Soft robotic manipulators offer huge potential for adaptive interaction with complex environments. Yet, unlocking their full capabilities requires rethinking design not as a linear process, but as an integrated, co-evolving relationship between morphology, control, and task requirements. This technical workshop brings together researchers exploring the frontiers of co-design—where body and brain are developed in synergy to achieve efficient and effective manipulation. We will spotlight emerging focus areas that promise to redefine the design process of soft robots. First, computational efficiency: making co-design algorithms more practical and feasible via more effective design parametrization, the use of design priors, and novel optimization strategies. Second, foundation models: exploring their use not just as data-driven generators of robot morphology, but also as design critics and generalist motion controllers—recasting co-design in the era of AI. The program combines talks from roboticists, AI researchers, and designers, with roundtable discussions and a co-design sprint. Together, we will envision soft robots whose form and function are co-developed to respond not just to task requirements, but to the physical and social complexity of human environments. By focusing on this deep coupling between human-centered design and task/context, the workshop aligns with the broader ambition of RoboSoft26 to address meaningful societal challenges.