The problem of Web service composition is the process of selecting, combining and executing existing services to establish reusable and versatile inter-operability applications. Currently, human beings perform manual Web service composition by reading information provided on service's Web pages. With the ever increasing number of Web services being made available on the Web, it is already beyond the human ability to analyze them and generate the composition process manually. This triggers an active area of research and development on automatic composition of Web services. Many researchers propose their composition approaches based on AI planning techniques. The major technical contributions of this paper are: 1) This paper first surveys the current problems and resolutions of semantic Web services and analyzes the relationship of the semantic Web, Web services and agents. Then an agent-based semantic Web service composition architecture is proposed. 2) The description logic is used, which is detailed as the formal tool of semantic Web and used to formalize the service composition problem, to represent the five different relationships of services. 3) An autonomic semantic Web service composition method is proposed based on planning algorithms by increasing limitation conditions and adopting agent strategies. By testing the approach to a simple, yet realistic example, the preliminary results demonstrate that the implementation provides a useful solution.