This is more a feedback than a question unless the author would like to clarify further. In offline metrics, AUC is described as a loss metric which makes no sense. You can train with log loss and then calibrate threshold for classification using AUC. One does not substitute another as the comparison is not relevant.
