Algebraic multigrid methods are known to be efficient and scalable preconditioners, not only for single-field problems but also for applications in multi-physics scenarios. The latter require special treatment for each individual field as well as an appropriate handling of the overall coupled block system. To archieve the construction of a general algebraic monolithic multigrid block preconditioner, the concept of segregated transfer operators is used to project information from fine to coarse levels including the coupling constraints on the off-diagonal entries of the block operator. Depending on the algebraic structure of the system, either standard block relaxation methods or saddle point smoothers are applied. This talk will discuss the theory behind physics-based algebraic multigrid block preconditioning and introduces a flexible software framework for the application to challenging simulation scenarios implemented in Trilinos/MueLu. These range from mixed-dimensional modelling (like the interaction of fibers with solids) to interface problems (e.g contact mechanics or fluid-solid interaction). Using algebraic multigrid methods in this context is challenging as there are several open questions regarding its core components like aggregation, prolongation and smoothing properties. In addition, the performance and scalability of the methods will be assessed.