Agentic AI for 5G networks: Unryo says it has ended the panic call between radio and Kubernetes

By Unryo · September 25, 2026.

Unryo, a Montreal-based observability company, has unveiled an agentic AI layer —AI systems that do not merely answer questions but run a sequence of investigative steps on their own— capable of correlating radio access network (RAN, the part of a mobile network that connects handsets to the antennas) failures with the state of the Kubernetes containers that today run much of the software on those antennas. According to the company's own press release, the promise is to replace the manual reconstruction of a fault with a single automated explanation, with every step traceable back to the source data.

The problem it claims to solve is a concrete one. Modern mobile networks are cloud-native: much of their network functions no longer run on dedicated hardware but as pods —the deployment units of Kubernetes, the de facto standard platform for orchestrating containers— inside clusters managed by the operator itself. When a cell degrades, the cause may lie in the radio or in the container infrastructure underpinning it, and each domain is currently monitored with different tools, forcing engineers to reconcile two consoles in the middle of an incident.

According to Unryo, its agent reads the alarm, walks the dependency chain from the affected radio object through to the network function, the pod and the node running it, and cross-references radio counters and service-interface signals (the signaling traffic between 5G network functions) with container-platform events at the moment of failure. The example the company offers: if a cell's throughput drops with no change in radio load or signal quality, the agent traces the cause to a Kubernetes node that began dropping packets a minute earlier, and classifies the subsequent pod restart as a consequence, not a cause.

As industry context, the company itself stresses that its reasoning is grounded rather than generative, and that the underlying correlation is deterministic: every conclusion rests on what it calls the Topology Data Fabric, a model built from vendor management platforms —it cites Nokia MantaRay and Ericsson ENM—, the Kubernetes API and telemetry protocols such as SNMP or gNMI. It is a direct answer to the objection that weighs most heavily on agentic AI in critical infrastructure: hallucination. Unryo also offers deployment on the operator's premises or in its private cloud, and presents the announcement as part of a broader platform, having already announced open APIs via TM Forum and the automatic assignment of incidents to the responsible team.

The announcement should be read with the caution any corporate press release demands. There are no figures on customers, funding or revenue, and no comparisons against rival AIOps platforms (automation of network and IT operations) such as Moogsoft, ServiceNow or Nokia's and Ericsson's own suites. Every claim about accuracy and reliability comes from Unryo itself and has not been validated by a third party: this is, strictly speaking, a product's calling card, not an independent study.

Our take: beyond the specific announcement, the case is a good illustration of where agentic AI is landing with less noise and more immediate usefulness. Not in the conversational assistant, but in automating the reconciliation work human engineers do today between tools that do not talk to each other. It is exactly the kind of routine, low-judgment task —cross-checking one console against another— that in our sector-by-sector review of AI and employment we flagged as the most exposed within technology: the winners are those who design the architecture and govern the integration of these agents; the loser is the manual correlation work that used to bridge the silos.

For telecom operators, which underpin increasingly critical infrastructure —everything from hospitals to logistics depends on mobile connectivity—, reliably automating fault diagnosis is not science fiction: it is the kind of incremental, boring, verifiable improvement that, multiplied across hundreds of similar cases in other sectors, sustains the underlying thesis that AI multiplies operational efficiency long before it brings us anywhere near curing diseases or extending life. The road to abundance is also built, brick by brick, in the engine room of the mobile network.