Current Status and Future Evolution of L4 Autonomous Network in the AI-Native Era

Autonomous networks (AN) are entering a critical stage toward L4 high-level autonomy, representing a key inflection point for the communications industry to shift from the "connectivity era" to the “AI-native era.” With the large-scale commercialization of 5G-Advanced and accelerated R&D of 6G, network nodes, service scenarios, and data traffic are growing exponentially. The traditional "manual + script" O&M model can no longer meet the operational demands of complex networks, becoming a major bottleneck for industrial upgrading. Meanwhile, the rapid iteration of new technologies such as large AI models, digital twins, and agentic AI is expanding the capability boundaries of traditional networks, driving their evolution from passive connection carriers to active intelligent service entities.

According to TM Forum, over 60% of the world’s leading operators have deployed L3 autonomous capabilities. Pioneers including China Mobile, Deutsche Telekom, Vodafone, and China Telecom have achieved L4 closed-loop operation in multiple high-value scenarios, signifying that autonomous networks have moved from concepts to large-scale deployment. Faced with the declining traffic dividends and the dilemma of "traffic growth without revenue increase", traditional models built on scale and cost efficiency are reaching their limits. Improvements in O&M efficiency alone are insufficient to offset stagnant revenue growth, and the core demand for autonomous networks has shifted from cost reduction to revenue generation.

Current Status of L4 Autonomous Networks

In 2025, industrial practices of L4 autonomous networks have broken through the limitations of traditional "automated scripts + rule engines", enabling a leap from "isolated intelligence" to "systematic intelligence". Based on TM Forum standards and practices of global top operators and vendors, the L4 phase now shows three core features.

First, AI deployment is evolving from scattered applications to systematic collaboration, with a full-lifecyle closed-loop as a key feature. L4 autonomous networks do not rely on the stacking of isolated AI modules; instead, they establish an autonomous, end-to-end closed-loop of perception, analysis, decision, and execution. This has been verified in the practices of leading global operators. China Mobile has achieved automated scheduling of 300 Tbps of daily IP network traffic and built cross-domain autonomous closed-loops based on its self-developed "Jiutian" network large model, reducing MTTR by 25%. Deutsche Telekom, in collaboration with Google Cloud, launched MINDR, a multi-agent system. Its early-deployed RAN Guardian Agent reduced the time needed to manage major events from hours to around a minute, an improvement of more than 95%. During the February Carnival season, it pre-checked 611 mobile sites and enabled automatic peak-load adjustments. Vodafone partnered with Cyient to build the VISMON™platform, which integrates multi-market and multi-vendor data to realize multi-agent collaborative optimization for radio access networks, improving spectrum efficiency by 19% and solving the pain point of traditional fragmented network management.

Second, agents are becoming a key enabler of AN L4 implementation, helping establish an "AI-led, human-supervised" paradigm. Unlike traditional AI that relies on manually defined rules, autonomous network L4 agents possess closed-loop capabilities, including goal orientation, autonomous planning, environmental adaptation, and continuous learning. Especially in multi-agent collaboration scenarios, they can efficiently complete complex tasks such as cross-domain fault self-healing and dynamic resource scheduling. ZTE and China Mobile jointly established the "Co-Innovation+" lab, building a collaborative system of expert-type and copilot-type intelligent digital employees, achieving over 90% accuracy in cross-domain fault root cause localization and a 21.34% reduction in MTTR. TM Forum’s Agent-to-Agent for Telecom (A2A-T) protocol provides a common language for agents across vendors and domains, reducing cross-domain integration cycles from months to days, driving multi-agent interaction from "manual integration" to "adaptive integration".

Third, the value measurement system is evolving from O&M KPIs to an operations-oriented KBI-KEI-KCI system, shifting focus from cost to profit. Operators’ goals are also expanding from basic O&M indicators such as shorter fault handling time and lower labor costs to the monetization of network capabilities. China Mobile Guangdong used NWDAF-based hierarchical assurance to provide low-latency cloud gaming packages for VIP users and improve Douyin data rates by 35.6%, directly driving ARPU growth. China Mobile Fujian’s home broadband complaint agent achieved over 90% demarcation accuracy and improved customer satisfaction by 22%. China Telecom empowered vertical industries through network large models, transforming autonomous capabilities into differentiated service advantages and winning multiple TM Forum innovation awards. These clearly show that AN L4’s commercial value lies not only in improving O&M efficiency, but also in enabling breakthroughs in operational revenue growth.

Future Trends of L4 Autonomous Networks

The real leap toward L4 AN lies in the integration of three paradigms: large models as the "cognitive brain,” agents as the "execution hub,” and digital twins as the “decision sandbox.” Large models overcome the limitations of traditional rule engines by enabling natural language understanding, expert experience accumulation, and long-horizon reasoning, driving the shift from "executing instructions" to "understanding intent.” Agents are goal-oriented with closed-loop perception, decision, execution and learning. Multi-agent collaboration supports cross-domain fault self-healing and dynamic resource scheduling. Digital twins provide a safe environment for validating AI decisions, ensuring reliability and reversibility through simulation.

Based on TM Forum standards, the practices of global leading operators, and the natural trajectory of technological evolution, autonomous networks are expected to evolve along three core trends over the next 3-5 years. Built on the three technical paradigms of large models, agents, and digital twins, these trends will reshape the underlying logic of network intelligence, advance the large-scale deployment of autonomous networks, and drive the industry toward high-quality development. The three core trends are as follows:

Trend 1: Architecture Evolves from Partial Intelligence to Full-Stack AI, with NE-Native Intelligence as the Strategic Core.

Today, most vendors’ autonomous solutions focus on AI at the NMS layer, resulting in delayed perception, slow response and cross-domain fragmentation. The essence of full-network L4 lies in NE-native intelligence: Every NE becomes an intelligent agent node with local decision-making, forming a distributed intelligent collaboration system and breaking away from centralized management.

ZTE proposes a four-layer agent architecture—NE, Network, Service, and Business—representing full-stack AI. In the future, the network will no longer be a "managed object", but a "living entity" with self-regulating capabilities. For example, wireless base stations equipped with built-in intelligent hardware and lightweight structured large models can perceive user mobility trajectories in real time and dynamically adjust beams. Integrating OTDR functions into optical modules enables second-level fault localization of fronthaul links, while embedded gSDU units support load-aware automatic on/off, achieving “zero load, zero power consumption”.

Going forward, those that first achieve NE-level AI-native capability and embed intelligence into every network node will gain a competitive edge in L4 high-level autonomous networks. This is also the inevitable direction for the architectural evolution of autonomous networks.

Trend 2: Operations Leap from O&M Efficiency to Revenue Growth, with Network-as-a-Service (NaaS) Unlocking a Second Growth Curve.

Against the backdrop of diminishing traffic dividends and intensifying homogeneous competition, operators are shifting from scale-driven to value-driven growth, with L4 autonomous network capabilities serving as a key enabler. The operational focus of autonomous networks is also shifting toward experience monetization. Networks will evolve from a bandwidth pipeline to a tradable, customizable intelligent service offering. Three major monetization paths are gradually taking shape.

First, monetization through differentiated experience packages. For example, customized services, such as low-latency guarantees for cloud gaming, uplink acceleration for live streaming, and SLA visualization for industrial leased lines are offered to different user groups, with charging based on experience quality.

Second, monetization via vertical industry empowerment. In sectors such as the low-altitude economy and the industrial Internet, 5G-A integrated communication and sensing networks provide drones with integrated “communication + sensing + navigation” services, with charging based on route and duration.

Third, monetization through data elements. After desensitization, network data can empower scenarios such as automaker site selection, retail heat analysis and financial risk control, building a “data bank” that unlocks data value under compliance.

The NWDAF-driven tiered and graded experience assurance solution jointly launched by ZTE and China Mobile Guangdong delivers an average of 3.65 assurance actions per user per day, providing personalized protection for tens of millions of users. This demonstrates the feasibility of turning network capabilities into commercial product and provides a replicable path for operators’ business transformation.

Trend 3: The Ecosystem Is Evolving from Single-Vendor Closed-Loop to Open Collaboration, with Standards Shaping Industry Competitiveness.

L4 autonomous networks span multiple domains including radio access, core network, transport, IP, services, and business. No single vendor can provide full-stack capabilities, so open collaboration has become essential. Over the next three to five years, competition in autonomous networks is expected to shift from products to ecosystems, with standards playing the key enabling role.

Global standardization is accelerating. TM Forum is promoting multi-agent collaboration and the A2A-T protocol, addressing coordination challenges through structured templates, event subscription, and fine-grained authorization. Leading operators, including China Mobile, Vodafone, and Orange, are actively involved. CCSA is advancing standards for large models and agents, building a "general + specialized" framework. 3GPP is enhancing AI security specifications, providing a foundation for AI-native security and standardized safeguards for trusted agent interactions and global collaboration.

ZTE, together with China Mobile, has established the "Co-Innovation+ Autonomous Network Lab" to explore cross-vendor technical cooperation. It promotes the compatibility of the A2A-T protocol with mainstream frameworks such as MCP and ACP and carries out practical verification to enable "plug-and-play" for cross-vendor agents. Looking ahead, those who build an open agent factory, shared test suites, and a unified governance framework will lead the ecosystem of autonomous networks, shaping the future of large-scale, trusted, plug-and-play cross-vendor and cross-domain agents.

Conclusion

The large-scale deployment of L4 autonomous networks marks the industry’s transition from an era dominated by manual operations to a new AI-native stage. We are at a historic inflection point: AI has evolved from a network add-on tool into the neural center spanning NEs, architecture and operations. The network is no longer a passive pipeline but a digital living entity powered by agent collaboration, capable of perception, reasoning, decision-making, and evolution.

Autonomous networks will follow three pillars—full-stack AI, open collaboration, and value-driven development—enabling real-time perception, autonomous decision-making, and continuous evolution. L4 is not the destination, but the starting point for autonomous networks to evolve toward higher-level intelligence. We call on global operators, vendors, standards organizations and vertical industry partners to collaborate in deepening full-stack AI innovation, building an open and win-win ecosystem for autonomous networks, and advancing autonomous networks toward higher intelligence, empowering the high-quality development of the digital economy and industries.