AI-agent communication network for 6G: vision, architecture, and key technologies

Abstract

The booming of artificial intelligence (AI) agents has brought about promising business scenarios for sixth-generation (6G) mobile networks, while simultaneously posing significant challenges to network functionalities and infrastructure. These AI agents can be deployed on end devices (e.g., intelligent robots and intelligent cars) or as digital entities (e.g., personal AI assistants). As novel service entities with autonomous decision-making and task execution capabilities, AI agents introduce potential risks of uncontrollable actions and privacy disclosures. AI agents also require new 6G capabilities beyond traditional communication, including multimodality information interaction (e.g., AI models and tokens) and support for service requirements (e.g., computing and sensing of data). In this article, we introduce the concept of AI-agent communication network (ACN), a new paradigm to enable global information interaction and on-demand capability provisioning for single or multiple AI agents. We first introduce the vision and architectural framework of ACN. Then, key technologies and future research directions related to ACN are discussed. Furthermore, we provide potential use cases to elaborate on how ACN can expand the service capabilities of 6G networks.

1. Introduction

In recent years, artificial intelligence (AI)-based large models have developed rapidly, evolving from single-modal large language models (LLMs) such as ChatGPT to multimodal vision language models (VLMs) such as GPT-4o, as well as vision–language–action (VLA) large models. AI agents built on large models, which are autonomous systems capable of executing certain tasks by interpreting intents, sensing the environment, planning, making decisions, and using tools, constitute an emerging technology (Wang L et al., 2024). There are various AI agents with different capabilities, including embodied AI agents (such as intelligent service robots and robotic dogs) and digital AI agents (such as virtual intelligent assistants). Multi-agent systems are widely applied in various industries and fields. Currently, these AI agents of diverse types and capabilities have been widely applied across various industries, effectively enhancing work efficiency (Huang et al., 2025; Jiang XY et al., 2025).

The integration of AI with communication is one of the six most promising business scenarios for sixth-generation (6G) mobile networks, as presented in the International Telecommunication Union (ITU) 6G report (ITU, 2023). Numerous studies have been conducted on the application of AI and large models in telecommunication networks (Mahmoud et al., 2024; Shahid et al., 2025). Zhou et al. (2025) reviewed key techniques and opportunities in the context of LLMs for telecommunications, including LLM-enabled generative applications with telecommunication domain knowledge, LLM-based classification applications involving network security and traffic classification, LLM-enabled optimization techniques, and LLM-aided prediction for telecommunications. Chen ZR et al. (2024) proposed an architecture design and system evaluation for large AI models in 6G wireless networks. Similarly, the fundamental principles, diverse applications, key challenges, and future research directions of wireless large-scale AI models were elaborated in detail by Zhu et al. (2025).

Specifically, large-scale AI has shown tremendous potential in network management and optimization (Shahid et al., 2025). Some research works focus on integrating large-scale AI with network management. Xu YF et al. (2023) developed an LLM-aided network configuration benchmark called CloudEval-YAML, providing a realistic and scalable assessment framework for YAML configurations in cloud-native applications. Dzeparoska et al. (2023) proposed an advanced method to automatically translate high-level user intent into executable policies via LLM’s learning capabilities, eliminating pre-configuration. Besides, how to use large-scale AI for network optimization has been widely studied. Du et al. (2024) proposed an innovative LLM-enabled mixture of experts (MoE) approach for network optimization, which uses the powerful reasoning capabilities to analyze the objectives and constraints of users, select specialized deep reinforcement learning (DRL) experts, and determine their decision weights.