Agentic AI Reasoning for Mobile Edge General Intelligence: Fundamentals, Approaches, and Directions

The rapid advancement of large language models (LLMs) has enabled an emergence of agentic artificial intelligence (AI) with powerful reasoning and autonomous decision-making capabilities. This integration with edge computing has led to the development of Mobile Edge General Intelligence (MEGI), which brings real-time, privacy-preserving reasoning to the network edge. However, deploying LLM-based agentic AI reasoning in MEGI environments poses significant challenges due to the high computational demands of reasoning and the limited resources of edge devices. To address these challenges, we propose a joint optimization framework for efficient LLM reasoning deployment in MEGI. First, we systematically review enhancement methods to identify mechanisms suitable for edge adaptation. Subsequently, we present a distributed framework that synergizes reasoning enhancement via adaptive CoT prompting with scalable deployment through a distributed MoE architecture. An important innovation of this approach involves modeling reasoning depth as a dynamic network resource variable, which is optimized jointly with expert activation and transmission power. This mechanism allows the system to dynamically regulate expert networks and reasoning complexity according to task requirements and device capabilities. Experimental evaluations in mobile edge environments demonstrate that the proposed framework effectively balances reasoning quality and resource efficiency. The results show that with less than one second of additional inference time, both accuracy and latency satisfaction rate can reach 90%, validating the practical viability of deploying sophisticated LLM reasoning in resource-constrained MEGI systems.

💡 Research Summary

The paper addresses the emerging need to run large‑language‑model (LLM) based agentic AI—systems that can plan, reason, and act autonomously—directly at the mobile edge, a paradigm the authors term Mobile Edge General Intelligence (MEGI). While cloud‑based services such as ChatGPT provide ample compute, they cannot satisfy the stringent latency, privacy, and data‑sovereignty requirements of many real‑time edge applications (e.g., fault detection, autonomous driving assistance, on‑device personal assistants). Deploying LLMs at the edge is challenging because (1) reasoning tasks often require multi‑step, high‑precision computation, (2) edge devices have limited CPU/GPU, memory, and battery capacity, and (3) wireless links are variable, adding unpredictable communication overhead.

The authors first survey existing techniques for enhancing LLM reasoning across three development phases: pre‑training (model scaling, new architectures such as Mixture‑of‑Experts (MoE) and multi‑agent systems), fine‑tuning (Supervised Fine‑Tuning, Reinforcement Learning from Human Feedback), and inference (Chain‑of‑Thought (CoT) prompting, self‑consistency). They note that most prior work treats inference as a static workload and does not account for the fact that reasoning quality is tightly coupled to the “depth” of CoT steps, which directly influences both compute and communication costs.

To bridge this gap, the paper proposes a joint optimization framework that simultaneously (a) adapts the reasoning depth of CoT prompting, (b) selects and activates a subset of expert networks in a distributed MoE architecture, and (c) controls transmission power for token exchange. The key conceptual innovation is to model reasoning depth (d) as a dynamic network‑resource variable. The total system energy is expressed as

\