Deterministic response is the core foundation for the safe operation of industrial production systems. However, with the increasing demand for intelligence, flexibility, and agility, ensuring the deterministic response of computing and control tasks while meeting new demands has become the primary issue that manufacturers urgently need to address. In response to this issue, this article focuses on AI-driven cyber–physical production systems (AI-CPPSs) and conducts research on the adaptive interaction mechanism of network, computing, and manufacturing resources with guaranteed performance. The efficient adaptive configuration of network, computing, and manufacturing resources is used to meet the response requirements of dynamic tasks. To achieve on-demand configuration of multidimensional resources for tasks, we first propose an AI-CPPS-oriented modeling method named the hourglass method, which redefines task models and multidimensional resources with resources as the core. Furthermore, through the proposed method of computing power quantification and a heterogeneous frame structure, we achieve the unified arrangement of network, computing, and manufacturing resources in the time dimension. Finally, to ensure the security and reliability of resource interaction, a multidimensional resource interaction mechanism is proposed for network computing control, namely, the quicksand mechanism. The experimental results indicate that the proposed quicksand mechanism can optimize resource utilization based on ensuring a deterministic task response.