In long-run planning, targets may be set further in the future than economic technologies can be estimated. How costly in hindsight is limited foresight? Following Los, we say k is an @e-horizon if knowledge of economic possibilities for the next K time periods permits decisions with results at most @e worse than optimal. For a dynamic Leontief model, we show that any T-period efficient path is almost as big as all feasible T-period paths during the first T - K periods. Thus, given K periods notice, we can plan an @e-optimal growth path no matter how technologies change after T.