The challenges in orchestrating complex industrial processes — whether in biomanufacturing, supply chains, or sustainable fuel production — stem from their inherently nonlinear dynamics, the opacity of real-time measurements in harsh environments, and the escalating demands for resilience amid global disruptions and decarbonization mandates. Traditional centralized control architectures, reliant on exhaustive sensor arrays and rigid optimization routines, falter under these uncertainties, leading to suboptimal yields, safety vulnerabilities, and inefficient resource use. As articulated in recent perspectives on agentic Artificial Intelligence (AI) in process systems engineering, these limitations underscore the urgent need for paradigms that harness distributed, adaptive decision-making to bridge sensing gaps, infer hidden patterns, and enact proactive decisions. In this perspective, we have revitalized the foundational Adaptive Agent-Oriented Software Architecture, the very blueprint that powered early distributed AI innovations like Apple's Siri — and evolved it into the Adaptive Agent-Oriented System Control (AAOSC) framework a decentralized control layer built around specialized autonomous agents coordinated through digital twins and real-time communication protocols. Rather than proposing new algorithmic components, AAOSC provides a unifying architectural approach for integrating hybrid physics-informed models, learning-based inference, and distributed control in cyber-physical process systems. The four case studies presented herein — spanning supply chain security, closed-loop biomanufacturing control, quantum-enhanced anomaly detection, and prospective optimization of digital twin for biomass-derived sustainable aviation and maritime fuels — collectively demonstrate AAOSC's prowess: reducing deviation durations, averting shutdowns in severe fault scenarios, and boosting efficiency through virtual sensing and decentralized reasoning, all while aligning with regulatory imperatives for safety and sustainability. Looking ahead, AAOSC charts a pathway toward cyber-physical ecosystems designed to support future regulatory-aligned deployment in safety-critical industries that not only minimize human fatigue-related risks — a leading cause of incidents in manufacturing — but also democratize advanced control for decentralized operations worldwide. Future implementations must prioritize hybrid edge-cloud deployments for low-latency actuation, robust cryptographic protocols for agent authentication in multi-stakeholder networks, and qualification and governance frameworks that would be required for compliance with U.S. Food and Drug Administration/European Medicines Agency (FDA/EMA) standards in regulated industries for biopharma and food production. Extensions could incorporate multi-logic decision ensembles within agents, quantum-accelerated reasoning for ultra-fast inference, and seamless integration with emerging generative models for predictive scenario generation. By thus extending agentic principles from consumer interfaces to mission-critical sustainability challenges, AAOSC promises to catalyze a new era of resilient, equitable industrial intelligence, where computational agents tirelessly safeguard and optimize the physical world.
Mansouri et al. (Tue,) studied this question.