In the long run, we are all dead. Nonetheless, even when investigating short-run dynamics, models require boundary conditions on long-run, forward-looking behavior. In this paper, we show how deep learning approximations can automatically fulfill these conditions.