In a recent paper, Zeijlstra (2012) argues that the theory of agreement should be revised so that the direction of valuation would always be downward: the element that contributes the value (e.g. in the case of predicate-argument agreement in φ-features, the nominal) would be required to c-command the element that receives its value derivatively (e.g. the verb or tense/aspect/mood marker) – rather than the other way around, as standardly assumed. In this short reply, I wish to demonstrate that Zeijlstra's proposal is unsuitable as a theory of φ-agreement (i.e., of morpho-phonologically overt co-variance in φ-features between a finite verb or a tense/aspect/mood marker and a nominal argument). I survey two empirical domains, from Tsez and Basque, demonstrating this point; I then briefly discuss the consequences of these facts for the empirical domains that Zeijlstra examines.