Adoption of AI is slow. It is slower than expected and the documented productivity gains are negligible, spawning fears of an AI bubble or at least that we have passed the peak of inflated expectations. While others might see cause for concern, tech optimists (and especially tech investors) view these developments differently. For them, this is the ‘calm’ before a massive AI deluge—a sign that when the disruption finally hits, it will be instantaneous, all-encompassing, and impossible to ignore. An example is a recent post from a tech entrepreneur whom I respect a lot, Martin Thorborg. He marshalls the go-to tech optimist credo: this is an exponential trajectory, and thus seems slow at first but then very quickly grows at a speed beyond imagination. This is a variant of Ray Kurzweil’s Law of accelerating returns.
The idea is that even very small increases become very large quickly if they are exponential, where the rate of change increases in proportion to the value of the function, ie. it is not linear as the growth we usually see. The problem with this view is that in nature, as in real life there are no true exponential functions. Bacteria can grow exponentially until they don’t have any more food. Viruses can spread exponentially in a population until the population develops immunity, There are no pure exponential functions in nature, no J shaped hockey stick. The exponential functions we see in nature are always S-shaped, part of a sigmoid function that has an exponential phase, which eventually gives way to a stagnant phase as it approaches its carrying capacity or technically its asymptotic limit. There will be a saturation point where a phase transition will occur.
The real discussion then, if we agree that AI is in an exponential growth phase, is not how amazing and wonderful it will get very quickly and very soon, but when AI hits its asymptotic limit and levels off into a steady phase.
I believe we have already hit it. At least for now. The boundaries or the asymptotic limit, of AI is not technological; it is defined by AI adoption. This is in fact a simple observation, no matter how advanced a technology is, it makes no impact if it is not adopted and used. That shifts our focus to the real challenges for AI. I spent many years thinking about this based on my experiences trying to implement AI and even wrote an Book on this subject. The hypotheses from this book were validated in a survey of Nordic AI usage that I led in 2025, parts of which are published in a whitepaper. Nordic organisations’ adoption of AI is challenged by many factors, but the technological capabilities is the least frequently cited limitation to use of AI. The most frequently cited limitations include solution characteristics, such as lack of transparency, unpredictability, next were human factors. These factors are the real boundaries that are defining AI adoption.
That does not mean that the exponential phase cannot be extended. It can. It is just not extended due to increase in technological capabilities, such as more impressive and faster LLMs, it is expanded by a greater ability to adopt and use AI. And that is achieved only by addressing the real challenges that organisations cite, not by reading science fiction and dreamin along with Ray Kurzweil, Elon Musk (full autonomy next year) or over optimistic tech invetstors. The AI revolution, and the extent of the exponential phase, is decided on the ground by addressing the real problems of real organisations use-case by use-case. So, let’s stop dreaming and let’s get started leveraging AI for a positive human future.
Photo by Saikiran Kesari on Unsplash