Between the BICEP2 and Keck Array experiments, we have deployed over 1500 dual polarized antenna coupled bolometers to map the Cosmic Microwave Background's polarization. We have been able to rapidly deploy these detectors because they are completely planar with an integrated phased-array antenna. Through our experience in these experiments, we have learned of several challenges with this technology- specifically the beam synthesis in the antenna- and in this paper we report on how we have modified our designs to mitigate these challenges. In particular, we discus differential steering errors between the polarization pairs' beam centroids due to microstrip cross talk and gradients of penetration depth in the niobium thin films of our millimeter wave circuits. We also discuss how we have suppressed side lobe response with a Gaussian taper of our antenna illumination pattern. These improvements will be used in Spider, Polar-1, and this season's retrofit of Keck Array.