Threads vs. Tasks
You can think about tasks in a variety of ways. One is to start from the notion of a thread and state that tasks are like lightweight threads, running on top of a pool of low-level operating system threads. This is quite an implementation-oriented definition that starts from machine-level mechanics and builds an abstraction out of it.
A better way to think about tasks is to see them as representations of future values that get computed by means of a function. As such, tasks are sources of asynchrony in your code. While the task is running, you can do other useful work, until the point you want to await the task’s result. Furthermore, tasks can be composed in a variety of ways. In the definition in the preceding paragraph, the word future is used intentionally. Computer science literature often refers to this mechanism as futures (“in the future, there will be a value”) or promises (“the tasks promise to compute a value”). Contrast the data oriented nature of tasks with regular threads to see the difference. Tasks provide a much nicer abstraction to perform computation in the background, allowing the result to be obtained in a variety of manners at a later point in time.
Tasks should really be thought of as the new threads. Disadvantage inherent to threads: their cost to create. The thread pool is used to reduce such costs and even to avoid excessive context switching by multiplexing multiple units of work onto the same physical operating system thread that gets reused. The introduction of tasks makes direct use of threads largely obsolete because they can be used for both long-running work and for short-lived operations. The latter case clearly is where you would have been using the thread pool before, although it suffers from limitations such as the ability to return values and compose operations. Tasks are a much better abstraction. Using the thread pool for long-running work used to be problematic because you were stealing away a worker in the thread pool for a long time. The engineering done to the task infrastructure (and the underlying revamped thread pool on top of which it’s built) means that tasks can now be used for long-running work, too.