An asynchronous agent is a sequential composition of asynchronous operations. Every asynchronous operation is considered to run as a part of an asynchronous agent, even if that agent contains only that single operation. An asynchronous agent is an entity that may perform work concurrently with other agents. Asynchronous agents are to asynchronous operations as threads are to synchronous operations.

However, an asynchronous agent is a purely notional construct that allows us to reason about the context for, and composition of, asynchronous operations in a program. The name “asynchronous agent” does not appear in the library, nor is it important which concrete mechanism^[1] is used to compose the asynchronous operations in an agent.

We can visualise an asynchronous agent as follows:

Asynchronous agents alternately wait for an asynchronous operation to complete, and then run a completion handler for that operation. Within the context of an agent, these completion handlers represent indivisible units of schedulable work.