Software is increasingly produced according to a certain goal and by integrating existing software produced by third-parties, typically black-box, and often provided without a machine readable documentation. This implies that development processes of the next future have to explicitly deal with an inherent incompleteness of information about existing software, notably on its behaviour. Therefore, on one side a software producer will less and less know the precise behaviour of a third party software service, on the other side she will need to use it to build her own application. In this paper we present an innovative development process to automatically produce dependable software systems by integrating existing services under uncertainty and according to the specied goal. Moreover, we (i) discuss important challenges that must be faced while producing the kind of systems we are targeting, (ii) give an overview of the state of art related to the identied challenges, and finally (iii) provide research directions to address these challenges.