Malaysia’s U Mobile has expanded an agentic operations platform built on Amazon Bedrock AgentCore to more than 150 documented scenarios across six business divisions. The system, called AiDaH, coordinates diagnostics across the telecommunications company’s hybrid systems and proposes corrective actions while keeping a person at the decision point before a live subscriber account is changed.

AWS published the customer architecture on 4 November. U Mobile’s internal measurements show trouble-ticket volumes falling by as much as 84.5 per cent in selected categories. The figure is not a general benchmark, but it provides a useful production example of agentic AI moving beyond response drafting into multi-system investigation.

From assistant to operational coordinator

A routine telco support case can require several systems. An employee may need to inspect interface logs, check subscriber eligibility through business-support APIs, consult policy material and decide whether a backend ticket is necessary. Traditional automation can help write the final response, but the investigation remains a sequence of manual lookups and context switches.

AiDaH is designed to coordinate that work. It can check configuration and eligibility, identify a likely point of failure and assemble a remediation plan. The support worker reviews the plan and confirms actions that will affect the live account. That division preserves a human checkpoint while allowing the agent to perform the slow evidence-gathering phase first.

In one example shared by AWS, a customer repeatedly failed to purchase an add-on. AiDaH traced the issue to a pending-order error and found more than 30 attempts across several channels over three days. It presented the root cause and proposed remediation together, replacing a manual investigation across separate systems.

A staged path to production

U Mobile began conceptual work in late 2025 with two pilot agents and Amazon Bedrock Knowledge Bases. An internal helpdesk pilot in the first quarter of 2026 expanded to four agents covering 20 scenarios. The platform then added three-tier routing and reached six agents by the middle of the year.

In the third quarter, AiDaH went live to 50 second-level call-centre and branch users with eight agents. It now covers more than 150 scenarios. That progression is significant because it shows a deliberate increase in scope: a small knowledge-backed pilot became an operational platform only after routing, security and integration layers matured.

Six layers behind one interface

Employees access AiDaH through Microsoft Teams and a web portal, reducing the need to learn another application. A REST interface also supports programmatic triggers. Behind that entry point, the platform uses a six-layer architecture on Amazon Bedrock AgentCore in AWS’s Asia Pacific region in Singapore.

The orchestration layer begins with identity checks, role-based access controls and guardrails. It then chooses a cost-appropriate route. AWS says about 70 per cent of requests are handled by deterministic keyword matching without a model invocation, around 20 per cent use a lightweight classifier and only 10 per cent require the full orchestrator. The design aims to reserve more expensive reasoning for cases that need it.

Managed AgentCore runtime reduces the infrastructure U Mobile must operate, while built-in memory maintains relevant conversational context between turns. Integrations connect the agents to internal systems, and the broader architecture adds knowledge, observability and security controls around the reasoning layer.

Reach, depth and shared controls

U Mobile describes its strategy through three pillars. Reach means agents can coordinate with backend systems in real time. Depth refers to multi-step root-cause analysis rather than a single retrieval or generated answer. Community means divisions use a shared, multi-tenant framework with role-based access and guardrails instead of building isolated assistants.

That shared platform is likely central to scaling. Every new operational domain still needs careful integration, permissions and testing, but teams do not have to rebuild routing, runtime and security foundations for each use case. Common controls can also make governance more consistent across divisions.

Why the human checkpoint matters

Telecommunications systems affect billing, connectivity and customer identity. A technically plausible remediation can still create harm if it is applied to the wrong account or under the wrong policy. AiDaH’s pattern keeps the human decision at the moment a proposed diagnosis becomes a live change.

That does not eliminate risk. The company still needs reliable identity propagation, least-privilege connectors, complete audit trails and tests for cases where source systems disagree. Measurements should distinguish between tickets avoided, tickets deferred and issues that reappear after an automated recommendation.

The reported reduction is nevertheless a notable signal. Agentic systems often look convincing in demonstrations because they can narrate a plan. U Mobile’s deployment is more interesting because it spans real operational scenarios, uses tiered routing to control cost and keeps authority separate from investigation.

A practical template for enterprise agents

The broader lesson is that production agent value can come from orchestration rather than a single model breakthrough. AiDaH combines deterministic routing, lightweight classification, full reasoning, memory, enterprise interfaces and human confirmation. Each layer handles a different part of the reliability and cost problem.

For organisations considering similar systems, a staged rollout remains the sensible path: begin with bounded scenarios, measure investigation quality, add integrations gradually and preserve human control over consequential changes. U Mobile’s journey suggests that the architecture can expand substantially once those foundations are in place.