As demand for high-capacity, low-latency connectivity surges, carriers are increasingly turning to advanced automation to manage complexity. According to a recent Omdia report, 41 percent of communications service providers (CSPs) now cite network management as the area where “agentic” AI will have the greatest impact. This shift is not limited to peripheral customer-care tasks. Instead, agentic AI is being positioned as a fundamental — infrastructure-level — driver of network reliability, efficiency, and scalability.
What “Agentic AI” Means for Network Operations
Traditional network automation typically relies on predefined workflows and static thresholds — triggers set by human engineers. By contrast, agentic AI introduces autonomous, goal-oriented decision-making: multiple AI agents monitor network telemetry in real time, interpret conditions, evaluate trade-offs (performance vs. cost vs. stability), and then execute corrective or optimization actions — sometimes without human intervention.
Recent academic work underscores the viability of this approach. A 2025 study titled Edge Agentic AI Framework for Autonomous Network Optimisation in O-RAN showed that an edge-based multi-agent system could drastically reduce outage risk in 5G deployments, compared with traditional fixed-rule networks.
Similarly, a contemporaneous paper, Agentic AI for Ultra-Modern Networks: Multi-Agent Framework for RAN Autonomy and Assurance, demonstrated that a distributed, collaborative architecture of specialized agents can not only manage routine tasks — like resource allocation or capacity balancing — but also preserve system-wide stability under dynamic conditions, avoiding unsafe policy changes that could degrade network-wide performance.
Thus, agentic AI brings:
- Real-time anomaly detection — analyzing continuous streams of telemetry to spot unusual behavior or precursors to failure.
- Predictive maintenance and pre-emptive interventions — forecasting hardware issues or capacity bottlenecks before they impact service.
- Closed-loop automation and self-healing — automatically executing configuration changes, resource reallocation, or fault remediation without human triggers.
- Dynamic resource optimization — adjusting RAN parameters, transport and core network resources, or fiber backbone configurations in response to shifting demand or network state.
Why the Shift Matters — and Why Now
The timing of this shift reflects broader structural pressures on telecom operators. As networks evolve to cloud-native cores, multi-vendor Radio Access Networks (RAN), edge compute, and hybrid fiber-wireless deployments, the volume, velocity, and variety of telemetry data have exploded. Manual or static-rule management — once sufficient — now struggles to keep pace.
Moreover, flat or slowly growing ARPU (average revenue per user) means carriers must optimize existing infrastructure, reduce operational costs, and find new ways to differentiate — not simply build more fiber or deploy more base stations. Agentic AI promises both lower OPEX (fewer manual interventions, fewer emergency field dispatches) and higher reliability — a critical factor for enterprise, wholesale, and latency-sensitive 5G segments.
Finally, advances in AI — including multi-agent systems, real-time telemetry ingestion, and digital twin modeling — have matured to the point where the risks of errant automation are more manageable and the potential return on investment is compelling.
Early Implementations: RAN, Fiber, and End-to-End Networks
Some of the earliest adoption of agentic AI-driven autonomy is in 5G RAN and transport layers. For example, Nokia reportedly uses AI-powered discriminative analytics to detect anomalies, forecast traffic surges, and automatically optimize resource allocation — including closed-loop fault resolution in its Digital Operations Center.
On the fixed network and fiber side, AI/ML-based closed-loop automation has been shown to improve resource allocation fairness, optimize throughput, and accelerate fault resolution — reducing outage durations and enhancing stability. Looking ahead, providers and vendors such as ZTE are contributing to global efforts to standardize autonomous network architectures, laying out reference models for “Level 4 highly autonomous networks” that integrate AI, digital twins, and modular orchestration to deliver end-to-end automated O&M across wireless, transport, and IP domains.
Challenges and Considerations
Despite clear momentum, several material challenges remain:
- Data quality and integration: Agentic AI requires consistent, high-fidelity telemetry across the RAN, transport, core, and, possibly, the edge. Fragmented or delayed data pipelines can undermine inference validity or lead to misaction.
- Governance, explainability and safety: Autonomous decision-making must be auditable and reversible. Operators must define boundaries, escalation paths, and oversight mechanisms before enabling full autonomy.
- Transition management and human factors: Migrating from legacy NOC workflows to agentic operations requires retraining, restructuring, and cultural change — from reactive operators to outcome-focused overseers.
- Interoperability and standards: Because many networks are multi-vendor and multi-technology, ensuring agentic components can interoperate across domains (RAN, transport, core) is essential — and efforts are still nascent.
The telecommunications industry is in the midst of a structural transformation: as networks grow in scale, complexity, and criticality, carriers are embracing agentic AI-driven autonomous network operations as the next logical evolution. By integrating real-time anomaly detection, predictive maintenance and closed-loop automation, operators can dramatically reduce operational costs, improve reliability, and support the demands of 5G, fiber, and future network technologies. While challenges — especially in data quality, governance, and organizational change — remain real, the trajectory is clear: the future of network operations will be far more intelligent, adaptive, and self-managing.
