AWS re:Invent 2025 - Making Level 4 Autonomous Networks a reality with British Telecom (IND205)

Transforming Telecom Networks with Agentic AI: British Telecom's Journey

Introduction

• Telecommunications companies face significant challenges in operating and managing their networks, including high costs, the need to fulfill the promise of 5G, and underutilized data.
• The presentation outlines British Telecom's (BT) vision and journey towards building autonomous, AI-powered networks using AWS's agentic AI strategy.

Telecom Network Lifecycle and Autonomy Levels

• Telecom network lifecycle consists of planning, deployment, services fulfillment, and operations.
• The industry has defined five levels of autonomy, with Level 4 being the focus of this work.
• Level 4 autonomy involves closed-loop, intent-driven networks with end-to-end service automation.

AWS Agentic AI Strategy

• The agentic AI stack includes infrastructure (e.g., AWS Trainium, Inferentia), AI and agent development services (e.g., Amazon Bedrock, Agent Core), and applications.
• Key components leveraged in this work include Agent Core, which provides primitives for building scalable, enterprise-grade agentic systems.

BT's Vision and Challenges

• BT's vision is to build a trusted, converged network (fiber, 5G standalone, Wi-Fi 7) that is AI-powered, intent-driven, and autonomous.
• Challenges include managing a vast network (20,000 macro sites, 4,000 KPIs per site), data silos, outdated processes, and a need to transition from network engineering to software engineering.
• BT's "DDOPS" (Data-Driven Operations) initiative aims to fix data, create a unified data view, and automate network operations.

Solution Architecture

1. Network Data Sources: Ingesting and curating data from various network elements (performance counters, alarms, topology, etc.).
2. AI-Powered Data Product Lifecycle: Using agentic data engineering, feature engineering, and a semantic data layer to create reusable data products.
3. Data-Driven AI for Network Applications: Deploying hyper-optimized ML models for anomaly detection, root cause analysis, and service impact analysis.
4. Agentic AI Applications: Building agents for root cause analysis, service impact analysis, troubleshooting, and RAN optimization.

Use Cases

1. Multivariate Anomaly Detection: Transitioning from univariate to multivariate anomaly detection using techniques like temporal pattern clustering and transformer models.
2. Root Cause Analysis and Service Impact: Using domain-specific community agents and inter-community collaboration to identify root causes and service impacts.

Benefits and Next Steps

• Key benefits include cost reduction, improved change efficiency, consistent data, and better customer impact identification.
• Next steps include expanding the solution to cover the entire network, improving coverage analysis and optimization, and enabling dynamic network slicing.

Related Sessions at re:Invent

1. Agent AI for Autonomous Networks
2. Domain-Specific Fine-Tuning
3. Using the Q Developer CLI for Telecom Infrastructure
4. AI Agents Framework for RAN Network Optimization (Workshop)