What happens when AI moves from beside the network to inside it?
At Telecom Infra Project’s recent Fyuz event, consensus was clear that AI-driven network autonomy isn’t a future roadmap item – it’s becoming operational reality.
Service providers are deploying AI workloads that demand autonomous network decisions, not just automated tasks. While the gap between current network capabilities and what AI-native applications require has become obvious, Fyuz 2025 revealed it’s closing faster than expected.
AI can no longer sit beside the network - it must live within it. The shift centers on agentic AI, models that can reason, plan, and act with autonomy rather than simply executing predefined tasks. These agents are becoming essential for network planning, real-time optimization, anomaly detection, and multi-vendor data ingestion. The consensus is that network capabilities are evolving from automating responses to optimizing for outcomes - autonomously.
But autonomy requires openness as its foundation – one that's moving well beyond RAN - across open LAN, transport, optical domains, and network API ecosystems. This means simplified management, open interfaces, and trusted automation that scales from labs to national networks. Equally critical is that programmability at every layer is becoming the foundation for networks that can adapt in real-time.
With this shift comes a familiar barrier: AI models struggle to generalize across heterogeneous, multi-country deployments, with the complexity of diverse network environments, vendor equipment variations, and regional requirements creating significant obstacles. That's where smaller, intent-driven foundation models trained on operator-specific network data are emerging as the answer, delivering the accuracy autonomous networks need while preserving privacy and sovereignty.
This autonomy matters because of what's coming next. XR applications, sensing capabilities, robotics coordination, edge inferencing - and the early building blocks of 6G - aren't distant possibilities. Service providers are building for these demands now, but supporting them requires more than automation can deliver. An operations focus therefore becomes key - specifically, how to move from autonomous network concepts to actual deployment. The path forward centers on modular architectures aligned with O-RAN standards, enabling operators to plug in AI/ML models and third-party applications without creating proprietary silos. For service providers, the benefits lie in gaining control points for autonomous decision-making without sacrificing flexibility.
Through our Amdocs Cognitive RAN platform, Amdocs is deploying these capabilities for operators - from tools like SON Code Expert that convert business intention into network parameters, engineering specifications into code and migrate legacy automations, to self-service capabilities that let field engineers build simple, repeatable workflows themselves. The result is expanded automation volume without central development bottlenecks. Beyond workflow automation, network planning is evolving with advanced radio digital twins. Amdocs deploys beam-level and floor-level simulations within operator workflows, letting engineers validate coverage, plan sites, and test scenarios before deployment - forming the foundation for automation that can run scenarios and optimize autonomously.
Fyuz 2025 showed that autonomous networks is moving from roadmap vision to operational deployment, enabled by open integration and open intelligence working together. The question now isn't whether this shift happens - it's which service providers move fastest to deploy the infrastructure that makes it work.
