You probably wouldn't use that model with voice agents