Videos
VIAVI on Autonomous Networks, Agentic AI, and Human Oversight
Insights
- Autonomous networks require human oversight and governance frameworks to ensure AI agents learn, adapt, and deliver trusted outcomes.
- Agentic AI enables many-to-many integrations across telecom environments, reducing complexity and accelerating automation at scale.
- The pace of autonomous network adoption depends as much on legacy infrastructure constraints as on technology readiness.
At MWC 2026, Per Kangru of VIAVI Solutions discusses the evolution from traditional network automation to autonomous networks powered by agentic AI. He explains that autonomous systems differ fundamentally from deterministic automation because they focus on achieving outcomes rather than following predefined steps, making human oversight and governance essential for building trust and accountability. The conversation explores how AI agents and orchestration frameworks can simplify complex integrations across telecom environments, enabling operators to scale automation more effectively. Kangru also highlights the challenges many telecom providers face as they modernize legacy infrastructure, noting that progress toward autonomous networks often depends on an operator's starting point as much as its technology investments. Looking ahead, he argues that the combination of intelligent agents, clearly defined skills, and strong oversight frameworks will be critical to unlocking the next phase of telecom transformation.
Per Kangru:
I could build automation, I could then look at what is being automated. That's basically a very deterministic process. Autonomous is really not deterministic. Intent is solve this problem. And if we want to solve this problem there's a million ways to do it. So this is very similar to that when you bring in a new employee in a company you trust that he is relatively skilled, he's been trained, he's been grown some things, he's been maybe on a set of on the job trainings and so on. But now we have an expectation that this individual will be able to achieve a certain outcome. But you don't necessarily dictate do it in these phases because then you can equal to get a robot to do it. Autonomous agents can be very much similar. So this agent before you can fully trust it. And it's not only one agent but it's composition of agents. So a composition of employees, can you actually use them to get the right outcome. That's where this oversight mechanism that you very, very, very accurately have in your framework and then your frontier telecom blueprint that allows you to really achieve that leverage. Because now you have that oversight that allows autonomous to work. You give it the feedback it learns, it avoids that mistake next time. So that human oversight there is essential at this point in time.
Samad Masood:
Are there good examples of telcos that have moved quite far ahead in the autonomous networks?
Per Kangru:
So short answer is yes. But what's more interesting is to understand as well, not just because people may be advanced, it’s why they are advanced. Because now this is really a matter of, if you start with a complete greenfield operator, as we see in a couple of examples, of course you build it as autonomous from the get go, but almost every other telco in the world, they have a brownfield. That brownfield has to evolve and you can't do that and just throw out everything and restart. So yes, you have an evolution. Almost every operator wants to go on that evolution journey. Some have come quite far, but others they, I mean, they are stuck with certain legacy assets. That really makes it hard to move. That's where if you measure someone only on where they are today, but not why they are where they are, it gets very uneven. I would say.
Per Kangru:
The key thing with agents, what they do is that it allows you to do many to many mapping between different use cases. Historically, we've been very good at saying, okay, I now have a use case set, I have a pipeline of things I need to do, and I basically do a sort of integration of it. I connect one thing A with thing B with thing C, then I can achieve an outcome. The problem with that is that it became very bespoke integrations. And when I needed many of them, it was really, really heavy to do that. But agentic together with then the right sort of orchestration framework for it allows you to do is to actually get many to many integrations with zero effort basically. So the way I try, I normally try to describe this is that agents, they have skills. These skills will then autonomously be used by their LLMs to solve a certain problem.
Per Kangru:
The real important thing is what skills do you need and how should you apply them? And my analogy here is that people cook food all the time in the world today. So people go to the supermarket, they buy ingredients to cook food. But now if I come with a new super ingredient, that's going to make the world's best pancakes, but no one knows how to use it, no one knows it's available that it can make the super pancakes. It's going to stand on the shelf in that supermarket unused, even if it’s the best thing since sliced bread. Same thing here with agents and skills. If we bring out a super agent with a super skill, but no one knows how to use it, any of the LLMs or private knowledge will never be able to use it if we can't articulate it very well. And that's where really questions on many of these things comes in. Can we first of all provide a set of agents with a set of skills that the system know how to use? Then we can achieve great outcomes. The next is now to say, can we now help people to get a little bit more of these intelligent skills, the new ingredients, new flavors to the fruit that makes the whole packaging even much better, much better outcomes.
