<p>This paper provides a novel analysis of asymmetries in gender agreement with conjoined subjects, accounting for a range phenomena in the Xhosa and Shona languages. For all Xhosa conjoined gender-mismatched singular DPs, the gender of plural agreement with &amp;P depends on whether the conjuncts share the feature [human], [inanimate], or [animal]. I argue that this reflects a stacked <i>n</i>P structure within Bantu nominals: whatever a DP’s visible gender, it includes an <i>n</i>P core with one of these three semantic genders. Thus the internal structure is [_nP1 n1 [_nP2 n2+root]]; n2 is always the bearer of a semantic gender. <i>n</i>2 values agreement if: (i) the conjoined DPs have mismatching visible n1 gender features but the same n2 cores, or (ii) the visible gender of the two conjuncts matches but is formally uninterpretable, assuming with Adamson and Anagnostopoulou (<CitationRef CitationID="CR1">2025</CitationRef>) that resolved agreement is based on the intersection of conjuncts’ interpretable features. A diagnostic for (un)interpretability of gender emerges: uniform gender singular conjuncts pair with semantic agreement iff their visible genders are uninterpretable, so excluded from the intersection. For others there is systematic matching agreement because they are uniformly interpretable, even in combination with arbitrary members since in such cases, <i>n</i>1 is <i>i</i>[entity]. I show that in Bantu languages with a large number of genders, those exhibiting matching plural agreement with uniform conjoined singulars are the exception, and not the rule.</p>