Crossing the ‘network autonomy chasm’ requires changing the business itself

The telecom industry has spent decades refining automation to improve efficiency and reduce operational costs. Yet, despite this sustained effort, most operators remain far from achieving true network autonomy. The issue is not a lack of technology or ambition. It is a matter of approach.

What the industry is facing today is not a gradual evolution but a discontinuity. There is a fundamental gap between the automation that has defined the past and the autonomy required for the future. Appledore Research calls this a “network autonomy chasm”, and it cannot be crossed by extending existing models.

The limits of today’s automation

Automation has largely meant encoding predefined workflows. Even in advanced deployments, automation systems execute predefined event-driven sequences, with humans responsible for exception handling, policy interpretation and cross-domain coordination. These systems are efficient but remain tightly coupled to static process logic.

True autonomy begins where this model breaks. Autonomous networks must be able to make independent decisions about network and service state and either fix the issues locally or ask for additional help. This implies continuous, flexible decision-making based on potentially conflicting requirements and under changing conditions, not just execution of predefined rigid tasks.

Mechanized workflows are inherently context-blind: They perform well under known conditions but struggle with change such as new service models, vendor integrations or shifting traffic patterns. In modern networks where services span multiple administrative domains, rely on disaggregated infrastructure and are deployed across hybrid cloud environments, these limitations become systemic. A single inconsistency in state between inventory, policy and real-time telemetry can cascade across services, forcing manual reconciliation.

Rethinking the role of orchestration

Closing this gap requires rethinking the role of orchestration. Traditional orchestration platforms are primarily workflow engines. They coordinate tasks across systems but depend on predefined execution paths and tightly integrated adapters. They are not designed to manage continuous change or to reason about system-wide state in real time.

In autonomous networks, orchestration must evolve into a model-driven control plane. This implies a shift toward declarative intent models, where the desired state is defined independently of implementation and continuously reconciled against actual state. Instead of executing workflows, the system maintains convergence between intent and reality.

TM Forum’s Open Digital Architecture (ODA) reflects this by promoting decomposed operational capabilities, standardized interfaces and intent-based interaction models that decouple service logic from infrastructure implementation. This shifts control loops from isolated domain-specific mechanisms to composable hierarchical loops across service, resource and infrastructure layers, coordinating across lifecycle stages while resolving conflicts under shared policy constraints.

Agentic AI needs governance

It is in this context that agentic AI becomes relevant. Unlike traditional AI and machine learning models that provide predictions or recommendations, agentic systems can actively participate in control loops. They can interpret high-level intent, generate actions, evaluate outcomes and adapt strategies over time. While traditional automation handles tasks, agentic AI provides the cognitive ability to manage the system.

However, this introduces a critical requirement: determinism at the system level. Autonomous networks cannot rely on opaque or non-reproducible behaviors. Every action must be traceable, verifiable and aligned with policy constraints.

This is why model-driven orchestration and agentic AI must be tightly integrated. AI provides the adaptability to navigate change, while the orchestration layer enforces the consistency and control necessary for carrier-grade reliability.

Autonomy is not about eliminating complexity but about governing it through software abstractions. Everything from service definitions to network functions must be model-based, versioned, validated and continuously reconciled. This is the only way to manage dynamic environments where change is constant and often unpredictable.

Transforming the business

This shift is as much organizational as it is technological. Crossing the chasm requires breaking the long-standing silos between NetOps & DevOps and aligning with ODA principles. Autonomy isn’t just a tool you buy; it’s a capability you build by merging software engineering principles with network operations. This necessitates a unified operating model in which service design, orchestration and assurance are no longer separate functions but rather part of a continuous lifecycle.

It also requires redefining metrics. Traditional KPIs assume that failures are inevitable. In an autonomous system, the objective shifts toward preventing issues through predictive and adaptive control. Success is measured not by how quickly problems are fixed but by how rarely they occur.

The transition to autonomous networks must be led from the top. It is not simply an operational improvement, but a transformation of the business itself.

Making the leap

The telecom industry has reached a point where incremental progress is no longer sufficient. Continuing to extend task-based automation risks increasing complexity without achieving system-wide autonomy.

Crossing the autonomy chasm is difficult, but necessary. The operators that succeed will be those that adopt a fundamentally different approach – one that combines model-driven orchestration, composable control loops and agentic AI into a coherent system.

By moving beyond the limitations of static workflows, we can finally align our networks with the dynamic needs of the digital era.