From LLMs to intelligent agents: AI evolves for next-gen wireless systems

A new wave of artificial intelligence is poised to redefine the landscape of wireless communications. A tutorial study introduces the concept of agentic AI as the successor to large language models, forecasting its critical role in enabling intelligent, autonomous, and goal-driven communication systems in future 6G and beyond.

The research paper titled “From Large AI Models to Agentic AI: A Tutorial on Future Intelligent Communications,” published on arXiv, explores how communications engineering is transitioning from using traditional AI models toward developing multi-agent systems capable of autonomous collaboration, planning, and execution in real-time network environments. Authored by Feibo Jiang, Cunhua Pan, Li Dong, Kezhi Wang, Octavia A. Dobre, and Merouane Debbah, the study lays a conceptual and technical foundation for the next stage of AI-enabled communications.

The study envisions future communication networks powered by a new generation of AI agents that go beyond passive data processing. These agentic systems will actively perceive, reason, and act in a decentralized fashion, enabling networks to adapt to environmental uncertainties, complex user demands, and global optimization goals.

What limitations of large AI models call for agentic intelligence?

Large language models (LLMs) such as ChatGPT and GPT-4 have demonstrated immense potential across domains. In communication networks, they offer support in protocol design, network management, and signal processing. However, the study underscores critical limitations of LLMs when applied directly to wireless systems.

The first major constraint is that LLMs lack real-time adaptability and interaction. They operate in a static, turn-based framework and cannot autonomously execute plans or respond dynamically to changing network conditions. Moreover, LLMs require massive amounts of pretraining data and are often disconnected from the physical realities of wireless systems, including energy constraints, spectral efficiency, and user mobility.

In contrast, future wireless networks will demand AI systems capable of real-time decision-making, distributed cooperation, and physical world interaction. These requirements go beyond language generation and involve strategic autonomy, continual learning, and coordination across diverse network nodes. This is where agentic AI emerges as a transformative force.

How does agentic AI work and why is it suited for 6G and beyond?

Agentic AI refers to AI systems modeled as autonomous agents - software entities equipped with the ability to perceive environments, make decisions, and act toward defined objectives. The study presents a structural hierarchy of such agents for communication networks, ranging from simple rule-based actors to complex multi-agent systems with memory, planning, and inter-agent communication.

These agents integrate sensing, reasoning, and acting components to perform tasks such as:

• Dynamic spectrum access and allocation
• Energy-aware routing and scheduling
• Mobility prediction and handover optimization
• Edge caching and computing orchestration

Unlike LLMs, agentic systems do not require fixed prompts or centralized inference. Instead, they operate in an event-driven, adaptive paradigm, similar to autonomous robots or reinforcement learning agents, enabling real-time responsiveness in communication environments.

The tutorial highlights how multi-agent reinforcement learning (MARL) and graph neural networks (GNNs) can power distributed optimization, allowing multiple agents to collaborate without centralized control. This architecture is especially suited for 6G scenarios involving UAV swarms, massive IoT networks, and intelligent vehicular systems.

The study also introduces the concept of communication-aware agent design, where the physical layer constraints of the network (latency, interference, bandwidth) are embedded into agent training and deployment. This grounding in real-world constraints ensures that agentic AI remains efficient, robust, and scalable under wireless conditions.

What are the key research directions and challenges for agentic AI in communications?

While agentic AI offers transformative potential, the study acknowledges significant technical and systemic challenges that must be addressed to realize its full integration into future networks.

• Modeling and Simulation: Developing realistic, large-scale environments for training and testing agentic systems is still in early stages. Simulators that include accurate physical and protocol models are required.

• Coordination and Scalability: As networks scale, agent interactions can become chaotic or unstable. Mechanisms for learning cooperation strategies, managing agent interference, and ensuring convergence remain open research areas.

• Security and Trust: Autonomous agents increase the risk of adversarial manipulation. Future systems must include mechanisms for secure coordination, anomaly detection, and fail-safe fallback.

• Interfacing with LLMs: The authors propose that agentic AI should not replace LLMs but complement them. LLMs can serve as cognition modules or knowledge repositories for agents, enabling natural language interaction and semantic understanding.

• Standardization and Benchmarking: The field lacks unified benchmarks or protocols for evaluating agent performance in communications. The paper calls for open datasets, collaborative competitions, and international standardization efforts.

The researchers emphasize the need for cross-disciplinary collaboration among AI scientists, communication engineers, control theorists, and policymakers to develop ethical, reliable, and technically sound frameworks for deploying agentic AI at scale.