Is this the best inference for agents?