Metareasoning is reasoning about reasoning, or deciding how to decide. Employing metareasoning allows one to defer design decisions (such as algorithm selection) to runtime, so that a robot or autonomous system can adapt its reasoning process to failures or changes in the environment or system state. In addition, by monitoring and controlling its reasoning process, the system can more efficiently use its limited computational resources. This talk will describe metareasoning and the process of developing a metareasoning policy. It will also discuss two metareasoning approaches that we have developed. In the first, metareasoning is implemented as part of a ground robot autonomy stack. The metareasoning approach gives the robot the ability to recover from path planning failures, which leads to better mission performance. In the second, a multi-agent team uses metareasoning to select collaboration algorithms in response to changes in communication availability, which reduces the time needed to perform collaborative search missions. Finally, this talk will discuss future plans for metareasoning research and engineering.