AI in telecom isn’t just an upgrade anymore – it’s a full-scale reshaping of how networks think, talk, and make money. And this is the part most people miss: the real game-changer is not a single AI feature, but an entire “agent network” where devices, services, and network functions all act like intelligent collaborators.
Event overview
At the 10th 5G Core Network Summit in Bangkok, Thailand, George Gao, President of Huawei’s Cloud Core Network Product Line, urged telecom operators to seize the new wave of AI-driven opportunities. His core message centered on building an agent-based communication network, where terminals and services work together through intelligent agents embedded in the network. During the summit, Huawei also launched its “Striding Towards the Intelligent World White Paper 2025” on cloud core networks, outlining seven key development trends around smart terminals, evolving service scenarios, and AI-powered core networks to help drive broader industry collaboration and innovation.
Why agent networks matter
Gao stressed that as user devices and service applications become smarter and gain their own agent capabilities, today’s networks must evolve to connect these intelligent “islands” into a unified ecosystem. Without this shift, operators risk ending up with fragmented experiences where apps, devices, and network functions cannot cooperate effectively, even though each is intelligent on its own.
Step 1: Making the network agent-based
The first step he proposed is to “agentize” the network using several types of agents: intent entry agents, network service agents, network capability agents, and network O&M (operations and maintenance) agents. In simple terms, this means turning the network into a system where software agents can understand user intent, call network capabilities, and manage operations autonomously, instead of relying solely on traditional static configurations.
This agent-based network supports an agent-to-agent (A2A) protocol and a model context protocol (MCP), which together define how agents talk to each other and share context. With these protocols, different agents—on devices, in the core network, and in services—can communicate directly, enabling much more intelligent coordination between terminals and applications.
The intent entry component plays a key role here by interacting with agents on the terminal side, so device-based agents can trigger network agents precisely when they are needed. For example, a user’s AI assistant on their phone could automatically request optimized video quality or low latency from the network without the user manually changing any settings.
Step 2: Building an agent communication network across ecosystems
The second step is to extend this concept into a full agent communication network that spans multiple ecosystems, not just one operator’s environment. To do this, A2A session management capabilities can be layered on top of traditional IP sessions, creating a structure where agent interactions are managed as first-class sessions in their own right.
Authentication also needs to evolve: instead of relying only on SIM card–based identification, the system can expand to digital identity authentication and authorization. This opens the door for agents linked to people, enterprises, or services (not just SIMs) to participate securely in the network, which some might see as both powerful and controversial in terms of privacy and control.
Four strategic actions for AI core networks
Gao outlined four major strategic actions to accelerate the evolution of operator networks into true AI core networks: intelligent network services, productized network capabilities, autonomous O&M, and diversified telco cloud computing power. Together, these steps are designed to shift operators from being simple bandwidth providers to becoming intelligent service platforms in the mobile AI era.
Intelligent network services
By leveraging AI agents, operators can deploy a wide range of intelligent services directly on the core network layer. One standout example is “AI Calling,” where AI capabilities are integrated with calling services to create an AI call assistant that can understand user intent, recognize emotions, and interpret contextual information in real time.
This assistant can help deliver call experiences that feel more personalized and efficient—such as summarizing calls, handling routine transactions, or guiding users while they talk. Built on the data channel (DC), the system can support interactive calling, allowing tasks to be completed in a closed loop during the call itself—for instance, confirming payments or booking appointments without leaving the call.
In this new service model, something as familiar as the traditional dial pad becomes a gateway into AI-powered services. That raises an interesting question: will users be comfortable with their everyday calls being mediated and analyzed by AI agents, or will that feel like too much automation in a very personal communication channel?
Network capability productization
Another critical move is to transform raw network capabilities into structured, sellable products so operators can move beyond charging mainly for data traffic. Instead, they can offer premium, differentiated service experiences with clearly defined performance characteristics for different user groups or applications.
Huawei defines five essential elements for successfully monetizing these experiences: experiences that can be clearly defined, guaranteed quality levels, offerings that are easy to market, a perceptible sense of exclusivity for users, and reliable service provisioning. Technologies such as media relay can significantly boost HD video usage and improve experiences for third-party services, effectively reshaping the traditional operator business model into one that sells quality and experience rather than just gigabytes.
Here’s where it gets controversial: if operators start segmenting and selling “better” experiences, could that blur the line with net neutrality principles in some markets? Some might argue it’s a natural evolution toward premium services, while others may see it as creating tiers of access and quality.
Network O&M autonomy
Gao also proposed using “digital employees” to push network operations and maintenance toward higher levels of autonomy. These digital employees—essentially AI-driven agents—can help achieve high O&M efficiency while maintaining network security and stability.
The AI core network brings together key technologies such as agents, digital twins, and an industry-first mixture-of-models (MoM) architecture. This MoM approach coordinates fast inference models and deep reasoning models so that routine tasks can be handled quickly, while complex scenarios trigger deeper, more thoughtful processing.
By fusing these capabilities, the AI core network can reach single-domain autonomy, meaning certain network domains can monitor and manage themselves with minimal human intervention. This accelerates the journey toward level-4 autonomous networks, where the network can predict, diagnose, and adjust itself almost entirely on its own—something that could be seen as both exciting and unsettling for traditional network engineers.
Diversified telco cloud computing power
The final strategic action is to integrate AI computing power directly into telecom infrastructure. Through resource pooling and unified orchestration and scheduling, operators can build a single cloud that supports both general-purpose computing and AI-focused computing.
Such a cloud can be designed to accommodate hundreds of billions of agents and huge volumes of service traffic as mobile AI use explodes. In practice, this means the network becomes a massive, shared AI computing substrate that can power everything from on-device assistants to industry-specific agent applications.
And this is the part most people miss: if operators own and operate this AI-centric computing fabric at scale, they could become foundational players in the AI economy, not just connectivity providers. But that also raises tough questions about concentration of power and how much control a handful of large operators and vendors might have over AI infrastructure.
Industry impact and ecosystem shift
In closing, Gao framed AI Core not merely as a technical enhancement, but as a fundamental reconstruction of the telecom industry’s ecosystem. Instead of incremental upgrades, AI Core represents a new architecture where agents, intelligence, and autonomous control become embedded into every layer of the network.
Huawei states that it is committed to ongoing innovation to help operators unlock fresh revenue streams and secure stronger market positions. The underlying idea is that operators who embrace AI core networks early could differentiate themselves and shape new business models, while late adopters risk being left behind in an AI-first world.
Your turn: what do you think?
Some people may see this vision as the natural next step for telecom: smarter networks, richer services, and better monetization opportunities for operators. Others may worry about issues like privacy, AI-mediated communications, and increased dependence on a few large technology providers.
So here’s the big question: Do you believe AI agent networks and AI core architectures will genuinely benefit end users and the wider industry, or will they concentrate power and control in ways that are hard to reverse? Do you agree with the idea of turning network capabilities into premium, tiered products, or do you think that could cross a line in how open and fair networks should be? Share where you stand—do you mostly agree, mostly disagree, or sit somewhere in the middle, and why?
