No subject

    (m1 >> m2) >> m3 === m1 >> (m2 >> m3)

Now, consider tell. We'd expect it to obey a law like this:

    tell w1 >> tell w2 === tell (w1 <> w2)

Combine this with the monad associativity law:

    (tell w1 >> tell w2) >> tell w3 === tell w1 >> (tell w2 >> tell w3)

And it's easy to see that '<>' must be associative in order for this law
to be upheld.  Additionally, the existence of identities in monads means
that there must be a corresponding identity for the monoid.

So anything that is "writer-like" and also satisfies the monad laws...
is going to be a monoid.

Now, it's possible what GP is actually asking about is more a question of
encapsulation.  Well, one answer is, "Well, just give the user specialized
functions which do the appropriate wrapping/unwrapping"; another answer is,
"if you let the user run a writer action and extract the resulting written
value, then he can always reverse engineer the monoid instance out of it".

Edward