I disagree with the original essay comprehensively. async/await is less composable (threads automatically compose by their nature; it is so easy it is almost invisible), a tie on expressive (both thread advocates and async advocates play the same game here where they'll use a nice high-level abstraction on "their" side and compare it to raw use of the primitives on the other, both are perfectly capable of high-level libraries making the various use cases easy), and I would say async being more "correct" is generally not a claim that makes much sense to me either way. The correctness/incorrectness comes from things other than the runtime model.
Basically, async/await for historical reasons grew a lot of propaganda around how they "solved" problems with threads. But that is an accidental later post-hoc rationalization of their utility, which I consider for the most part just wrong. The real reason async-await took off is that it was the only solution for certain runtimes that couldn't handle threading. That's fine on its own terms, but is probably the definitive example of the dangers of taking a cost/benefit balance from one language and applying it to other languages without updating the costs and the benefits (gotta do both!). If I already have threads, the solution to their problems is to use the actor model, never take more than one mutex at a time, share memory by communicating instead of communicating by sharing memory, and on the higher effort end, Rust's lifetime annotations or immutable data. I would never have dreamed of trying to solve the problems with async/await.
Basically, async/await for historical reasons grew a lot of propaganda around how they "solved" problems with threads. But that is an accidental later post-hoc rationalization of their utility, which I consider for the most part just wrong. The real reason async-await took off is that it was the only solution for certain runtimes that couldn't handle threading. That's fine on its own terms, but is probably the definitive example of the dangers of taking a cost/benefit balance from one language and applying it to other languages without updating the costs and the benefits (gotta do both!). If I already have threads, the solution to their problems is to use the actor model, never take more than one mutex at a time, share memory by communicating instead of communicating by sharing memory, and on the higher effort end, Rust's lifetime annotations or immutable data. I would never have dreamed of trying to solve the problems with async/await.