In this paper, we propose a plan execution architecture which supports different task semantics. This way, each goal that is identified during the deliberation cycle of an agent can be satisfied through tasks defined in different semantics. The capability of supporting different task semantics provides two main advantages. The first advantage is the reuse of legacy artifacts within agent plans. The second advantage is simplifying the adaptation of an agent architecture to different standards of a business organization. In order to integrate various task semantics within a plan execution architecture, we have used a smoothly revised version of a previously articulated workflow model into which different task semantics would be reduced before execution. We have integrated hierarchical task network and OWL-S semantics into our plan execution architecture to test the strength of it in terms of support for executing different task semantics in an agent architecture. We think that such a plan execution architecture will contribute to the industrialization of agent architectures.