Many modern engineering applications involve materials with heterogeneous and often highly anisotropic material properties with applications ranging from thermal diffusion over groundwater flow or electrostatics to solid mechanics. Numerical simulations of such systems are challenging, since finely resolved finite element discretizations usually result in poorly scaled and ill-conditioned linear systems of equations. To ensure efficiency and scalability on high-performance computing platforms, iterative solvers preconditioned with algebraic multigrid (AMG) methods are widely employed. However, conventional AMG techniques typically do not account for material heterogeneity or anisotropy, resulting in low convergence or even failure of the iterative solver. Although recent progress has been made for mesh-induced anisotropy, the robust treatment of genuinely heterogeneous and anisotropic materials remains an open problem. In this talk, we present a new smoothed-aggregation strategy that incorporates material information directly into the AMG coarsening process. In contrast to classical strength-of-connection measures, which do not account for coefficient jumps and directional anisotropy, the proposed approach is robust, easy to apply, and effective across a wide range of scenarios. For problems with strongly directional material behavior, we demonstrate that the iteration counts become independent of the material contrast. We also discuss the implications of Jacobi-based prolongation smoothing for ill-conditioned fine-level operators and introduce a matrix-filtering procedure based on a targeted dropping scheme. This filtering controls operator complexity throughout the multigrid hierarchy, ensuring that the computational cost of the preconditioner remains low. We show examples for scalar problems as well as ongoing work of extensions to vector-valued problems such as elasticity.