This
"waterfall had all this effort go into a product only to get feedback at the end" was Winston Royce's own diagnosis of the problem with the waterfall model he formalized.
That's part of it. In more practical terms, assigning people to tasks tactically as a project progresses is probably 80% of the advantage of Agile.
You are also spot on that you can start an Agile project and find your architecture is a pile of poo 2/3rds of the way through, in part because you can start without the waterfall set of specs.
A lot of the dissatisfaction in the "Scrum is cancer" article is about estimation, and I agree: Trying to improve estimation is a fool's errand. Your highest risk tasks are generally what create the most value, solve the hardest problems, etc.
More emphasis on retrospectives, and accepting that estimation fails most of the time, is what a lot of Scrum-as-a-cargo-cult misses.