Telecom has cycled through this before. 5G was supposed to be the leap. Then came cloud-native cores, then a wave of AI-driven automation, each pitched in turn as the shift that would finally make networks run themselves. None quite delivered on that promise.

Rohit Chowdhary, Head of Advanced Consulting Services EMEA at Nokia, believes a fourth shift is underway, one that could fundamentally change how networks are built, operated and maintained. “It’s a year of agents,” he says.

For an industry that has historically relied on large engineering teams, lengthy deployment cycles and manual processes, the shift could be significant. Operators have spent years pursuing the vision of autonomous networks, systems capable of monitoring, optimizing and repairing themselves with minimal human intervention. The ambition has long existed. What has been missing is the technology to make it possible.

THE ROAD TO AUTOMATION

Telecom doesn’t lack for technological upheaval. The past decade alone brought the 5G rollout, the standalone versus non-standalone debate, and a heavy bet on cloud-native infrastructure. Chowdhary treats those shifts as groundwork for what’s coming.

The first job was prying networks loose from rigid hardware and rebuilding them on software that could be deployed and upgraded without the usual drag. That drag has been real. “It takes us weeks to deploy a network. It takes us months to scale it across a country, across multiple nodes,” Chowdhary says. “It’s not very efficient, and it takes a lot of human effort.”

Software-defined networks opened the door to automating deployment, testing, migration and upgrades. AI added another layer, sifting through operational data at a scale and speed no human team could match. But a gap remained. AI could flag anomalies, parse alarms, and forecast failures. Acting on any of it was still left to people.

WHY AGENTIC AI CHANGES THE EQUATION

The gap between insight and action is why agentic AI matters to operators. Traditional AI is built to interpret data. Agentic systems are built to act on it. Chowdhary calls that distinction the core of the shift.

“With AI, you can analyze data, you can create intelligence, but you cannot do the things,” he says. “The doing part of it was part of the agentic framework.” In practice, the work gets split across specialized agents. One monitors alarms across the network. Another diagnoses the root cause. A third decides on the fix. A fourth carries it out.

“It’s like multiple humans talking to each other and doing a job,” Chowdhary says. “It started becoming multiple agents talking to each other and doing a job.” These agents coordinate through emerging protocols, exchanging information and making decisions without a human approving each step. That is the difference between automation and autonomy.

TM Forum’s autonomous network model lays out several levels of network autonomy. Many operators today sit at Level 2 or 3, where automation supports human decision-making but doesn’t replace it. Chowdhary expects agentic architectures to move operators toward Levels 4 and 5, where networks largely run themselves. “This is very powerful,” he says.

THE UNEXPECTED WORKFORCE SHIFT

Agentic AI is also changing the way telecom workforces look. The first wave of AI adoption demanded specialized data scientists. Large language models have lowered that bar. Companies still need the headcount. What they need has changed.

The skills required to build and manage these systems are changing. Chowdhary says this has meant bringing software engineering expertise back into teams. “I had to go back and hire software stack engineers in my team,” he says. “This is something I used to do 20 years back.”

It’s the same question playing out across every industry touched by AI: how much to automate versus how much to retrain. Chowdhary doesn’t see engineers being replaced. He sees experienced telecom professionals becoming the ones who train, validate and improve the AI itself. “We are seeing how it is, first of all, reducing the manual effort,” he says. “But then the humans who are involved in doing the jobs are helping to build this AI.”

Engineers move from running operations to designing and managing the systems that run them. “My faith in that process is becoming more robust,” Chowdhary says. “There is a way forward where we can take all that knowledge sitting in engineering teams and operations teams and repurpose that for making our AI more intelligent.”

THE EDGE BECOMES THE NEW BATTLEGROUND

The next challenge sits outside telecom altogether. As AI moves from chatbots and software into machines and physical environments, networks need to process information close to where it’s generated, not in a data center several hundred miles away. Chowdhary frames this as the shift from generative AI to physical AI. Autonomous vehicles, robots, drones, industrial automation and smart infrastructure all need decisions made in milliseconds, which rules out a round trip to the cloud.

“When you have a robot doing a job for you in a factory or maybe a port or airport, it has to process that query locally,” he says. “That query cannot go to the cloud.”

The trend is also shaping partnerships across the industry. Nokia recently expanded its collaboration with Nvidia, whose chips are increasingly being used for AI inference closer to end users. For telecom operators, that means networks are taking on a larger role in supporting computing workloads. “We have to provide processing and compute very close to that robot,” Chowdhary says.

Governments are factoring this into how they build digital infrastructure, too. Across the Middle East, public sector bodies are pouring money into mission-critical networks for emergency services, transport, utilities and public safety. Chowdhary says the conversation with governments has moved beyond basic connectivity to how that infrastructure can support broader economic and social goals. “You have a network, but you have this monetization layer which is helping them build more and more digital use cases on top of that platform,” he says.

That opens the door to smart city applications and AI-enabled public services. But for governments running mission-critical networks, uptime still outranks revenue. “I think, honestly, for public safety networks, mission-critical networks are more important because that is their prime goal,” Chowdhary says.

THE OUTLOOK AHEAD

If the past decade was defined by the race to build faster networks, the next may be defined by the race to make them autonomous.

The biggest unknown is how quickly society accepts AI systems that move beyond software and into the physical world. Autonomous vehicles are already running in parts of the United States and starting to appear elsewhere. Robotics keeps advancing. AI-powered infrastructure keeps gaining ground.

Trust is the harder problem. “The key fear is the security part,” Chowdhary says. “How do you put the proper guardrails around it?” Even so, he’s betting on a fairly short timeline. “My rough guess? Maybe five years from now, you are seeing quite heavy physical AI around you.”

What that timeline implies is bigger than faster connectivity. It points toward networks built to help machines think, collaborate and act on their own.