Telecom AI agents need a live, structured map of the network, argues João Miguel Antunes, Celfocus’s head of autonomous networks. His point is practical: an agent cannot reliably identify the cause of an outage if it cannot tell which equipment and services depend on one another.
Watch Desk analysis
What happened
In a partner-content article published by Mobile Europe on 27 October, Antunes challenges both AI sceptics and believers who expect a sufficiently large model to make sense of raw network data. He argues that the missing ingredient is context: an up-to-date digital twin representing network resources and their relationships.
His example is a power failure at one site. It could trigger an environmental alarm, alarms from nearby radio equipment and signal loss in dependent transmission nodes. Treating those as separate events risks opening several incident records for one underlying problem. Antunes says tracing the network topology could connect them to a single incident and root cause.
Why it matters
The argument shifts attention from model size to the unglamorous work of making operational data accurate, connected and current. Antunes estimates that roughly four-fifths of the effort is data engineering, including reconciling different systems’ names for the same physical equipment.
For network operators, the stakes are more than tidier dashboards. An agent’s proposed diagnosis is only useful if engineers can see the causal path behind it. Antunes argues that systems should first propose actions for a person to approve, with each decision recorded, rather than leap straight to autonomous operation.
Our read
This is a useful corrective to the idea that a bigger model can compensate for a poor map. In the article, the agent’s role is to investigate by querying network context, testing a causal explanation and showing its reasoning. The digital twin is not a decorative 3D visualisation; it is the structured account of what exists and what depends on what.
The most useful questions for an operator considering an agent are straightforward: what should it know about the network, where does that information come from, how often is it updated, and can it show the causal path it followed? If those answers are vague, the demo is getting ahead of the groundwork.
What to watch
- Whether operators publish results using incident-detection time, identified root causes and field interventions, rather than automation percentages alone.
- How network digital twins are kept accurate as equipment and dependencies change.
- Whether agents can show engineers a reviewable causal trail before any action is taken.
Discussion spark: Should network operators require an agent to show a verifiable causal path before acting, even if that makes automation slower?
Sources and evidence
- Source update (27 October 2026, 05:47 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.