One model writes your code; others test and review it. You manage the work from your terminal or phone. The job agent adds independent testing and review before delivery.