One distinctive feature of human intelligence is a high level of flexibility. Human thought is flexible roughly in the sense that , "there is no end to the kinds of problems human reason can deal with" (Horgan & Tienson, 1996). However, no theory to date has adequately explained such unique capacity. Recently, Evolutionary Psychologists have confronted this challenge by building models that have the potential to generate human flexibility via interaction of modules and learning (Barrett, 2005; Carruthers, 2006a; Sperber & Hirschfeld, 2006; Sperber, 1994). The key idea is that our cognitive system can learn to self-assemble, out of our sophisticated adaptive toolbox, new mechanisms that solve novel problems. In this paper, I identify a serious information routing problem, “the nativist input problem”, distinct from the a priori and really real input problems previously launched against evolutionary psychology by Fodor(2000) and subsequently solved (Barrett, 2005; Pinker, 2005). The nativist input problem is, briefly, a crippling limitation to the range of contexts in which a massively modular architecture can handle information routing reliably. I argue that it undermines successful self-assembly required for these models to explain human flexibility, highlighting nativism as one of the most problematic commitments of evolutionary psychology.