Position: Cross-Layer Co-Design Is Necessary to Safely Deploy Agentic AI in Real-Time Systems

This position paper argues that the rapid advancement of Generative AI (GenAI), driving the emergence of reasoning capabilities in Large Language Models (LLMs), is creating a paradigm shift in autonomous cyber-physical systems where AI-based agentic models are integrated into real-time decision making. Concurrently, the demand for real-time (RT) computing and connectivity in industrial, automotive, and robotic applications remains critical. We argue that the inevitable convergence of these two domains exposes a fundamental mismatch between probabilistic agentic AI workloads and deterministic real-time requirements.We delineate two primary modes of convergence: the isolation of resource-intensive agentic workloads from RT tasks that need deterministic execution on shared platforms, and the direct integration of agentic AI into time-sensitive decision making within cyber-physical systems.For each mode, we identify key applications, outline fundamental challenges such as resource contention and temporal unpredictability, and survey state-of-the-art solutions spanning compute platforms, communication stacks, and application-level strategies.We conclude that future progress requires a cross-layer co-design approach, encompassing new system-level timing models, real-time agentic frameworks, and novel certification methodologies, to bridge the gap between the probabilistic nature of GenAI and the deterministic needs of real-time workloads.