Cyber–Physical Systems (CPSs) are complex and heterogeneous systems. Interacting with the physical world makes CPS unpredictable because of unanticipated events. Therefore, a CPS needs to reason these events autonomously. Henceforward, suitable reasoning mechanisms, such as rational agents with deliberative capabilities, should be selected and integrated into the CPSs. In this way, the integrated multi-agents can reason on environmental changes to find a plan that sustains the system’s operation for CPS. In addition, a layered architecture can pave the way to integrate the rational agents on embedded hardware and control the CPS. To this end, this paper presents an architecture, discusses a reference implementation and elaborates on the high-level integration of agents and CPS. Moreover, a complex and heterogeneous case study is provided to validate the effectiveness of rational agents in conducted experiments. Firstly, the rational agents utilising the belief–desire–intention (BDI) model required approximately three times less development time than simple-reflex agents. Secondly, the proposed approach resulted in up to 3 times less description complexity in language expressiveness. Lastly, the product quality is improved up to 66% by the rational agents and BDI model. As a result, our approach is beneficial to designing multi-agent CPS where it is aimed to use low-level control and high-level reasoning in a single platform.