Operational pressure inside telecom networks continues to increase as scale and complexity outpace traditional workflows. Microsoft’s use of Network AI points to a different approach, where AI takes responsibility for a large share of routine operational work that previously required continuous human effort. That shift changes how organizations think about efficiency, placing less emphasis on scaling teams for repetitive tasks and more on embedding intelligence directly into operational processes and systems with AI agents.
These systems can take ownership of defined tasks such as incident response or field coordination, managing workflows end-to-end and supporting engineers with real-time information. Rather than acting as passive tools, agents function as active participants in operations, helping drive networks toward more consistent and responsive outcomes.
Greater autonomy also introduces new operational requirements. Visibility into agent activity, clear policy boundaries, and structured human oversight remain critical, particularly when actions affect live network environments. As AI becomes more deeply embedded in operations, the balance between automation, control, and accountability is emerging as a defining consideration for telecom operators.
Denizcan Billor:
Internally in Microsoft, we've been working on NetAI for over three years. I think people greatly underestimate how they can leverage AI to do day-to-day operations in their business. The quicker you can integrate AI part of your workflows, part of your operations, part of your business' intelligence, you will get this compounding of value. So if your network grows by 10X, if it grows by 100X, it just doesn't scale to grow a workforce with 10X or 100X for completing these repetitive tasks. Instead of scaling your workforce for volume of issues, you have to scale your workforce from slices of understanding. No matter how big your network is, you still will need core groups of people who have a fundamental end-to-end understanding of the infrastructure, the failure modes.
Iain Thornhill:
That's why we put together a framework, which is called NOA Network Operations Agent Framework. So MyOS is a fully autonomous agent at the point that someone has such as a fiber incident, such as a fiber cut, he's triggered and he then manages the flow to get the person assigned to go out to the site, he manages the quality throughout the entire process whilst the field force are out there fixing it. But more than that, he can even provide realtime information to the field force engineers about what's actually happening on the network as they fix it.
Denizcan Billor:
MyOS is really cool just as Iain talked about because it actually took a problem that we had engineers working on it 24x7 and then we could have an agent do it. AI can do something fundamentally that code could not do and that is actually that AI can understand knowledge. AI can actually understand your intention in a way you could never do in code. These agents are literally like coworkers who are working 24x7 on ticket queue and driving the network to a healthier state.
Iain Thornhill:
Putting a strong control plane at the top of this is incredibly important and that means you need to have good observability of agents, it means you need to have good optimization of agents. But on top of that and probably most important of all, you need to have good rules, policy, and security on your AI orchestration framework.
Denizcan Billor:
When agents are going to perform a change that's irreversible or a change that they haven't been given jurisdiction to do, they ask for help from humans and the humans then guide them. So it's very conceivable that within four or five years, there will be these AI systems that will look like super intelligence to us. They will be able to analyze all parts of the network, all layers of the network, investigate the devices, investigate the logs, and give us insights we could never get before.