Symbolic execution of programs is a versatile software analysis technique that recently enjoyed much renewed interest. Traditional usage of symbolic execution in a logic-based setting includes formal verification and model checking, however, symbolic execution engines for Java or bytecode provide also a basis of glass-box test case generation, visualisation of program dynamics, or resource analysis. In this talk we present yet another take on symbolic execution: we show that symbolic execution, when viewed as an advanced forward analysis performed on the abstract syntax tree of a program, can be combined with first-order reasoning, partial evaluation, and usage analysis to implement provably correct program specialization and compilation.