The analysis presented in this paper provides a development of the idea that subject agreement markers in Bantu can be analysed as pronouns. With reference to Swahili and Herero, it is shown that the interpretation of subject markers is dependent on the context in which they are found and that it is, similar to ordinary pronouns, constrained by locality, directionality and class features. The relevant context includes both the wider, pragmatic context as well as the relation between overt subject and agreeing subject marker, and thus includes word-order variation between subject-verb and verb-subject structures. Lexical restrictions on the interpretation of subject markers interact with different ways in which logical subjects can be syntac...