Channels are a safe way for goroutines to communicate. Safe here means, that the channel itself performs all the necessary locking needed for any number of goroutines to read or write "at the same time" to and from the channel. Other than that it does behave like a plain pipe. You write into it and can read from it "on the other side". Creating a channel does not create a goroutine. A goroutine is only created if you use the "go" command, like "go f(42)". This would run the evaluation of f(42) in a separate goroutine, which behaves like a thread, but uses less resources. So in neither of your examples, goroutines are created. You can optionally add buffering to the channel. A write to a channel without a buffer blocks, until the value you write to the channel is read. If you have buffers, you can fill them without blocking, even if the values are not consumed immediately. Without the buffer, the first example would get stuck, as the first write to "messages" would block until there is a read on that channel - which is never performed. But due to the buffering, the two writes to "messages" are performed without blocking, and the values can be read after that. A third read on the then empty channel would also block infinitely.

The flow of data through channels can be confusing until the concept becomes familiar, but the behavior is well-defined, so the behavior is as easy or difficult to predict as of any program of the given complexity.