Outside Android, Kotlin will always be playing catch-up with platform features, having two ways of doing the same thing for features introduced before the platform went their own way, FFI that only goes one way, the native variant is being redone due to the clever idea of being originally incompatible with JVM memory model, it is officially a way to sell InteliJ licenses.
There is a Herb Sutter talk at CppCon about the downsides of reference counting.
Basically the solution to those problems is to make use of hazardous pointers or deferred destructions on background threads, which are a kind of lightweight tracing GC.
Besides which I don't want to have to keep track of weak references, at that point I would rather just have a full ownership model: https://github.com/apple/swift/blob/main/docs/OwnershipManif...
Swift has the benefit of actual value types, no GC, and IME it's much faster. But it only runs on Apple's OS (technically there's support for Linux but people say it isn't good). Kotlin has access to all Java packages and runs on the JVM, and with Kotlin multiplatform JetBrains is trying to make it run on almost anything.