Many geometric structures (such as Riemannian, conformal, CR,
projective, systems of ODE, and various types of generic
distributions) admit an equivalent description as Cartan geometries.
For Cartan geometries of a given type, the maximal amount of symmetry is realized by the flat model. However, if the geometry is not (locally) flat, how much symmetry can it have? Understanding this "gap" between maximal and submaximal symmetry in the case of parabolic geometries is the subject of this talk. I'll describe how a combination of Tanaka theory, Kostant's version of the Bott-Borel-Weil theorem, and a new Dynkin diagram recipe led to a complete classification of the submaximal symmetry dimensions in all parabolic geometries of type (G,P), where G is a complex or split-real simple Lie group and P is a parabolic subgroup. (Joint work with Boris Kruglikov.)