Telenor wins Level 4 AN hackathon with intent-based network slice solution

For communications service providers (CSPs) investing in increasingly intelligent, dynamic and automated network operations, AN Level 4 is an important milestone. Part of the six-level taxonomy developed by TM Forum members so that CSPs can measure their progress from manual to fully autonomous network operations, it is significant because it represents a major shift towards true autonomous decision-making.

But achieving AN Level 4 is complex and requires new processes, technologies and operating models. So, during TM Forum’s DTW Ignite event, Google Cloud organized an Autonomous Networks Hackathon where CSPs could build use cases supported by TM Forum code frameworks and Google Cloud’s AI infrastructure and technology, including Gemini 2.5 and agentic systems.

CSPs typically focus on “high-value scenarios” – such as fault management, service quality assurance, network optimization and energy efficiency – where automation can deliver the greatest impact on cost, customer experience and network resilience.

The hackathon, which took place over two days, centered on CSPs’ efforts on enhancing network automation, sustainability or customer experience, with Jio, Telefonica and Telenor taking part.

Analyzing network slice performance

The winning team from Telenor combined expertise in AI from Weiqing Zhang, Senior Research Scientist, and Claudia Battistin, AI program Director; in networking from Min Xie, Platform Lead and Senior Research Scientist; and AI in the Cloud from Peiqing Zhang, Senior Expert, Cloud and AI, Telenor, as well as Stefan Dubbelman, Active Monitoring Specialist, Emblasoft.

Together they tackled an obstacle to network operations center (NOC) automation they encountered when using Telenor’s internal experimentation network platform to perform 5G slice monitoring and to recommend solutions that proactively enhance customer experience.

The team required network intelligence to perform monitoring, orchestration, and assurance. However, network components within a heterogeneous and multi-vendor environment are not designed for easy data exchange and collaborations. Nor was it realistic to replace existing equipment with new intelligent, AI-capable components.

The Telenor team believes that “telecommunications companies should promptly integrate MCP and A2A protocols to avoid repeating the industry’s pattern of delayed adaptation to major architectural changes, such as the current shift from REST APIs to AI-native protocols”.

Their solution, which is currently a proof-of-concept, therefore uses MCP and A2A to facilitate network self-governance, enabling AI agents to autonomously discover, interpret, and manage network resources while coordinating workflows. In this way network functions can be made discoverable and actionable within agents’ decision-making cycles, thereby supporting self-managing networks capable of predicting, preventing, and resolving issues while continuously optimizing performance.

In addition, the complexity of the experimentation platform, "caused by running advanced and less-mature technical solutions from multiple vendors, makes it difficult to remain stable and reliable all the time. So, we frequently experienced errors and failures,” says Min Xie.

Faced with unknown and unprecedented errors, Telenor’s teams often resorted to manual troubleshooting to identify and resolve faults reactively when operating and managing slicing services on the trial network. The team’s experience of the challenge of automating the NOC in a complex environment suggested the process would benefit from using Intents and AI agents.

Delivering with intent

TM Forum defines Intent as “the formal specification of all expectations including requirements, goals, and constraints given to a technical system”. For Telenor’s network slice NOC, intents serve as high-level commands expressed in natural language that are automatically translated into network policies and configurations to, for example, self-configure or self-heal networks.

The hackathon provided the opportunity to use AI agents and standardized API interfaces to enable exchange between different network components and to further automate observability and monitoring capabilities.

“AI agents [help] address the challenge, because now we can offload all sorts of the intelligence tasks to AI agents by providing the necessary APIs,” according to Peiqing Zhang.

Deploying an active monitoring system in the experimentation platform means AI agents are alerted if the slice performance drops below a pre-defined threshold for each service level metric. The AI agents then query the monitoring system for the required metrics via APIs.

A wide array of service level metrics are available to the agents, covering both the control plane and user plane performance, according to Telenor.

“What we realized was the importance of very detailed metadata for the metrics to make it possible for the AI agents to understand and classify the severity of the alert”, says Stefan Dubbelman.

The metrics can be provided either as real-time or historical data, depending on whether there is a need for further fault analysis, for example, or recovery actions.

“We adopted a multi-agent architecture, consisting of specialized AI agents, each responsible for a specific stage in the network service lifecycle — such as alarm classification, root cause analysis, workflow optimization, customized slice recommendations, and intent management,” according to Weiqing Zhang. “What’s more," he adds, "these AI agents can share data, invoke each other’s capabilities, and pass tasks along without relying on manual intervention or a single coordinator. Each agent focuses on its own role while collaborating to complete complex business processes through orchestration, thereby achieving efficient and automated business closure.”

In Telenor’s solution, these agents work together to process and analyze operational data from partner monitoring tools, including those of Emblasoft, quickly pinpointing issues, identifying root causes, and initiating corrective actions. For instance, the framework can trigger the creation of a new network slice, which is then validated through direct customer interaction.

Intent Management agents are central to this process, explains the Telenor team, converting high-level business objectives—expressed in natural language—into precise network commands that are executed automatically, minimizing manual intervention. For usability, the solution includes a business customer portal that visualizes KPIs, interacts directly with customers, checks account balances, provides tailored recommendations, and—once approved—generates intents via the TM Forum Intent Management API (TMF921). This ensures service delivery that is intelligent, automated, user-friendly, and fully standards-compliant.

Partnering to close the loop

Min Xie emphasized the importance of Telenor’s long-term collaboration with technology partners in addition to Google Cloud and TM Forum, which fed into the solutions developed during the hackathon.

“In order to realize AN Level 4, we need to close the loop between the network and [the] AI agents framework. One direction is from the network to AI agents,” she explains. This entails exposing “the network data, KPIs and metrics provided by Emblasoft software, which specializes in monitoring and testing” and is supporting Telenor in its development of dynamic and automated network slicing.

“The other direction is from the AI agents to the network, with AI agents giving recommendations in the form of intents,” according to Min Xie. The “intents will be sent from AI agents to the network for execution via the network orchestration components, which are provided by Nokia,” she explains.

The hackathon revealed “there is huge potential to advance further, and we are prepared to proceed with a more mature solution,” Min Xie concludes.

“The Google Cloud Autonomous Networks Hackathon was a powerful showcase of how innovation and collaboration are driving the future of autonomous networks,” said Aaron Boasman-Patel, Vice President of AI, Labs & Innovation at TM Forum. “In just two days, three talented teams used TM Forum’s assets and Google’s advanced tooling to build forward-looking solutions that tackled real-world challenges.”