I agree with this perspective. Focusing on what AI cannot do today misses the bigger picture. Adoption is being driven by access, and with billions of people connected through smartphones, distribution is already solved. Capability always follows.
The 10x yearly improvement estimate makes sense when you look at the full stack. It is not just models improving, but also GPUs, inference optimization, data efficiency, and deployment. When all of that compounds, linear thinking completely fails.
The declining cost per token is especially important. Once models with near–frontier capabilities can run locally or cheaply at scale, AI stops being a premium product and becomes infrastructure. At that point, progress feels sudden, even if it was inevitable.
Speed is the real takeaway. This is no longer Moore’s Law territory. A 10x yearly curve changes everything, and most people are still underestimating how fast that reality arrives.