Truly autonomous networks and operations Based on TM Forum’s Open Digital Architecture,28 CSPs have significantly transformed to embrace open architectures, including disaggregation of hardware and software layers to enable virtual networks that can be managed, programmed and updated remotely. This reduces reliance on proprietary hardware while opening opportunities for advanced network automation powered by genAI. Autonomous networks deliver a “Zero X” experience – zero wait, zero touch, zero trouble – through AI-enabled self-management, optimization, configuration and security.29 TM Forum’s AI Maturity Model identifies six levels (0-5) of network automation,30 with the highest level offering “closed- loop automation capabilities across multiple services, domains (including partner domains) and the entire life cycle via cognitive self-adaptation”.31 This enables use cases such as proactive fault identification, resolution and management and zero-touch operations. The potential value of Zero X has been estimated as $794 million annually for an average CSP .32 China Mobile exemplifies this ambition, aiming for “highly autonomous network” (level 4) automation by 2025,33 where manual intervention is reserved for strategic decisions and process oversight. Fully automated networks require an open architecture across vendors, including network management interfaces that are currently specific to original equipment manufacturers (OEMs). Removing these barriers will enable CSPs to harness AI’s full potential, driving unprecedented levels of automation, efficiency and customer experience across the telecom landscape. Autonomous tech stack reinvention Traditional IT management and programming methods rely on centralized points (such as an application) to access and process data, using coded business logic and producing consistent outcomes for given inputs. However, technology has now advanced to a point where it can enable non-deterministic solutions that disaggregate the model (data) from the controller (decision-making capabilities), eliminating the need for a single interface to connect the two. This transformation requires agentic AI architecture, including foundation agent orchestration, continuous training and feedback. These components can build an intelligence wrapper that acts as a dynamic control plane capable of understanding context, making decisions and improving over time. However, guardrails are required for responsible use. The implications are transformational. The rise of an AI agent-operated “super platform” is anticipated, supplanting the current coded “super app”. This will free businesses from constantly managing technical debt and allow them to focus on experimentation and innovation. Future reference architectures will depend on decisions around the location of the control plane for specific scenarios – whether it should be closer to data sources at the edge, or to cognitive, connected and creative decision engines. Artificial Intelligence in Telecommunications 18

Artificial Intelligence in Telecommunications 2025