Toward AI-Augmented Applied Systems: Emerging AI and LLM Directions Across Trajectory Prediction, Cybersecurity, and Engineering Automation

Abstract

This paper reviews fifteen recent publications (2015-2026) spanning trajectory prediction, cybersecurity and trust management, classical machine learning and algorithmic optimisation, and engineering automation, and asks a forward-looking question: where is artificial intelligence — particularly Large Language Models (LLMs), foundation models, and agentic AI systems — positioned to extend each of these research directions next? For each of four thematic clusters we first summarise the current method documented in the reviewed sources, then discuss a concrete, currently-emerging AI direction grounded in recent (2025-2026) literature: LLM- and foundation-model-based trajectory reasoning for Cluster A; agentic, LLM-driven security-operations copilots for Cluster B, including a reported 6.5x improvement in malicious-alert identification over human analysts in one production deployment; learned and graph-neural-network-based approaches to the classical optimisation and graph-isomorphism problems in Cluster C; and natural-language-to-workflow agentic automation for the low-code and embedded-systems work in Cluster D. Throughout, we explicitly separate what the reviewed papers themselves established from this review's own forward-looking analysis, and close with a discussion of the reliability, interpretability, and adversarial-robustness risks that accompany each proposed direction.