Our concern in this paper is with the question of how irrational an intentional agent can be, and, in particular, with an argument Stephen Stich has given for the claim that there are only very minimal a priori requirements on the rationality of intentional agents. The argument appears in chapter 2 of The Fragmentation of Reason.1 Stich is concerned there with the prospects for the ‘reform-minded epistemologist’. If there are a priori limits on how irrational we can be, there are (...) limits to how much reform we could expect to achieve. With this in mind, Stich sets out to determine what a priori limits there are on irrationality by examining `a cluster of influential arguments aimed at showing that there are conceptual constraints on how badly a person can reason’ (p. 30). Stich aims to remove the threat of a priori limits on the project of reforming our cognitive practices by showing, first, that these influential arguments are bad arguments, and, second, that at best there are only minimal constraints on how irrational we can be.2 We aim to show three things. The first is that Stich’s own arguments against strong a priori limits on how badly a person can reason are unsuccessful, because Stich fails to take into account that the concept of rationality is an epistemic, not just a logical concept, and because he fails to take into account the connection between having a concept and being able to recognize conceptually simple inferences involving the concept. The second is that the position Stich argues for, on the basis of Richard Grandy’s principle of humanity, turns out not to be distinct from the one he rejects. The third is that, in any case, the position that Stich rejects in order to preserve some scope for the project of improving our reasoning is not only no danger to that project but must be presupposed by it. (shrink)