The results of his collaboration with Hardy (the approximations for the partition function and the circle method) make it clear he knew well what he was doing. Nobody was doing anything similar at the time.