Microsoft and South Korean telecoms operator KT have developed an AI system that investigates network problems, helping engineers diagnose faults faster while leaving decisions and actions to people. The useful detail is the boundary: the agents can inspect equipment, but cannot issue commands to change it.
Microsoft AI Watch analysis
What happened
KT’s AIO-Net platform analyses live network data and draws on the company’s operating procedures and the experience of its network specialists, known as Meisters. Microsoft says the system has been in use since August. KT reports that it detects issues in two to three minutes, down from more than ten, and that its agents investigate an incident and produce a report in an average of five minutes, about half the time a manual investigation took.
The agents can flag incidents, inspect equipment using read-only commands on an allow list, and suggest a solution. They cannot take action. Microsoft says that restriction is enforced in the application, not merely in the prompt. KT plans to expand use from 70 units of critical equipment to 280, then into access networks closer to subscribers.
Why it matters
This is a concrete example of AI being used in operational infrastructure where a bad instruction could cause a real outage. The system is designed to shorten investigation, while keeping the response in engineers’ hands. That division is more consequential than the label “agentic”, which is doing plenty of work these days.
Our read
The notable idea is not simply that AI can sift through network alerts. KT says its specialists train and review the system, and its agents are technically barred from changing equipment. That makes this a more useful deployment to watch than an automation demo with the safety boundary left to a polite request.
What to watch
- Whether KT’s reported diagnosis times hold as deployment expands.
- How AIO-Net performs in access networks closer to subscribers.
- Whether the read-only restriction remains in place as the system gains new capabilities.
Discussion spark: For AI in critical infrastructure, is a hard technical bar on taking action enough, or should the system also face independent testing before it expands?
Sources and evidence
- KT ‘Meisters’ and Microsoft engineers build AI agents to spot network issues before customers do – Microsoft Source (11 November 2026, 16:16 UTC)
not affiliated with or endorsed by Microsoft