Collaborative AI’s offer great potential for explorative and creative methods. While previous work in HCI and ML mainly focuses on exploiting either human or machine capabilities, the concept of collaboration suggests work on equal terms to achieve synergy effects as seen in collaborative creativity. However, the uncertain nature of creative problems raises new questions regarding the definition and design of such systems. We present a collaborative system for mood board design based on a state-of-the-art contextual bandit structure that is able to iteratively adapt to changing objectives, and autonomously selects explorative/ exploitative strategies to propose suitable contributions to the current mood board. Besides the technical implementation, we discuss the need of design professionals in the design of such collaborative systems as well as open questions based on our results.