The idea is you need to understand your workload and run tasks on schedulers meant for that workload. You'd make sure to move that work to a context for long running tasks.

Not unlike how you might use a non-virtual thread pool in Java.... but it seems wrong to imply that you no longer need to think about this stuff.