question everywhere.

But owning is not the same as owning the frontier model. Those weights are expensive, they age in months, and the race for them is not where a small country wins. Sovereignty lives a layer higher up, in the harness: the orchestration, the governance, the context, and the tools that turn a raw model into a capability someone dares put into production. That layer we build ourselves, and because it is model-agnostic, we can place each task on the right model. When we need raw capacity, we rent a frontier model. When sovereignty weighs heaviest, we run a fine-tuned or distilled open-weight model on metal under our own jurisdiction, where we own the weights, sign the mandate, and log the run ourselves. Not an infrastructure decision dressed up as strategy, but an intention that has been given somewhere to live.

The pandemic years taught some of us how it feels when a door opens, and how fast the room behind it grows. The next doors open faster. Execution will walk through them on its own.

Deciding which doors, and why: that is the job now. That is why imagination belongs firmly planted in the leadership group.