Why not both? The URL is just a URL - it does not need to reflect your underlying data model. There would probably be a canonical URL for use when the category context isn't available (say, "consoles"), but why not have multiple URLs lead to the same information?
> "despite the appeal of the OP's argument, one can hardly deny the success of Amazon's product listings in spite of their ugly URLs"
But they're really not ugly. In fact, given the complexity of the system they represent, they are remarkably human-friendly.
In an ideal world all ideas, all businesses, and all use cases can be fulfilled by simple URLs like "example.com/shockingly-unique-identifier", but we don't live in that world. Amazon has constructed human and machine-relevant URLs. The author's argument can be applied to many sites, but I don't think Amazon is one of them.