Verified, then logged.
Sun 22 Nov 2026 · 00:34 KST / 15:34 UTC한국어 | ENSearch /

Breaking02 / IT

U Mobile runs its AiDaH helpdesk on Amazon Bedrock AgentCore

Seen past the blurred back of a headset-wearing support agent, rows of call center desks stretch toward windows overlooking a tropical city and a rooftop cellular antenna. AI-generated image

AWS on 5 Nov 2026 published the design of AiDaH, the helpdesk that 50 call-centre and branch users at Malaysian telecoms operator U Mobile use to diagnose problems. U Mobile built it on Amazon Bedrock AgentCore, AWS's platform for building and running AI agents.

Requests take the lowest-cost route first. AWS said about 70% end in a keyword match with no model call. A lightweight model classifies 20%, and the full orchestrator handles only 10%.

Eight agents pass work to one another over the Agent2Agent protocol, according to AWS. AgentCore Gateway connects U Mobile's internal systems through Model Context Protocol (MCP), a standard interface that gives AI models access to tools. A new business domain adds a gateway and its MCP tool definitions. The stages before and after that step stay the same.

Hosting. AiDaH runs in the AWS Singapore region, AWS said. All computing takes place in private subnets of a virtual private cloud (VPC), an isolated network inside a public cloud. Internal systems connect only through a VPN encrypted with IPsec, a protocol suite for securing network traffic.

Actions that change a subscriber's account pass through a person. AWS said AiDaH drafts the proposed fix, which runs once a human support agent confirms it. Sensitive subscriber data is masked at the gateway before it reaches a model.

U Mobile began work on the concept in late 2025 with two pilot agents, according to AWS. A pilot in the first quarter of 2026 started with four agents and 20 scenarios. The system had eight agents when it went live in the third quarter. It now covers more than 150 scenarios across six business divisions.

U Mobile's own measurements show trouble tickets falling by up to 84.5% in key categories, the AWS post said. The passage reporting that figure names neither the categories nor the measurement period.