Algebraic multigrid (AMG) preconditioners are often employed in large- scale computer simulations to achieve scalability. AMG construction can be costly, sometimes as much as the solve itself. We discuss strategies for reducing expense through reuse of information from prior solves. The information type depends on the method. For smoothed aggregation AMG this includes aggregation data, whereas for energy minimization it includes sparsity patterns and prior interpolants. We demonstrate the effectiveness of such strategies in parallel applications.