A. Does x exist?
B. Do I know that x exists?
We can complete these questions by exchanging the variable x for a meaningful word, e.g. God, Russell's teapot, a highest prime number, or whatever. A-type questions are ontological, while the B-type questions are epistemological. That is, A-type questions concern what exists, and the B-type questions concern what we know to exist. This is a subtle but important distinction. Atheism and theism address an A-type question, while agnosticism addresses B-type questions.
While it's fairly uncontroversial what it would mean for something to exist, almost everyone has different standards regarding what it means to know that something exists. Most self-described agnostics harbour a rather naive empiricist epistemology: seeing is believing. They know that something exists when there is evidence verifying, confirming, or otherwise supporting the claim that it exists; the 'naive' part of their epistemology is that they rarely recognise or appreciate the theory-ladenness of such evidence. Further, most agnostics qualify their knowing in probabilistic terms, so rather than saying 'I know that x exists', they say 'I know that x probably exists'.
The problem, then, is that agnostics have a tendency to muddy the waters concerning the ontological question of whether something exists, because they tacitly substitute the epistemological question of whether they know that thing exists. This, I think, is largely a consequence of an implicit theory of rationality that demands that they should not take a position on an A-type question unless they can take a position on the corresponding B-type question. In practice, this can lead to some rather Orwellian conflicts between words and action; they refuse to answer A-type questions about whether a particular god exists, but still behave just as though that god doesn't exist.
Agnostics are mostly non-confrontational. They're usually atheists, at least judging by demonstrated behaviour, but want to avoid connotations of intolerance and certainty that go with terms like knowing. To resolve the Orwellian tangle they've gotten themselves into, and to try and diffuse arguments on a controversial issue, they usually end up trying to undermine the notion of truth itself. Sometimes truth is redefined into more pragmatic terms (truth is what is useful), and other times truth is thrown out all together (usually by confusing the existence of truth itself for knowledge of the truth).
The attitude seems to be that if truth itself can be undermined, then everyone can just stop arguing and get along. In my opinion, this is a rather counter-productive strategy. A common feature of most violent and totalitarian ideologies is that rational argument is futile. When people cannot address their differences through rational argument and a common goal of discovering the truth, then they tend to use swords instead.