The abundance of software that will be available in the next future will promote the production of appropriate integration means (architectures, connectors, integration patterns). The produced software will need to be able to evolve, react and adapt quickly to a continuously changing environment, while guaranteeing dependability through (on-the-fly) validation. The strongest adversary to this view is the lack of knowledge on the software, notably on its structure, behaviour, and execution context. Despite the possibility to extract observational models from existing software, a producer will always operate with software artefacts that exhibit a degree of uncertainty in terms of their functional and non functional characteristics. Uncertainty can only be controlled by making it explicit and by using it to drive the production process itself. This calls for a production process that explores available software and assesses its degree of uncertainty in relation to the opportunistic goal G, assists the producer in creating the appropriate integration means towards G, and validates the quality of the integrated system with respect to the goal G and the current context. In this paper we discuss how goal-oriented software systems can be opportunistically created by integrating under uncertainty existing pieces of software.