A crucial step in the aggregation of amyloidogenic sequences is the transition from a relatively disordered state to an assembly-competent structure (referred to as the N* state). Using a well-calibrated model for intrinsically disordered proteins (IDPs), we show that these N* states bear considerable resemblance to the fibril polymorphs (U-bend and S-bend) characterized in experiments. The N* states appear as excitations on the monomer free energy landscape and are only sparsely populated. By relating the populations of the N* states to fibril formation time-scales using an empirical relationship derived from lattice simulations, we can explain nearly quantitatively the relative aggregation propensities of Ab40 and Ab42, the two most common isoforms implicated in Alzheimer’s disease. Despite having similar global topographies, the energy landscapes of Ab40 and Ab42 exhibit different extents of local roughness, which we show have profound implications in dictating selfassembly. Using kinetic transition networks, we further show that the N* states are obligatory along the favored assembly routes. The N* theory seems quite general, and also explains the length-dependent polymorphism in the fused in sarcoma (FUS) peptide sequence.