One of AI's greatest challenges today is not the model itself. It is its adoption: the friction between humans and machines.
Today's AI models alone are enough to develop our industry for the next ten years. A significant part of Silicon Valley is well aware of this and would agree to slow down, but geopolitics decides otherwise.
The gap between the raw computing power available and what is actually deployed in our society is already considerable. Most companies only use a fraction of what the models can already do, and the broader public is barely beginning to grasp what is at stake.
Piling on more power will not close this gap. At best, it widens it.
Each new generation of model widens the gap between what is technically possible and what is genuinely usable day to day. We are accumulating capacity that no one has the time to learn how to exploit.
Let us take the time instead to properly adopt these tools. There is no need to wait for the next model to get serious results. A well-designed system, built with lighter models used in the right place, often performs as well as the latest releases, sometimes better.
The real margin for progress today is not a more powerful model. It is the mastery of the system that runs it.