For years, telecom operators have relied on automation and machine learning to manage increasingly complex networks. These tools have made operations faster and more efficient, yielding improvements from alarm correlation to predictive maintenance. Yet, even with these advances, networks often remain reactive — detecting what’s wrong, but not fully understanding why or how best to fix it.
This is where the next necessary evolution in network management begins: artificial intelligence (AI) agents and agentic AI.
Rules and Models Still Matter
Rule-based automation has been the foundation of telecom operations for decades. It’s reliable, fast, and perfect for repetitive tasks: If event X occurs, then perform workflow Y.
These deterministic rules deliver consistency and predictability, which are critical for large-scale operations.
Machine learning (ML) added intelligence to that foundation. By analyzing vast volumes of telemetry, ML models can detect anomalies, identify performance trends, and predict failures before they occur. They’re invaluable for pattern recognition and forecasting. But both approaches have limits. Rules can’t adapt to new or ambiguous scenarios. ML can identify a pattern but doesn’t grasp the context or causality behind it. In complex networks, in which symptoms span multiple domains (optical, IP, and access), troubleshooting often requires more: the ability to reason.
The Rise of Reasoning
AI agents, powered by large language models (LLMs) and domain-specific intelligence, bring that reasoning layer. They can interpret alarms, correlate data across layers, test hypotheses, and determine the best corrective actions, all while continuously learning from outcomes.
Where rule engines execute predefined actions and ML models predict what might happen, AI agents can think through scenarios and decide what should happen next.
Consider a fixed access network, such as a passive optical network (PON). When several optical network terminals (ONTs) show degraded performance, how do the different techniques perform?
- A simple rule might restart an ONT and hope it fixes the issue.
- An ML model can detect and predict anomalies based on past performance.
- An AI agent, however, performs a more thorough investigation. It analyzes optical power, configuration history, and recent traffic changes, reasons through dependencies, and identifies a misaligned topology or damaged splitter as the root cause. It then recommends or executes the optimal fix, safely and autonomously, even before the customer calls to report an issue.
That’s the power of contextual reasoning.
Why Reasoning Matters in Fixed Access Networks
Fixed access networks are the backbone of broadband connectivity. They’re also complex systems, in which a single issue can ripple across multiple layers. A drop in optical power may look like end-user equipment failure but could stem from external temperature fluctuations or issues with the optical distribution network (ODN), like damaged connectors or splitters.
AI agents can use reasoning to connect cause and effect across these domains. Such agents can move from handling alarms to reasoning root causes — analyzing symptoms, verifying hypotheses, and acting through open interfaces. This leads to both faster fault resolution and self-healing networks.
With digital twins, AI agents can simulate actions, assess risks, and test “what-if” scenarios without impacting live subscriber traffic. This adds a critical safety and integrity layer, transforming automation into a confident, data-validated decision-making process.
How Agentic AI Reinvents the NOC
Telecom network operations centers (NOCs) manage vast data, alarms, and services, often leading to alarm fatigue and fragmented troubleshooting. Agentic AI transforms the NOC by providing reasoning, autonomy, and collaboration:
- Smarter alarm management: Correlates and prioritizes alarms, bringing only actionable issues to the surface and reducing noise.
- Autonomous root-cause analysis: Understands the “why” behind anomalies, simulates fixes via digital twins, and recommends or executes corrective actions safely.
- Proactive coordination: Anticipates issues, optimizes traffic, maintains key performance indicators (KPIs), and collaborates across functional domains.
- Knowledge retention: Retains lessons from every incident, building a knowledge graph for consistent, scalable troubleshooting.
- Human-on-the-loop collaboration: Engineers supervise and refine decisions, ensuring trust and accountability.
The NOC evolves from reactive operations to autonomous network operations, with AI agents handling most tasks while humans guide strategy.
Human expertise remains central as operators integrate agentic intelligence into their access architectures. Engineers set goals, guardrails, and policies; AI agents handle the scale and speed. It’s not about replacing people — it’s about augmenting them with intelligent partners that never stop learning.
Rule-based systems will continue to automate the predictable. Machine learning will keep driving foresight. But AI agents bring reasoning — the ability to sense, think, and act — turning networks into adaptive, self-managing systems. This is the true path from automation to autonomy in telecom networks.
Sireesha Kora, Director of Engineering
Nokia
Sireesha Kora is Head of R&D for SDN Controller applications and AI & ML in Fixed Networks, Nokia. Sireesha brings 15+ years of extensive experience working in the data and telecommunications industry. She started her career as a Software Engineer & Architect designing software for PON systems and has subsequently led several R&D teams for GPON, XGS-PON * NGPON2. Sireesha has been instrumental in bringing Cloud, network automation and AI/ML technologies into Fixed Access and has worked extensively with several service providers around the globe in this area. Sireesha is recognized as a Distinguished Member of Technical Staff for her significant contributions in Fixed Access innovation.
