The concepts of exchange coupling in magnetic materials, and the most common approximations of this theory which are used in realistic studies are majors topics of this talk. The validity of certain popular approximations in magnetic materials with a different degree of magnetic itineracy will be shown. In particular, the focus on removal of long-wave length approximation and static/rigid spin approaches. The limitations of bilinear considerations, as in the Heisenberg model, are shown in detail. Realistic considerations are done using a linear response scheme. The majority of results are obtained for elemental 3d ferromagnets (Fe, Ni) and also for the recently discovered antiferromagnetic iron pnictides/selenides. Our results indicate that all approximations mentioned above are highly material-specific. We find that the conventional static Heisenberg model is best justified for systems with large local moments and relatively weak magnons in a long-wavelength limit. This results in a very limited applicability of this model for quantitative descriptions of real magnets. We also introduce a concept for dynamic exchange coupling and show that in Fe, for example, the first nearest-neighbor exchange parameter is rather static, while already the second neighbor already has a significant energy dependence in the spin wave energy range. The long wave approximation also strongly alters the distribution of coupling strength among the nearest few neighbors. We also discover that famous exchange anisotropy of the exchange coupling in iron pnictides has strong dynamic component, which allows us to provide new interpretations of the spin wave spectrum of these materials and introduce a new type of magnetic fluctuation. A detailed comparison with the experiments will be shown.