Nowadays, classical configurations of civil transportation airplanes are well optimized, with further improvements expected to be of the order of a single drag count. To achieve it, the entire airplane has to be taken into account during the aerodynamic shape optimization, which requires handling complex geometries with computer-aided-design software. So far, the integration of this software and aerodynamic shape optimization has been difficult, due to the framework complexity and problematic gradient evaluation. In this paper, the authors propose to substitute the computer-aided-design engine with a surrogate model of the surface deformation. The resulting framework does not require calls to the computer-aided-design software during the optimization. In addition, the surrogate model has been analytically differentiated and coupled with a mesh-adjoint method based on the radial-basis-functions to enable an efficient gradient-based optimization. The methodology is validated using two cases: a three-dimensional wing geometry in inviscid, transonic-flow conditions and a three-dimensional wing–body–tail configuration in a turbulent and transonic regime.