Hindi-Urdu displays an asymmetry with respect to the availability of Closest Conjunct Agreement. It is available only to objects and not to subjects. Agreement with subjects is always agreement with the full conjunct. We argue that this asymmetry in Conjunct Agreement is related to another asymmetry between subject and object agreement in Indo-Aryan languages: object agreement never involves person. We derive these properties of object agreement from the fact that object agreement is an instance of dissociated agreement, agreement that takes place independent of case-licensing. As a result when the probe (T) accesses the direct object goal, the person features of the goal have already been deactivated by the case-licenser (v) and T must look inside the DP at the phi-P, where only gender and number features are available. This yields the absence of person features in object agreement. With subjects, T is both the case-licensor and phi-agreement trigger. Hence the person features of the subject are visible to T. By a similar logic, the features of conjoined objects are not visible to the probe and a subpart must be identified whose features are visible. The identification of the subpart is subject to linearity considerations and we present a mechanism that allows for this. The resulting proposal sheds light on the distribution of features within the DP and the proper analysis of dissociated agreement. It is also a first step towards an integration of linearization and structural considerations in the treatment of agreement.