An Agent reference can be pid or name.

Each function of the Agent module takes an agent as the first argument:

get(agent, fun, timeout \\ 5000)

but this can be a couple of different things. The documentation defines the agent type as:

agent :: pid | {atom, node} | name

#The agent reference

name ::
  atom |
  {:global, term} |
  {:via, module, term}

#The agent name

So really five different types of values are valid. The pid is a value returned from calling start_link.

> {:ok, agent_pid} = Agent.start_link(fn -> [:thing] end)
> Agent.get(agent_pid, fn (state) -> hd(state) end)
:thing

The agent can be referenced by a name atom also. In the documentation example that name atom is the __MODULE__ for the module that wraps access to the agent. The name for the agent is declared by passing a name option into the start_link function call.

> Agent.start_link(fn -> [:thing] end, name: __MODULE__)
> Agent.get(__MODULE__, fn (state) -> hd(state) end)
:thing

I'm note sure about the {:global, term} and {:via, module, term} values that seem to be valid. I would love to see examples for those.