Microsoft's forward deployed engineers embed at KT to build network-diagnosis AI agents

A Microsoft Source feature published November 11 details how forward deployed engineers from Microsoft Frontier Company worked inside KT, South Korea's major telco, to build AIO-Net, an agentic AI platform for network operations. The engineers sat in KT's offices alongside its development team, writing code and jointly building the processes for collecting, applying and rolling back operator feedback.

The agents, built on multiple models in Microsoft Foundry on Azure, were trained on KT's operating procedures and by its veteran network experts, called Meisters. KT says detection time dropped from over 10 minutes to 2-3 minutes, and agents produce an incident report in about five minutes, roughly half the manual time. Agents are limited to read-only commands on an allow list, enforced in the application rather than the prompt, and decisions stay with human operators. KT plans to expand from 70 to 280 units of critical equipment.

Why it matters for FDEs: KT's own team framed the engagement as co-development rather than outsourcing, which is the core FDE pitch. The project also shows the practical work FDEs own in regulated, high-stakes settings: capturing expert knowledge, designing feedback loops, and hard-coding guardrails so agents cannot take unsafe actions.