Agentic Artificial Intelligence in Chemical Engineering, Process Systems Engineering, and Process Control: A Systematic Review of Emerging Perspectives and Challenges

This is the authors' abstract. We don't add key points for this paper.

Agentic artificial intelligence is gaining significance in chemical engineering. Many process decisions involve coordinated actions rather than isolated predictions. These decisions are constrained by physical limitations, uncertain measurements, safety protocols, and human supervision. This PRISMA-guided systematic review asked where AI agents are applied in chemical engineering and process control-related problems. It also asked which agent families are used, what evidence supports their contributions, and what limitations condition deployment. Scopus was searched in the title, abstract, and keyword fields on 20 April 2026. Eligible records were peer-reviewed articles published from 2022 to 2026, indexed in the Chemical Engineering subject area, and explicitly relevant to AI agents. Evidence maturity, reported limitations, and risk of overinterpretation were extracted for each study. The included studies were synthesized into four domains. These are safety and risk; digitalization and process systems engineering workflows; control, scheduling, and operations; and molecular, reaction, and materials design. Most evidence still comes from simulations, computational studies, or prototypes rather than from plants in operation. The review therefore identifies six conditions for deployment backed by auditable engineering evidence. These are industrial validation, safety guarantees, digital twins grounded in ontologies, reproducible evaluation of LLM agents, governance of the interaction between engineers and agents, and integration across scales.