AI Agents & Agentic AI in Telecom – the Next Frontier for Global Operators

Introduction 

The global telecom industry today is facing dual challenges of surviving and thriving. From both the operators’ and users’ points of view, while the 4G story has been a glorious one, the 5G story is not something to write home about. The 5G hype train has quietly left the station without delivering the promised high revenue services built on top of capabilities such as ultra-low latency and extremely high bandwidth. To compound matters further, the demand for mobile data is unrelenting and rising exponentially, especially in the context of new services enabled by AI.

As a result, the telecom industry itself is transforming from being the playfield of just the traditional players to a more vibrant democracy in which players such as Meta, Google, Amazon, Microsoft and others have a seat on the telecom standardization high table and are influencing network architectures to enhance operational efficiency, network resiliency and better user experience as AR, VR and AI-enabled services proliferate. This new reality is putting huge existential pressure on the telecom network operators and forcing a rethink on network design, management and optimization for better spectral efficiency (bits/Hz), richer user experience (QoS), lower operational expenditure (opex), and higher revenue per user (ARPU). To these ends, the recent advances in agentic AI can play a significant role.

For decades, telecom operators have relied on human RF engineers to manually carry out manual drive tests, parameter tuning, and building rule-based automations to keep networks running smoothly and efficiently. However, the 5G and future networks are too complex, dynamic, and data-heavy for traditional methods of operating the networks. In order to implement capabilities like zero-touch provisioning, self-healing, autonomous management and higher energy efficiency, there is an urgent need for building agentic AI frameworks and over-the-top applications for the existing and the evolving networks.

This article brings out the differences between AI agents and agentic AI (agentic AI applications) in terms of their designs and applications in the modern telecom networks, and goes on to establish the complementary utility of both for the global network operators.

AI Agents versus Agentic AI

AI agents and agentic AI are fundamentally different in their scope, intelligence and impact. An AI agent is essentially a software bot (automation) programmed to handle a specific, narrow task. Think of it as a junior RF engineer in a NOC (Network Operations Center), who efficiently executes a specific task by following the steps written in the ‘rulebook’. Some examples of agent tasks are:

• Alarm-Based Optimization:  A cell site in downtown Chicago detects SINR degradation. An AI Agent steps in and automatically increases transmit power by 2 dB.
• Load Balancing and Parameter Tuning: In Manhattan, an LTE/5G site becomes overloaded with RRC connections. An AI Agent tunes handover thresholds to the neighboring cells to rebalance traffic.
• Coverage Optimization: A site in suburban Dallas faces coverage complaints. The AI Agent adjusts antenna tilt after analyzing OSS counters and drive test reports.

AI agents are reactive (Figure 1). They only respond to familiar events, e.g., congestion alarm, KPI breach, etc., according to the rulebook or a pre-trained machine learning (ML) algorithm/model and they stop once the job is done. While they are faster, more consistent and less error prone than humans, they are unable to multi-task, plan ahead and adapt to a dynamic environment. These limitations limit the utility of AI agents.

Figure 1: AI Agent Workflow

Agentic AI is a step-change over agents. An agentic AI application can both Reason and Act. It goes beyond linear automations and limited reasoning to goal-driven, adaptive intelligence. Consider it like a Senior Solution Architect, who doesn’t just fix issues but also strategizes, predicts, and plans. Instead of being told exactly what to do, like in the case of agents, the agentic AI application is instead given a high-level objective, such as “Maximize user experience at the Rajiv Chowk Metro Station during peak hours.” The application will then (Figure 2):

• Plan and Reason: Break the goal into sub-tasks, like coverage, interference, capacity, etc. that can be accomplished via collaborating agents.
• Run simulations to evaluate different strategies, such as antenna tilt, carrier aggregation, small cell activation and new small cell locations, etc.
• Choose the best approach, execute it, and learn from the outcomes.

Figure 2: Agentic AI Workflow

The following examples illustrate the capabilities of agentic AI:

• Autonomous Goal Execution: Consider a cluster-wide goal to maximize 5G Quality of Experience (QoE). Agentic AI breaks the goal down into sub-goals like interference reduction, mobility robustness, and capacity enhancement and executes them simultaneously.
• Proactive Planning: Prepare for traffic surges during IPL matches, Valentine’s Day, or, New Year’s Eve by predicting mobility and demand.
• Dynamic Tool Selection: An operator is facing coverage drops in Miami. Instead of relying only on OSS data, agentic AI pulls in CBRS spectrum usage data, weather forecasts (hurricane alerts), and UE reports to optimize more intelligently.
• Multi-step Reasoning: An operator is considering enabling carrier aggregation across its 600 MHz and 2.5 GHz layers. Agentic AI runs simulations of handovers, user QoE, and interference before deciding the best rollout strategy.
• Self-reflection: If a coverage tilt adjustment worsens drop calls, agentic AI rolls back and tries out a new strategy without human intervention.
• Multi-agent Collaboration: Different agents focus on interference mitigation and load balancing but they collaborate as orchestrated by the Orchestrator agent for holistic cluster-level optimization.
• Cross-domain awareness: Agentic AI application can integrate external tools and data sources like weather forecasts, ticketing data, or social media sentiment for better optimization.

Figure 3 compares a few characteristics of AI agents and agentic AI.

Figure 3: AI Agent and Agentic AI comparison

Why AI Agents and Agentic AI Matter to the Global Telecom Operators

For telecom operators across the globe, the use of AI Agents and agentic AI isn’t just a theoretical debate. Their adoption directly impacts how networks are built, optimized, and monetized in one of the most competitive markets. While AI Agents drive Immediate Operational Savings, agentic AI is fundamental to the adaptive and autonomous networks of the future.

Today, all global operators are under intense pressure to deliver better 5G performance while keeping costs under control. Average revenue per user (ARPU) has remained flat for years, even as customers demand faster data rates, broader coverage, and richer digital services.

AI Agents – Do More with Less

In this environment, AI Agents act as the first line of efficiency — taking over repetitive, rule-based tasks like alarm handling, antenna tilt adjustments, neighbor relation corrections, or routine KPI monitoring. By doing so, they can help reduce operational expenditure (opex), free up engineering talent for more complex challenges, and improve the day-to-day reliability of the RAN. In essence, AI Agents help carriers “do more with less,” a critical survival strategy when margins are progressively tightening.

Agentic AI: Building Networks of the Future

Cutting costs is only half the battle. To stay competitive against global peers and even hyperscalers like AWS, Google Cloud, and Microsoft, global operators need networks that are intelligent, adaptive, and self-evolving. This is where agentic AI comes into play. Unlike AI Agents, which follow predefined rules, agentic AI can autonomously sense, reason, and act at scale — learning from patterns, predicting future issues, and optimizing not just individual cells but entire clusters or nationwide footprints.

For instance, an agentic AI application could predict a surge in traffic during a Super Bowl event, reallocate spectrum resources in advance, balance energy use in real time, or even coordinate between terrestrial RAN and satellite backhaul for rural coverage. This shift transforms the network from a reactive system into a proactive, self-driving platform — one that is resilient, scalable, and ready to support new revenue streams like private 5G, IoT, connected cars, and immersive experiences.

AI Agent and Agentic AI Strategic Alignment

The adoption of agentic AI also aligns directly with telecom industry’s shared goals and policy initiatives. Some of these are depicted in Figure 4.

Figure 4: Key Applications Areas of Agentic AI in Telecom

The industry alignment is best reflected in their collaborative efforts towards standardizing architectural frameworks for building and interfacing agentic AI applications with the access network, core network and the management layer. The 3rd Generation Partnership Project (3GPP) service assurance (SA) study groups have laid a solid foundation for AI applications in ‘5G Advanced’ telecom networks and even beyond. See Figure 5 for their respective study items in release 18. This work is being developed further in releases 19 and 20 in order to harness the evolving capabilities of AI/ML.

Figure 5: 3GPP Release 18: AI-related Study Items
(Source: IEEE Comsoc CTN, 6 September 2023)

Conclusion: The Future is Hybrid

Both AI agents and agentic AI applications are useful in their own way and it’s not the battle of one against the other. The optimal approach should be to choose horses for courses: while agents handle repetitive, site-level optimizations, agentic AI applications orchestrate across clusters, cities, or nationwide networks. For telecom networks, AI agents deliver automations and agentic AI delivers autonomy. As operators prepare for 6G, Open RAN, CBRS spectrum sharing, and sustainability goals, the adoption of agentic AI is not just evolutionary, it’s essential for staying competitive.

About the Author

Mohinder Pal (MP) Singh is a proven technology leader with over 30 years of R&D leadership experience in large Tier-1 Indian and multinational telecom/IT organizations, specializing in software and product engineering, R&D operations, system engineering, and project/program management. He has been recognized for building high-caliber telecom and software R&D organizations, leading successful research collaborations with the academia and delivering impactful R&D outcomes aligned with India’s national technological priorities.

His current interest is to lead indigenous, cost-effective and sovereign AI/ML-driven innovations for India’s strategic technological infrastructure, e.g. Communications, Defense, Power, Agritech, Homeland Security, etc.

You may reach him on LinkdIn https://www.linkedin.com/in/mpsingh1912/