In 'Wittgenstein on Following a Rule' McDowell's objection to the idea that language use just involves contingent agreement among speakers in their dispositions to go on in the same way, rather than some linguistic community in a richer McDowellian sense seems to be this. If the former view is right, we can never have more than "inductive" certainty that the rest of our community uses the word the same way. Hence, when we apply a certain term in a certain way, e.g. when we say "arthritis is inflamation of the joints" we can only be `inductively' certain that this expresses a truth - it's logically possible that everyone in our language community uses the word differently.
But why is this a problem? This supposedly bad consequence seems directly *true* in the arthritis case. Maybe it's worse to say that you can only be inductively certain that 2+2=4, since it's logically possible that your whole language community uses the word differently. But - come to think of it- don't we individuate language communities by common linguistic practice. So, arguably, if any community were to count as your linguistic community it would have to agree with you about many (most?) assertions that are really central to you, which you feel confident about. So the worry about the rest of our community using "2+2=4" differently enough for it to express a falsehood seems very very slender.
p.s. does anyone know if McD thinks he has a transcendental argument for the existence of other people, from the claim that we can have meaningful thoughts, and hence must belong to some non-private-language community?