I believe Java virtual threads/project Loom fits what you are describing. No separate async APIs, everything is coded using a thread based model. The user decides between using platform/OS threads (thus delegating scheduling to the kernel), or using virtual threads and letting the JVM take over scheduling.