Yes, we’ve leaned toward minimizing operational overhead. Taking the useful parts of a mesh (xDS endpoint and routing updates) into the client has worked extremely well in practice and has been very reliable, without the extra moving parts of a full mesh.