The term anaphor–agreement effect refers to the descriptive generalization that anaphors cannot control agreement on the verb. The general consensus in the literature is that the anaphor–agreement effect is a universal phenomenon. Contrary to that consensus, in this article, I present data from a set of languages where an anaphor does indeed control agreement on the verb, in violation of the anaphor–agreement effect. To explain this dual nature of anaphors in the world's languages, I propose a derivational-timing analysis that derives not only the anaphor–agreement effect but also those problematic cases in which it seems to be violated. © 2022 John Wiley & Sons Ltd