Every speaker seems to be saying that that concurrency should not be explicitly managed and that thread management should go the way of explicit memory management.
I agree but for some reason I feel uneasy about it. I suspect because I wonder about how people can effectively reason about code with non-deterministic execution order.
That was one thing that was interesting about the GPars syntax that consisted of declarations of identity, execution tasks and then datafeeding and shutdown code. The tasks are described without identity so it is clearer that you can't assume what is happening with the task except through its output.