In the near future we will be surrounded by a virtually infinite number of software applications that provide services in the digital space. This situation radically changes the way software will be produced and used: (i) software is increasingly produced according to specific goals and by integrating existing software; (ii) the focus of software production will be shifted towards reuse of third-parties software, typically black-box, that is often provided without a machine readable documentation. The evidence underlying this scenario is that the price to pay for this software availability is a lack of knowledge on the software itself, notably on its interaction behaviour. A producer will operate with software artefacts that are not completely known in terms of their functional and non-functional characteristics. The general problem is therefore directed to the ability of interacting with the artefacts to the extent the goal is reached. This is not a trivial problem given the virtually infinite interaction protocols that can be defined at application level. Different software artefacts with heterogeneous interaction protocols may need to interoperate in order to reach the goal. In this paper we focus on techniques and tools for integration code synthesis, which are able to deal with partial knowledge and automatically produce correct-by-construction service-oriented systems with respect to functional goals. The research approach we propose builds around two phases: elicit and integrate. The first concerns observation theories and techniques to elicit functional behavioural models of the interaction protocol of black-box services. The second deals with compositional theories and techniques to automatically synthesize appropriate integration means to compose the services together in order to realize a service choreography that satisfies the goal.