The separation of intradomain and interdomain routing has been a key feature of the Internet's routing architecture from the early days of the ARPAnet. However, the appropriate "division of labor" between the two protocols becomes unclear when an Autonomous System (AS) has interdomain routes to a destination prefix through multiple border routers-a situation that is extremely common today because neighboring domains often connect in several locations. We believe that the current mechanism of early-exit or hot-potato routing-where each router in an AS directs traffic to the "closest" border router based on the intradomain path costs-is convoluted, restrictive, and sometimes quite disruptive. In this paper, we propose a flexible mechanism for routers to select the egress point for each destination prefix, allowing network administrators to satisfy diverse goals, such as traffic engineering and robustness to equipment failures. We present one example optimization problem that uses integer-programming techniques to tune our mechanism to improve network robustness. Experiments with topology and routing data from two backbone networks demonstrate that our solution is both simple (for the routers) and expressive (for the network administrators).