- Operators are pursuing autonomy not just for efficiency, but because experienced network staff are retiring faster than they can be replaced
- The next challenge is not building autonomous networks, but managing thousands of competing AI-driven control loops operating across shared infrastructure
- Success will depend on trust, sovereignty and human expertise — not technology alone
The telecom industry likes to talk about autonomous networking as a technology challenge.
Per Kangru sees it differently.
The VP network performance and automation solutions at Viavi believes the industry's biggest obstacle is not artificial intelligence, orchestration software or even network architecture. It is people.
Or more precisely, the growing gap between the complexity of modern networks and the number of people available to operate them.
"Attracting people to do a manual operations job in a NOC is becoming almost impossible," Kangru said during a recent interview with me for Fierce Network TV (FNTV). "Many operators are facing a situation where experienced staff are retiring and there simply aren't enough new people coming in with the right skills."
That challenge sits at the heart of the telecom industry's accelerating push toward autonomous networking.
For years, autonomy was treated as a future aspiration. Today it is becoming an operational necessity.
Yet Kangru argues that telecom operators face a fundamentally different challenge from hyperscalers.
Cloud companies largely build the software they operate. Telecommunications providers generally purchase equipment and software from a broad ecosystem of vendors and then attempt to operate it as a coherent whole.
"If you're a hyperscaler, the same teams that build the software often operate it," Kangru explained. "Telecom operators don't work that way. They buy equipment and software from vendors. That means autonomy often involves retrofitting autonomous operation into systems that were never originally designed for it."
That distinction helps explain why autonomy appears to progress more naturally inside cloud environments than traditional carrier networks.
The challenge is not simply brownfield versus greenfield deployment. It is organizational.
"Operators know how to buy and operate systems," Kangru said. "They generally don't build them. That creates a very different starting point for autonomy."
The hidden driver: sovereignty
The labor challenge is increasingly colliding with another industry trend: operational sovereignty.
For years, operators could offset skills shortages through outsourcing models and global operations centers. Increasingly, however, national regulations are demanding that critical infrastructure be operated by security-cleared nationals within specific jurisdictions.
That changes the equation.
"The options available to operators are becoming fewer," Viavi's Kangru said. "In many countries you can't simply solve the problem by importing labor or moving operations elsewhere. The question becomes how you continue operating increasingly complex infrastructure with the resources available."
Autonomy, therefore, is no longer merely a cost-reduction exercise.
It is becoming a strategy for operational sovereignty. Regular readers will know this is something of a hobby horse of mine. I call it OpSov, short for Operational Sovereignty: the ability of an organization, industry or nation to continue operating critical digital infrastructure using its own people, processes and resources. Viewed through that lens, autonomous networking is not simply about efficiency. It is increasingly about ensuring that networks can continue to function even as labor pools shrink, skills become scarce and regulatory requirements limit traditional outsourcing models.
Why autonomy is really about business agility
The industry often frames autonomous networking around operational efficiency.
Kangru believes that misses the larger opportunity.
The real value of autonomy, he argues, lies in business agility.
Future enterprise customers will increasingly expect infrastructure to behave like software: dynamic, programmable and instantly available through APIs.
"If operators cannot provide that level of flexibility, customers will move to providers that can," Kangru said. "The question is no longer just whether you can fix a fault. The question is whether your operational model can support the business agility customers expect."
That requirement extends beyond telecom and into the broader AI infrastructure economy.
As enterprises consume AI, cloud and edge resources dynamically, networks must become equally dynamic.
The challenge is not merely transporting traffic. It is continuously matching infrastructure resources to changing business demands.
The agentic AI problem nobody has solved
Perhaps the most intriguing part of the discussion emerged when the conversation turned toward agentic AI.
Much of the current industry narrative focuses on individual AI agents.
Kangru believes the real challenge begins when organizations deploy hundreds or thousands of them simultaneously.
"It's very easy today to build a proof of concept around agentic AI," he said. "The problem comes when you have many autonomous control loops operating against shared infrastructure."
In that environment, individual decisions may appear rational while collectively creating instability.
One agent optimizes a service. Another optimizes a different service. A third responds to a customer request. A fourth attempts to rebalance capacity. Each may be behaving correctly in isolation while producing unintended consequences at system level.
"How do you ensure hundreds or thousands of agentic loops manage shared resources appropriately?" Kangru asked. "How do you understand when things are about to go wrong, and how do you intervene?"
The challenge becomes particularly acute in mission-critical environments.
Academic researchers have proposed market-based approaches where autonomous agents effectively bid against each other for resources and priorities.
Interesting in theory. Potentially dangerous in practice.
"What happens when a critical emergency service competes with another workload?" Kangru said. "You cannot have a situation where the optimization worked perfectly but the emergency call failed."
The industry has not yet fully answered that question.
The risk of losing the human touch
The debate around autonomy is often framed as humans versus machines.
Kangru sees a subtler challenge.
If every operator trains models on similar data, deploys similar architectures and relies on similar AI systems, networks may become increasingly indistinguishable from one another.
"The concern is that operators could lose their personality," he said.
Historically, companies differentiated themselves through people, processes and institutional knowledge.
Autonomy requires that knowledge to be translated into software.
Kangru describes the process as "digitizing silent intellectual capital" — capturing decades of operational experience and converting it into repeatable decision-making frameworks.
Done correctly, autonomy can preserve organizational identity. Done poorly, it risks reducing operators to interchangeable infrastructure providers.
"If operators lose that identity, they lose part of their brand value."
A new role for testing
The shift toward autonomous systems is also changing Viavi's role.
Historically, telecom testing focused on validating compliance with standards and specifications. Autonomous systems introduce a different challenge. There is often no single deterministic outcome to test against.
Instead, testing increasingly becomes an exercise in understanding how systems behave under unexpected conditions.
"In the past you could read a specification and build a test plan," Kangru said. "With autonomy, the challenge is understanding how the system behaves when conditions become unpredictable."
That means validating not only whether a system works, but whether it fails safely.
It also means developing ways to challenge autonomous systems intentionally, exposing edge cases and interactions that traditional testing methodologies were never designed to address.
The result is a profound shift in the role of companies such as VIAVI. Testing is no longer simply about proving compliance. It is increasingly about building confidence.
Trust becomes the new infrastructure
For all the discussion around AI, automation and orchestration, the interview repeatedly returned to one underlying theme: trust.
Kangru argues that the industry's next challenge is not building autonomous networks, but building confidence in them. Operators once had to learn to trust IP networking. They later had to learn to trust virtualization and cloud-native architectures. Now they must learn to trust autonomy.
"The technology can provide the capability," Kangru said. "The challenge is giving operators the confidence to use it."
That may ultimately prove to be the most important lesson from telecom's autonomous future.
Autonomous networking, it turns out, is not a networking problem. It is a people problem.