Networks today are frequently tested by changing traffic demands, whether from professionals moving between corporate hubs and their homes or one-time events drawing large crowds to a specific venue. In a traditional network, engineers need to manually reallocate resources, which can lead to service interruptions or degradation, impacting the end user experience. With the power of artificial intelligence (AI) applied in networks, these challenges are addressed as autonomous networks detect, predict and resolve surges before most users even notice a hiccup. This isn’t a distant vision — it's the operational promise that's driving one of the most significant shifts in telecommunications today.

This evolving approach represents how telecom networks are becoming more and more sophisticated — and how the industry is shifting away from manual, rule-based operations to embracing intent-based autonomous networks. This change is the latest step down the path toward self-configuring, self-optimizing and self-healing networks. It allows mobile network operators (MNOs), vendors and other telco stakeholders to set and achieve new possibilities in spectral and operational efficiency, as well as predictive network maintenance.

Key Elements of Autonomous Networks

Autonomous networks are not just a futuristic concept — they are the cornerstone of the next generation of telecom operations. Automation is a critical enabler in the journey toward autonomous networks, but it’s important to note that automation and autonomy aren’t synonymous — true autonomy extends beyond automation. To understand what makes a network truly autonomous, it helps to start with how the industry defines autonomy. TM Forum, the industry association that developed the Autonomous Networks framework and its maturity levels, describes autonomous networks through two fundamental dimensions

The first is Self-X capabilities: the operational engine of autonomous networks. These are self-configuration, self-healing and self-optimization — working in concert so that the network can manage itself across its full lifecycle without manual, step-by-step intervention. The second is the Zero-X experience: the outcome those capabilities make possible. Zero wait, zero touch, zero hassle — for both operators and end users.

Structurally, the architecture is defined by three operational layers — Business, Service and Resource — connected by closed control loops that drive continuous, automated decision-making across each layer. Rather than a single centralized system, the architecture is built on Autonomous Domains: a set of systems or platforms that can fulfill closed-loop automation of certain network operations. Holding these domains together is an intent-driven interface model: operators input high-level business objectives, and the network’s automation layers translate those intents into concrete actions — without requiring manual intervention at each step.

The industry looks at network autonomy as a maturity journey across six levels (L0-L5), from fully manual operations at L0 to complete closed-loop autonomy across all services, domains and lifecycle stages at L5. This graduated model gives operators a structured, measurable path — anchoring automation investments to defined capability thresholds. The industry is making efforts to advance through these stages: in June 2026, the TM Forum surveyed 80 global operators and found that 20% expect to reach Level 4 or above by 2027, while 81% are targeting Level 4 or above by 2030.

True network autonomy is not achieved by deploying many individual tools. Instead, it relies on a continuous, intelligent lifecycle governed by business goals. A highly autonomous L4/L5 network operates through five tightly integrated cognitive stages:

1. Intent (The Goal): The governing authority of the loop. Operators set a high-level business or operational goals (e.g., "minimize energy consumption across the network while optimizing performance and coverage") and the network autonomously translates these "intents" into dynamic policies, minimizing manual step-by-step intervention.

2. Awareness (The Senses): Beyond basic monitoring. Through observability frameworks, the network continuously analyzes multi-dimensional telemetry — metrics, logs, traces, topology — to build a real-time, comprehensive understanding of its operating environment and state.

3. Analysis (The Brain): Leveraging predictive AI and machine learning, the system evaluates the gap between the network's current status and the defined objective. It predicts upcoming bottlenecks and performs root-cause analysis on anomalies before they impact services.

4. Decision (The Logic): The system generates and simulates multiple resolution strategies, autonomously selecting the best action to satisfy the objective without violating conflicting SLAs or policies.

5. Execution (The Action): The decision is seamlessly executed. Using standardized Open APIs, the system triggers precise configurations, scaling or self-healing actions across multi-vendor domains — closing the loop without requiring human intervention.

The application of these stages allows networks to dynamically adapt to changing conditions and ensure optimal performance as well as reliability. The engine fueling these stages and the journey toward fully autonomous networks is the power of AI and machine learning (ML).

The Role of AI/ML and Telco-Grade LLMs

AI and ML leverage vast amounts of data housed within telco networks to identify patterns, predict trends and automate decision-making, enabling networks to operate efficiently and respond to real-time demands. Large Language Models (LLMs) extend these capabilities by interpreting complex intents, analyzing telemetry data and providing actionable recommendations. Telco-grade LLMs, fine-tuned with domain-specific knowledge, offer the precision, reliability and explainability required for mission-critical environments.

Telco-grade LLMs can be used for network planning and optimization by analyzing urban development trends and predicting future capacity needs, ensuring that infrastructure investments are aligned with demand. LLMs can also enhance cybersecurity by detecting and mitigating threats in real-time through advanced anomaly detection and automated response protocols. Performance optimization can be achieved by identifying unusual patterns that indicate issues like sleeping cells or network overload. This process allows for early detection and corrective actions to sustain optimal network performance.

By integrating AI/ML and telco-grade LLMs, autonomous networks shift from reactive automation to proactive, adaptive and monetizable connectivity — allowing networks to achieve performance expectations while unlocking new revenue opportunities. For example, a telco-grade LLM might ingest thousands of network telemetry signals and surface a plain-language recommendation to an engineer, such as: "Sector X is showing early signs of congestion consistent with a local event; recommend preemptive reallocation of spectrum from the adjacent sector." AI can further improve observability across the RAN, enabling timelier and more accurate anomaly detection based on valuable insights into the network’s behavior. Additionally, LLMs can dynamically analyze network traffic, predict congestion and reroute data. These capabilities can be packaged as premium usage APIs, allowing businesses to ensure reliable, high-performance connectivity. Premium-tier billing can be applied for deterministic latency and guaranteed quality of service for high-priority traffic or mission-critical traffic.

Agentic AI: A Leap Forward

Agentic AI is a significant development in the evolution toward fully autonomous networks. Rather than a single, centralized AI making all decisions, agentic systems deploy modular AI agents — each responsible for a specific domain while collaborating with other agents to improve overall performance.

When considering agentic AI in the context of TM Forum's levels, the primary distinction is less about when AI appears and more about what it's permitted to do. From L0 to L3, AI functions mainly as an analytical layer — identifying anomalies, forecasting traffic, recommending actions — but human engineers keep ownership of the final decision. The step into L4 is where agents start to change the equation: rather than presenting options for a human to execute, agents can translate high-level business intent directly into network action. L5 will extend this further — removing the human-dictated objective, with agents deriving and pursuing goals based on their own network insights.

Samsung is making tangible efforts to realize highly autonomous networks with its comprehensive, AI-powered, O-RAN compliant automation solution, CognitiV Network Operations Suite (NOS). The core of this solution is the AI Agent Fabric, a multi-agent system that integrates Generative AI and specialized domain knowledge to drive true end-to-end autonomy. Rather than relying on static scripts, this architecture operationalizes a continuous cognitive loop, aligned to the TM Forum’s cognitive stages. Throughout the process, the LLM-powered AI agents leverage Retrieval-Augmented Generation, Model Context Protocol, knowledge ontology, cross-domain data and digital twins to boost its context awareness, decision making and execution. This approach ensures seamless integration of advanced Agentic AI technologies into the network lifecycle.

By unifying this intelligence, CognitiV NOS empowers operators to automate the entire network lifecycle — spanning planning, installation, maintenance, troubleshooting, optimization and user experience management — accelerating the journey toward complete L4/L5 autonomous networks.

Industry-wide Standardization

With all this momentum toward autonomous networks, the TM Forum and other standard bodies like the O-RAN Alliance, the AI-RAN Alliance and 3GPP provide the foundational groundwork for scalable, secure and interoperable networks. The TM Forum, all anchored by AI-native ODA, is driving the development of frameworks that align business objectives with technical implementations, ensuring that autonomous networks deliver tangible value to operators and end-users. The O-RAN Alliance is focusing on open and interoperable architectures, enabling multi-vendor collaboration and reducing vendor lock-in, while the AI-RAN Alliance is on a mission to transform networks through AI innovation and collaboration. 3GPP is defining the technical specifications that ensure seamless integration of autonomous capabilities across global networks.

Samsung currently holds multiple chair and vice-chair positions across 3GPP Technical Specification Groups and Working Groups, and co-chairs multiple working groups within the O-RAN Alliance. Samsung is also a founding member of the AI-RAN Alliance, currently serving as the Vice Chair of its Board of Directors and leading the Alliance's Agentic AI Task Group, focused on exploring the application of agentic AI for autonomous network operations and management.

Together, these organizations and the industry leaders involved with them are ensuring a unified ecosystem that fosters innovation, accelerates deployment and ensures that autonomous networks can meet the needs of operators and enterprises. This collaborative effort is essential for realizing the full potential of autonomous networks.

Embracing the Autonomous Future

The transition to autonomous networks marks a new era for the telecommunications industry. With the help of AI/ML, Agentic AI and Generative AI, MNOs will be able to automate their network operations while continuing to deliver scalable, efficient connectivity aligned with their individual business goals.

Organizations that move deliberately and meaningfully toward AI — like those around the globe already leveraging the power of Samsung CognitiV NOS — will be best positioned to reduce costs, respond faster and deliver better experiences for all stakeholders, from operators to vendors. As the industry continues to evolve, these technologies will play a critical role in driving innovation and ensuring networks can dynamically respond to evolving demands.