This is The Road to Autonomous Networks. Every month, the signals that show where the network is actually heading.
AI agents already run on live networks at dozens of operators, and none of those operators has taken the human out of the loop. When TM Forum asked, 72% said they trust their AI while only 14% could prove it.
We think the reason is simple. An engineer steps back only when they can check what the agent saw and why it acted. In most networks, that information is still scattered across separate vendor tools.
So the next step sits underneath the agents, in the data they read. NGMN is now asking the industry to fix exactly that.
— Vicent R. Soler Ruiz, CEO @ Kenmei
TeckNexus: 50 operators use AI agents on live networks. All of them still keep a human in charge.
TeckNexus looked at AI agent deployments at 50 operators and kept only the ones it could verify. In 54% of them, the agent suggests an action and an engineer carries it out. In 42%, the agent handles routine work on its own.
Only one operator has been verified going further than that.
In a 2025 white paper, Appledore Research projected up to $60 billion in operating savings from agentic AI by 2030. The agents are there, and so is the business case. Engineers stay in the loop because they still need to check what the agent does.
Sources: AI Agents in Telecom: What 50 Global Operators Are Actually Doing, TeckNexus · Role of Agentic AI in Telecommunications, Appledore Research
TM Forum: 72% of operators trust their AI. 14% can prove it works.
Most operators believe their AI works. Few can show what data it used or why it made a given decision.
Analysys Mason found the same gap in a study of 50 large operators. About half are behind on putting AI into daily operations. The main reason they give is network data that is hard to access and hard to trust.
In August, NGMN asked vendors and operators to agree on one common way to describe the network. Agents would then read the same data, whoever built the equipment.
Sources: Trust and assurance: the key to scaling AI deployments, TM Forum and IBM · Telco AI Readiness Index: 2026 results, Analysys Mason · Agentic AI for Autonomous Mobile Networks, NGMN
Moldtelecom and Kenmei: handover success steady at 99.9% on 4G and 5G C-band
This summer we worked with Moldtelecom's engineers on their 4G network and their new 5G C-band layer. Handover success now holds at 99.9%. The weakest went from 14 successful handovers out of 100 to 91.
We also analysed several months of download data. It showed that the radio network can carry a lot more traffic than it does today.
Source: Moldtelecom and Kenmei turn new C-band spectrum into stronger 4G and 5G performance, Kenmei
----
The Road to Autonomous Networks is a monthly newsletter for the people who run mobile networks:
CTOs, heads of network, RAN and operations leaders.
Each edition picks the market signals that show where network automation is really going.
