This is incorrect. async/await as implemented in Swift requires runtime support, including new runtime entry points to manage tasks and actors, changes to core runtime functionality such as type metadata, and tie-ins with the OS kernel. Some of these are covered in the behind-the-scenes talk.
Despite any surface-level similarities, Kotlin's technology stack is completely different from Swift's. Comparison at a technical level is not likely to be meaningful, especially given that the technical challenges for back-deploying a Swift Concurrency are mostly understood.
The performance and semantic impact of doing this is hard to evaluate without having done all of the work to back-deploy. This is one of several "unknowns" that would require detailed evaluation. It could very well end up being the case that it is possible to back-deploy, but the experience of running on older OS's is poor.
There's some description of the technical issues above, as well as the aforementioned blog post on evolving in an ABI-stable world. This is not the first time that a new language feature has not been available on previous OS versions: opaque result types came in Swift 5.1 and are not available on Apple OS's prior to iOS 13 / macOS 10.15 Catalina, due to the same technical issues.
Frankly, I don't think any more technical details will change the discussion in any way. I could write many paragraphs about all of the issues we face, the approaches we think will work, and the unknowns still out there, but the best possible outcome from doing that is that I spend a whole lot of time writing it and the takeaway is "yep, that's a bunch of work, they weren't kidding." There is no silver bullet, no singular insight, that will suddenly make the work to backward-deploy become trivial.
It is completely reasonable for folks to ask about the limitations and to report on how the lack of backward deployment affects them. As Ted said, those of us at Apple hear you, and we appreciate the feedback. It's fine for that to be the entire point of this thread.
Doug
[EDIT: Another post came in, so I've added a paragraph about performance/correctness considerations]