By: Aaron Boasman-Patel

For more than a decade, the telecom industry has spoken about autonomous networks as a destination, something mapped neatly across maturity curves, debated at conferences, and positioned as just over the horizon. What we are now seeing across the industry is not the arrival of fully autonomous networks end to end, but the steady emergence of autonomy as an operational capability, already taking shape in live environments, already solving real problems, and already influencing how networks are run day to day.

Recent TM Forum data reflects this shift, although it needs to be read carefully. According to “Assessing CSPs’ progress towards Level 4 autonomous networks: Benchmark Report”, around one fifth of operators report that they are operating at Level 3 or above at an overall level, and while Level 4 is beginning to appear in more advanced scenarios, it is not yet widespread, nor is it consistently deployed across domains. What matters is not the headline number, what matters is the direction of travel.

Across production networks, we are seeing repeatable, scalable use cases where systems are able to act with a high degree of independence within defined boundaries, handling complexity in ways that would previously have required constant human intervention. These are not isolated experiments, and they are not theoretical constructs. They are practical implementations, grounded in operational need, and they are quietly reshaping expectations of what is possible. The industry has not changed its ambitions. It has changed its approach.

From Automation to Understanding

For years, automation in telecom was largely about efficiency, replacing manual tasks with predefined workflows and rule-based processes that could execute faster and more consistently than people. As networks became more dynamic, more distributed, and more interdependent across domains, rigid automation struggled to keep up. Systems could follow instructions, but they could not adapt meaningfully when conditions changed beyond what had been anticipated. The shift we are now seeing is from pilot to production, and from execution to understanding.

Operators are no longer asking systems simply to do what they have been told. They are asking them to interpret intent, to assess context, and to make decisions that balance multiple objectives at once, whether that is performance, resilience, cost, or energy efficiency.

Intent driven control sits at the heart of this shift. When intent is defined in a way that systems can continuously interpret and enforce, it creates the conditions for autonomy to scale, not because every action is predefined, but because every action is aligned to an outcome. The challenge, as many operators will recognize, is not defining intent in isolation. It is applying it consistently across domains that have historically evolved in silos, with different data models, interfaces, and operational ownership.

Artificial intelligence plays an important role here, although not in the way it is often portrayed. The value is not in handing over control to opaque systems, but in augmenting the network’s ability to perceive, to anticipate, and to prioritize. AI helps interpret vast amounts of telemetry, identify patterns that would otherwise be missed, and evaluate trade offs in real time. Crucially, it operates within guardrails defined by architecture and policy, reinforcing rather than replacing operator control.

Autonomy, in this sense, is not about removing humans from the loop. It is about changing how control is expressed, moving from manual intervention to continuous, policy driven oversight.

Turning Capability to Measurable Value