The telecommunications industry has spent years teaching networks how to manage themselves. Now comes a rather more complicated challenge: teaching them how to work together.
Artificial intelligence is already transforming network operations, from fault detection and service assurance to energy optimization and radio network management. Increasingly, these functions are being performed by intelligent agents capable of analyzing conditions, making decisions and executing tasks with progressively less human intervention.
But there is a fundamental problem. An intelligent agent operating inside one vendor’s radio access network is of limited use if it cannot communicate effectively with another agent responsible for transport, core network operations or service assurance.
Autonomy within individual network domains is one thing. Coordinating autonomous operations across an entire multivendor network is quite another.
That is the challenge behind OpenAN, an industry initiative jointly advanced by Huawei, China Mobile and other partners to establish an open framework for intelligent agent collaboration.
At the Decoding China event on October 27, Fang Xiang, President of Huawei’s Wireless Product Line, outlined the latest progress of the initiative and its ambition to accelerate the commercial deployment of Level 4 autonomous networks.
At the heart of OpenAN is A2A-T, or Agent-to-Agent for Telecom, a protocol designed to give intelligent agents from different vendors and network domains a common language for discovering capabilities, negotiating tasks and coordinating operations.
The proposition is straightforward: a genuinely autonomous network cannot depend on a collection of intelligent but isolated systems.
From individual intelligence to collective autonomy
The emergence of agentic AI has introduced a new dimension to the industry’s longstanding autonomous networking ambitions.
Traditional automation executes predefined processes. Intelligent agents can go further, interpreting objectives, selecting appropriate capabilities and coordinating the steps required to complete a task.
Consider a network experiencing deteriorating service quality. A radio network agent might identify a congestion problem, but resolving it could require changes to transport capacity, service policies or other network resources outside its immediate control.
In a conventional operating environment, those activities may require separate management systems, predefined integrations or human coordination.
In an agent-based environment, the ambition is for the relevant systems to discover one another, exchange information, negotiate responsibilities and execute an agreed response.
That requires more than powerful AI models. It requires interoperability, consistent semantics and clearly defined operational safeguards.
A2A-T builds on the broader Agent-to-Agent protocol, extending it to address the particular requirements of telecommunications networks.
These extensions include mechanisms for identifying operational scenarios, negotiating tasks and results, and applying authorization controls to potentially sensitive network operations.
They also address reliability. Rather than restarting an entire interaction when an individual exchange fails, the protocol supports negotiation and recovery mechanisms intended to reduce unnecessary processing, latency and token consumption.
Security is equally important. Network operations cannot simply be delegated to autonomous software without appropriate restrictions on what those agents are permitted to do.
A2A-T therefore incorporates additional controls for high-risk activities, including authorization and constraint mechanisms.
The objective is not merely to enable agents to communicate, but to establish the conditions under which operators can trust them to collaborate.
Three tracks toward an open ecosystem
OpenAN is being advanced through three complementary industry activities: standardization, open-source development and ecosystem collaboration.
The first is being coordinated through TM Forum, where industry participants have been developing common rules for telecom agent interaction.
The A2A-T standardization effort began with the publication of IG1453 Version 1 in November 2025, followed by IG1453A in January 2026 and IG1453 Version 2.1 at DTW Ignite in Copenhagen in June.
These specifications address three essential requirements: semantic alignment, task negotiation and the management of high-risk operations.
Semantic alignment is particularly significant. Two agents cannot collaborate effectively if they interpret the same operational objective differently. Establishing common business vocabulary and knowledge representations is therefore an essential foundation for interoperability.
The second track involves converting those standards into practical software components through the Linux Foundation’s networking open-source ecosystem.
OpenAN includes an A2A-T software development kit, designed to simplify protocol implementation, together with registry and orchestration components that support agent registration, discovery and task coordination.
The architecture is intended to support private deployment, allowing operators to retain control over sensitive operational interactions rather than routing them through external cloud services.
Huawei and China Mobile have contributed initial open-source components, including the SDK and registration and orchestration capabilities.
Further development is planned around reusable scenario libraries and agent ontologies, potentially reducing the amount of integration work required to introduce new autonomous networking applications.
The third track is industry collaboration, facilitated by the GSMA, bringing together operators, software suppliers and equipment manufacturers to develop and validate practical use cases.
In September 2026, GSMA kicked‑off the “Open Autonomous‑Network Agent Ecosystem Project” with a launch ceremony held at the China National Convention Center in Beijing. Representatives from GSMA, China Academy of Information and Communications Technology (CAICT), China Communications Standards Association (CCSA), China Mobile, China Telecom, China Unicom, Huawei, ZTE, Inspur and AsiaInfo attended the ceremony to witness the official initiation of this industry initiative.
That breadth of participation is important. An interoperability framework controlled by a single supplier would do little to solve the industry’s underlying fragmentation problem.
OpenAN’s ability to deliver on its ambitions will ultimately depend on whether competing suppliers implement compatible capabilities and operators can demonstrate that those capabilities work together in production environments.
Huawei’s RANSpirit provides the commercial foundation
While cross-domain agent collaboration remains an evolving industry objective, Huawei points to its existing RANSpirit deployments as evidence that substantial network autonomy is already commercially achievable within individual domains.
RANSpirit combines a telecommunications foundation model with digital twin technology to support intelligent radio network operations.
According to Huawei, the technology has now been deployed across more than 25 large-scale networks worldwide, covering over 1.2 million cells and assisting more than 2,000 engineers.
Those figures provide a useful indication of the scale at which autonomous network management is already operating.
They also illustrate the next challenge.
Optimizing a radio network through an intelligent agent can deliver operational benefits, but the wider promise of Level 4 autonomy depends on coordinating activities across the entire network.
An agent responsible for radio performance must eventually be able to collaborate with systems managing transport, core infrastructure and other operational domains.
The commercial success of individual intelligent systems therefore creates a new requirement: an architecture capable of connecting them without imposing another layer of proprietary integration.
This is where OpenAN and A2A-T are intended to play a central role.
Huawei says it will continue opening native network capabilities, supporting on-demand orchestration and working with industry partners to develop cross-domain collaboration.
The strategic implication extends beyond radio network management. If operators can combine autonomous capabilities from multiple suppliers, they could progressively automate increasingly complex operational processes without requiring every component to come from the same vendor.
The real test is interoperability at scale
There is an important distinction between demonstrating autonomous capabilities and achieving an autonomous network.
An operator may have highly advanced AI systems managing individual network functions while still relying on conventional integration and human intervention to coordinate activity between them.
The transition to Level 4 autonomy requires those boundaries to become increasingly permeable.
It also introduces difficult questions about accountability. When multiple agents participate in an operational decision, operators need to know which systems have authority, how conflicting objectives are resolved and what happens when an action produces an unexpected result.
A common interaction protocol is a necessary part of the answer, although it cannot, by itself, resolve every question of operational governance, security or commercial interoperability.
The significance of OpenAN is that it attempts to address these challenges collectively rather than leaving every operator and vendor to develop its own approach.
Its combination of common standards, open-source components and industry collaboration offers a potential route from isolated AI applications toward a coordinated operational architecture.
For Huawei, the initiative also represents an opportunity to build on its established single-domain autonomous networking capabilities while participating in a broader multivendor ecosystem.
For operators, the potential prize is considerably larger: the ability to introduce intelligent capabilities from different suppliers while retaining control over how those capabilities interact.
The telecommunications industry has spent decades building networks in which equipment from different manufacturers can exchange traffic according to common standards.
The next challenge is to make the intelligence managing those networks interoperable as well.
OpenAN’s ambition is to establish the common framework that makes that possible.
And if Level 4 autonomous networks are to move from isolated commercial achievements to widespread deployment, that may prove just as important as the intelligence of the agents themselves.