This is really cool, but it's already slightly out of date... because of course it is, with the speed of AI development :D . Google's "Inner Monlogue" system is built upon SayCan, and it allows the AI to not only think through solving a problem, but to reason through it one step at a time, making it robust to sudden unexpected changes (a change in request, an inability to complete a step it had planned, a change in the environment, even a change in the language of the request).
We're currently in the nascent stages of what I'm confident will directly lead to the first AGI, and while I'd estimate that's about 10 years away at minimum, the staggering acceleration of AI research has me optimistic that it might even be sooner than that.