Model checking is becoming a popular verification method that still suffers from combinatorial explosion when used on large industrial systems. Currently, experts can, in some cases, overcome this complexity by selecting appropriate modeling and verification techniques, as well as an adapted representation of the system. Unfortunately, this cannot yet be done automatically, thus hindering the use of model checking in industry. The objective of this paper is to sketch a way to tackle this problem by introducing self-adaptive model checking. This is a long term goal that could lead the community to elaborate a new generation of model checkers able to successfully push forwards the scale of the systems they can deal with.