Autonomic Computing, self-management based on high level guidance from humans, is increasingly being accepted as a means forward in designing reliable systems that both hide complexity from the user and control IT management costs. Effectively, AC may be viewed as policy-based self-management. We look at ways of achieving this, with particular focus on agent-oriented software engineering. We propose utilizing MaCMAS, an AOSE methodology for specifying autonomic and autonomous properties of the system independently. Later, by means of composition of these specifications, guided by a policy specification, we construct a specification for the policy and its subsequent deployment. We illustrate this by means of a case study based on a NASA concept mission and describe future work on a support toolkit.