Service-oriented computing is a new paradigm to improve the performance of information systems in environments demanding for flexibility and interconnectivity. Additionally, the support of intra- and inter-organizational processes has become crucial for the competitiveness of companies. Recent research approaches propagate to plan processes automatically to improve quality and flexibility. This paper discusses the application of AI planning for service composition in a service-oriented architecture. Three alternative levels of abstraction for service descriptions are defined and their impact on automated planning is investigated in regard to the provider and requestor role of service composers. It is shown that different abstraction levels of service descriptions are necessary to take full advantage of AI planning in a service-oriented architecture.