AWS re:Invent 2025 - Ericsson Innovation: Optimizing Mobile Networks & Unified Development with AWS

Ericsson's Transformation with AI on AWS

Optimizing Mobile Networks with AI

• Ericsson is leveraging a combination of traditional ML and agentic AI solutions to optimize mobile network performance for their CSP customers
• Key components of their approach:
  • Cognitive loop framework for network operations, with 5 stages: measurement, assurance, proposal, evaluation, and execution
  • Specialized agents across the cognitive loop, including:
    • Measurement agents to capture network data
    • Classifiers and root cause analyzers for assurance
    • AI recommender systems and agents for proposals
    • Reinforcement learning "cell shaper" agents to optimize coverage, capacity, and quality
  • Integrating these agents with AWS services like Amazon SageMaker, Amazon Bedrock, and others to enable scalable, observable, and trustworthy AI-powered network operations

Empowering Engineers with AI

• Ericsson's "System Comprehension Lab" aims to address the challenge of "comprehension bound" product development, where the complexity of mobile networks exceeds human capacity for understanding
• Key objectives:
  • Consolidate Ericsson's institutional knowledge and data assets to better design mobile network products
  • Layer real-world network data and insights on top of this foundation
  • Empower engineers to build better products through AI-powered tools and agents
• Technical approach:
  • Decouple front-end tools from Ericsson's internal data and systems using an MCP gateway
  • Build specialized AI agents and models, including a custom large language model tailored to Ericsson's proprietary frameworks
  • Integrate agents with downstream engineering tools and enable access where engineers work (e.g. IDEs, custom front-ends)
  • Ensure observability, scalability, and trustworthiness through capabilities like agent tracing, identity management, and memory

Key Takeaways

• Ericsson is transforming their mobile network offerings and internal engineering processes by deeply integrating AI, enabled by their partnership with AWS
• Their approach focuses on building trustworthy, scalable, and observable AI systems that can handle the immense complexity of modern mobile networks
• Specific results include:
  • 30% improvement in network coverage and capacity through reinforcement learning "cell shaper" agents
  • Accelerated feature delivery and developer productivity through AI-powered engineering tools and agents
• The collaboration between Ericsson and AWS showcases how enterprises can leverage the breadth of AWS services to tackle complex, domain-specific AI challenges