If AI solves a programming problem, it automatically solves any other problem.

I'm not saying that it will break the laws of physics, but almost everything short of. That's why in my primitive experiments and more research I primarily focus on models and agents that can automate programming.

And in this sense, there is not much difference between a model that can perfectly and on the first try come up with and create any program and a full-fledged AGI. Any task of interacting with the physical world (using a robotic hand to pick up a pen from the table, wash a toilet, catch and spank all the scoundrels in the world, build an optimal economic system, coordinate markets) is an engineering task that consists of components that are understandable to us: reasoning, planning, computer vision, mechanism design and cryptography, the list can go on forever.