A key limitation of traditional algebraic multigrid methods is the fundamental assumption regarding smooth error. Recently, there have been many investigations into weakening this assumption in search of greater robustness. In this talk, we discuss progress to date in the development of self-correcting (or adaptive) algebraic multigrid methods that automatically determine smooth components. We highlight the motivation for our approach and some of the key benefits, including improved robustness and a graceful degradation to simplicity.