AI-Native networks and autonomous cores: how telecoms Is re-architecting the network for the AI era

The use of AI to bring autonomy into telecoms network operations was a dominant theme at this year’s DTW 2026 event in Copenhagen. While “AI” and autonomy” are deeply connected, there are specific industry initiatives behind, on the one hand, autonomous networks and, on the other, AI-native networks. While autonomous network operations has already established a clear framework through TM Forum's AN levels, AI-native networking is an emerging concept.

The two developments are deeply connected. Both seek to address the growing complexity of modern telecoms networks by shifting from human-led operations towards intelligent, agent-driven systems capable of diagnosing issues, making decisions and ultimately taking action autonomously. Together, they represent a fundamental transformation in how networks are designed, operated and maintained.

Two panel discussions at DTW explored the themes of, on the one hand, AI-native networks and, on the other, AN in core networks.

Beyond cloud native: defining the AI-native network

Given that AI-native networks is a relatively new term, panellists were asked to give their vision of what exactly it means.

For Gao Yu, President of Intelligent Telco Cloud Domain at Huawei, AI-native represents something much more significant than simply deploying AI applications on cloud-native infrastructure. "AI native is more than just an AI app running in cloud. This is too simplistic. think it's a reconstruction of the service layer, the network and the infrastructure layer."

Huawei's vision involves a comprehensive transformation across the telecom stack. At the service layer, traditional network functions evolve into agents capable of communicating and collaborating autonomously. At the network layer, virtual network functions (VNFs) and cloud-native functions (CNFs) give way to AI-native functions（ANFs）. At the infrastructure layer, operators must rethink both hardware and software architectures to combine traditional computing resources with AI processing capabilities.

The concept reflects a broader industry shift. While cloud-native technologies helped operators automate and modernise network deployment over the last decade, panellists agreed that AI introduces a fundamentally different set of requirements.

Jill Lovato, Director of Marketing Communications at the Linux Foundation contrasts the relative maturity of cloud native with the still-emerging definition of AI native.

"Cloud native has been around quite a bit longer. When it comes to AI native, I sort of feel like it's still being defined."

She argues that AI-native networking is principally about re-architecting operations so that AI agents perform much of the operational workload while humans become supervisors and governors of increasingly autonomous systems.

"It's about re-architecting the way networks are operated to take the human out of the picture, but to kind of refocus the job of human and have the AI agents do a lot of the work."

Complexity driving the need for AI and autonomous network operations

For CSPs AI-native is needed to address network complexity which has grown to beyond what humans can efficiently manage.

David Low, director of AI and machine learning at Singtel, highlighted the operational reality facing telecoms operators today. "It's getting more and more complex in terms of the number of parameters, the dependencies, the configuration decisions that need to be made by humans." The challenge is not that humans cannot manage modern networks, he explained, but that relying exclusively on human decision-making limits speed, agility and operational efficiency.

Boonchoung Tansuthepverawongse, head of Network Data and OSS Management Unit at AIS, noted that the deployment of a new 5GSA core network – and rolling out services such as network slicing and and private networks – has been the catalyst for deploying AI and AN capabilities. Without increasing the size of the team, deploying autonomous systems was the only option for managing the complexity of the new 5G core.

Shen Chengying, president of Core Network MAE Domain at Huawei, pointed that as network scale and complexity continue to grow, operations and maintenance (O&M) staff are under increasing pressure—making it critical to improve O&M efficiency. Meanwhile, core network failures tend to have wide-ranging impacts, so high stability must be guaranteed.

AI therefore becomes a mechanism for accelerating diagnosis, decision-making and resolution. "This is where AI Native comes into pictures to be able to do the diagnosis, to be able to do the decisioning, and finally to act on the certain recommendation or fixes required."

However, operators remain cautious about fully autonomous decision-making. While agentic AI is being deployed today for diagnosis and recommendations, execution remains largely under human oversight. "That still requires our engineers to do the final check before we roll out the deployment or the patches to the system."

Reconciling probabilistic AI with deterministic networks

One of the most interesting debates centred on a fundamental tension between the deterministic requirements of telecom network and the probabilistic nature of Generative and Agentic AI.

Qian Bing, Deputy Director, Cloud and Network Department at China Telecom's reckons this is one of the industry's most important challenges. "Telecom networks require 99.999 certainty and high availability. But large models are essentially about probability, and they can be black boxes full of hallucinations."

This challenge becomes especially acute as operators seek to move beyond isolated AI applications and realise a much broader vision. towards multi-agent environments capable of autonomous planning, learning and execution. "We are moving from network-oriented passive maintenance to service-oriented proactive operations by large models." The ultimate goal is a telecom environment capable of intent understanding, task planning, continuous learning and autonomous execution.

Singtel is also grappling with the challenge of deploying probabilistic technologies in a deterministic environment. The operator has established standardised "skills" for AI agents and implemented guardrails designed to control the probabilistic behaviour of large language models.

Open collaboration becomes essential

Panellists made a strong case for deeper collaboration between CSPs, vendors and industry associations such as TM Forum and the Linux Foundation

The Linux Foundation's Lovato argued that opensource collaboration will be critical because the technology is evolving too quickly for organisations to work independently. "We just need to iterate faster. And if we're all doing it together, we're not going to be duplicating efforts."

In June, Linux announced the Open Autonomous Networks (OpenAN) project, a collaboration with Huawei, China Mobile and the GSMA to develop a common technical foundation for collaborative telecom AI agents.

Huawei's Gao Yu similarly emphasised the importance of cooperation across operators, vendors, standards bodies and open-source communities.

"We need to work together to define the way and define the step by step how we can go to the AI"

Autonomous Networks move from vision to execution

While AI-native networking is an emerging theme and focus, autonomous networking is progressing more rapidly and moving from vision to execution.

Olta Vangjeli, TM Forum’s autonomous networks programme director, says that the implementation of AN L4 can be divided into two phases. The first phase, through 2027, will primarily focus on automating high‑value scenarios, with a particular emphasis on fault management and network change scenarios within the core network domain. The second phase, from 2028 to 2030, will focus on achieving autonomy within the core network domain.

There is a big focus on fault management as a high value scenario within core networks. Shen Chengying, President of Core Network MAE Domain, explained how automation can dramatically improve fault management by analysing root causes and accelerating service restoration. Huawei’s AI-driven core network AN L4 solution can reduce mean time to repair (MTTR) to less than 15 minutes in certain scenarios, according to Chengying.

AIS has already achieved significant progress in core network fault management and operational stability, according to Boonchoung, AIS's core network AN level assessment reached a score of 3.6 last year. Huawei's cloud‑network visualization solution has significantly improved the efficiency of cross‑layer root cause analysis (RCA) and signaling storm prevention solution has reduced network assessment time from months to minutes. Meanwhile, AIS is also conducting trials of a fault management agent, which will cut fault handling time from hours to minutes. It is aiming to extend Level 4 autonomy across the entire core network to totally eliminate service interruptions.

That objective reflects the broader sentiment expressed during both discussions. Whether framed as AI-native networking or autonomous networking, the industry's goal is not automation for its own sake. Rather, it is about creating networks that are more resilient, efficient and capable of dealing with growing complexity without relying on ever-expanding human operational teams.

Videos of the full panel discussions can be viewed here -

Why network autonomy is critical for the future of core networks

How AI-native networks will reshape telecom operations