This is The Road to Autonomous Networks. Every month, the signals that show where the network is actually heading.
This month they converge on one point. The agents are ready for the network. The data foundation underneath them is years behind.
Operators want autonomy.
The surveys confirm the ambition is everywhere. But most network data still can't be read by an agent in any reliable way.
In Copenhagen, with Microsoft, we put our answer on the table: a Fabric-native data product that turns multi-vendor network counters into one semantic layer an agent can actually reason on.
Enjoy the reading!
— Vicent R. Soler Ruiz, Co-founder & CEO, Kenmei
GSMA Intelligence: 82% of operators think they see the whole network. 41% actually do.
Two GSMA Intelligence surveys point at the same gap.
82% of operators say they already have RAN-to-core correlation.
Drill one layer down and only 41% have a true end-to-end data architecture across domains.
Meanwhile 65% rate agentic AI as transformational or high-value, and 70% of planned deployments target this year.
The ambition is set for 2026.
The data foundation it runs on hasn't caught up.
Sources: GSMA Intelligence with Radcom, Service Assurance Trends in the AI Era (Jan 2026) ; GSMA Intelligence, Telco AI: State of the Market Q4 2025 (Jan 2026)
Telstra's chief architect: autonomous networks stall until the network can describe itself
Telstra, one of Australia's largest carriers, put a name to what's missing: a knowledge plane.
Before agents can safely run a network, the network needs one accurate, shared description of every part and how it connects, written so software can read it.
The industry calls it ontology.
Skip it and clever agents act on a blurry picture.
Telstra's bet for 2026 is foundation first, autonomy after.
Source: IEEE ComSoc Technology Blog, "Will 2026 be the 'Year of the AI Ontology' for telecoms?", 3 Apr 2026
NTT Docomo: AI agents now run in commercial production across more than a million network devices
In February, NTT Docomo moved AI agents into everyday commercial use to keep its mobile network healthy.
The agents watch more than a million pieces of equipment in real time.
On the hardest faults, the kind that used to need an engineer digging in by hand, response time dropped by more than half.
The distance between roadmap and operating model is closing faster than most plans assumed.
Source: NTT DOCOMO press release, 25 Feb 2026, "DOCOMO Begins Commercial Deployment of Agentic AI System for Network Maintenance"
Inside Kenmei: the data foundation these signals keep describing, shipped at DTW with Microsoft
At DTW Ignite in Copenhagen, with Microsoft, we launched the Network Performance Data Product on Microsoft Fabric.
Counters from every vendor, across every radio access technology, reconciled into one semantic layer the agent reads once.
450+ KPIs you can treat as contracts. Activation in under an hour.
The same governed dataset powers dashboards today and Fabric Data Agents tomorrow.
The signals from this newsletter showed a gap between data ambition and foundation.
We built our product to bridge this gap.
→ More details in our blog post
