In this paper we present a Causal Artificial Intelligence Design (CAID) theory that borrows notions from Classical philosophy for modeling intelligent agents. Principles introduced by this theory are used for extending a goal-driven BDI architecture and implementing what we call Causal Agent. This architecture incorporates causal formalisms like Pearl's Do calculus and C+ which are adapted to Semantic Web knowledge representations. Our approach includes an ontological agent description that enables and justifies the instantiation of agents as part of a plan. An experimental prototype used for validating experimentally our approach is commented.