This position paper is concerned with the challenge of engineering multi-scale and long-lasting systems, whose operation is regulated by sets of mutually-agreed, conventional rules. The core of the problem is that there are multiple, inter-dependent dimensions of flux, with numerous contextual factors to take into account. These dimensions of flux include, on the one hand, the set of rules itself; and on the other, the system components (population), their social network, and the operating environment. However, there appears to be no ‘one size fits all’ optimum ruleset for all combinations of population, social network and environment; nor (given the contextual factors) is there a planning-type algorithm that can compute an ‘ideal’ ruleset for any particular combination of population, social network and environment. These features of the problem suggest that recent advances in machine learning and evolutionary computation can provide the instruments for facilitating self-adaptation of a rule-based system over different timescales. This paper proposes that the integration of concepts from socially- and biologically-inspired computing can pave the way for eventual development of a computational framework that will enable principled (methodological) development of sustainable adaptive rule-based systems.