This paper reports our experiences with extending Interface Automaton (IA) with shared variables by lifting IA’s intuitive notion of refinement and compositionality to shared variables. Although there are existing works that introduce shared variables to IA, they typically support a very restricted notion of sharing, e.g., the value of a shared variable is only defined for the duration of an atomic operation, with no ability to persist values for subsequent operations. When attempting to formulate the semantics of shared variables that could persist their values across operations, we encountered numerous challenges when defining a notion of refinement that respected compositionality. We conjecture that, even for a basic notion of variable persistence, concurrent shared variable accesses between automata create a tight data dependency that prevents a compositional reasoning. We discuss the generality of this negative result in relation to other concurrency theories.