Autonomous Networks: How Aircom is Making Agentic AI Real for the RAN with raNora

An Interview with Khurram Chaudhry, General Manager & CTO, Aircom

Autonomous networks have moved from vision to strategic priority for communications service providers (CSPs). Agentic AI is widely seen as the technology that will get them there, yet most of the industry is still at the stage of demonstrations and roadmap slides. The hard part is turning intelligent decisions into reliable, repeatable improvements in network performance, operational efficiency and investment effectiveness.

Aircom has taken a different route. Rather than adding a conversational interface to existing software, it has built raNora: an agentic AI platform purpose-built for the radio access network, grounded in decades of RAN planning and optimization expertise and designed from the outset for the governed, auditable autonomy that operators actually need.

We spoke with Khurram Chaudhry, General Manager & CTO at Aircom, about why a purpose-built software framework is the missing piece in the autonomous-network conversation, how raNora brings agentic AI into live RAN engineering today, and what the path to L4/L5 autonomy looks like from here.

Q1. Autonomous networks are attracting enormous attention. What do you believe is missing from much of the discussion?

Most discussions start with AI agents and how many tasks they can automate. In a radio network, deciding what to do is only part of the challenge. The system has to understand many parameters as well as the consequences of changing one part of the network on another.

A language model cannot replace validated radio-engineering methods, and it should not be asked to. An autonomous workflow is much more than that. It’s a framework that connects the reasoning of an AI agent to trusted network data; to proven engineering functions that can calculate alternatives and quantify expected effects; and to the governance that decides when a change is allowed to be executed.

That is the gap we set out to close with raNora. Our view is simple: without a purpose-built software framework, autonomy is very difficult to achieve. You can demonstrate an agent answering questions about the network in an afternoon. Getting from there to agents that act safely in a live RAN, with every step traceable, is an engineering program. That program is what Aircom has been executing.

Q2. What is raNora, and how does it differ from the agentic AI initiatives operators are seeing elsewhere?

raNora is Aircom’s independent agentic AI platform for the radio network. It is designed to interpret engineering objectives expressed in plain language, identify the required data and engineering functions, perform the necessary analysis, and provide engineers with supporting evidence and recommended actions.

Three design decisions set it apart. First, raNora is grounded in the operator’s own data. Agents reason over what the network actually looks like, not over a generic description of one. Second, it is governed by human checkpoints. Reading and analysing the network is autonomous; any action that changes the network inventory or triggers a significant computation pauses for an engineer’s explicit approval. Third, it is explainable end to end. Every answer carries its evidence, every step is logged, and every decision can be reconstructed after the fact.

Those are not add-ons. They are the reason raNora can be used by a real operator engineering team rather than just demonstrated to one. Many agentic initiatives in the industry are still general-purpose assistants layered over a dashboard. raNora was designed as RAN engineering infrastructure from day one, and can be deployed on-premises within the operator’s environment, ensuring that sensitive data does not leave their environment.

Q3. Why is RAN domain knowledge such a decisive advantage when building an agentic platform?

Radio networks are highly interconnected systems. A configuration change intended to improve coverage in one area can increase interference in a neighbouring area or alter mobility behaviour. A recommendation that looks reasonable against a single KPI can be wrong when it is evaluated in the wider network context.

Experienced RAN engineers understand those trade-offs and raNora translates that expertise into something an agent can actually use. The result is: agents reason; engineering functions calculate; and policies decide what may be executed.

Aircom’s position is unusual. We have engineering specialists, established planning and optimization tools that operators have trusted for years, and now we have an agentic platform that connects them. Autonomy should build on proven radio-engineering principles rather than expecting an AI model to rediscover them. Having all three capabilities inside one organization is what allows us to move faster than companies that have only one of them.

Q4. Operators already run planning and optimization software. How does raNora turn those existing capabilities into part of an autonomous network?

By making established engineering functions consumable by agents. raNora interprets engineering objectives, identifies relevant tools and data, coordinates activities, and presents results with supporting evidence. The engineering tools continue to perform the specialized calculations, while agents bring the different activities together. This allows raNora to span the full RAN lifecycle rather than a single task.

Planning and optimization are usually managed as separate disciplines with separate tools and datasets, yet they are inherently connected. Because Aircom’s capabilities cover both, a raNora workflow can compare expected and observed behavior in one engineering context. When investigating a localized coverage or capacity issue, it can correlate geolocated measurements, traffic distribution and antenna configuration with planning predictions, and assess alternatives such as antenna changes, capacity redistribution, physical redesign or an additional site, taking neighboring-cell effects, performance trade-offs and investment implications into account.

Under the hood this means the platform has to handle asynchronous execution, job tracking, versioned results and explicit failure handling, because many of these computations are heavy and long-running. It is not a matter of adding a chat window to existing software. The engineering capabilities have to be made reliable enough for an automated workflow to depend on them, and that is the work we have invested in.

Q5. Governance and trust come up constantly in operator conversations. How does raNora make autonomy safe enough to deploy?

We treat autonomy as something you earn progressively, not something you switch on. raNora is structured around a clear separation between observing, recommending and acting. Agents can query, audit and analyse the network freely because those operations are read-only and fully scoped to what each user is permitted to see. The moment an agent proposes a change, the workflow pauses: the engineer sees the proposed change set, the evidence behind it and the expected impact, and nothing is written to the network until it is approved. The same applies to computationally significant actions, where the engineer confirms the scope before any resources are committed.

Everything is traceable, so an operator can audit not only what the network looked like but how a recommendation was reached. The platform provides controlled access to information and protects sensitive network data. The entire platform, including the models themselves, can run inside the operator’s own environment.

These features let operators adopt agentic AI in a regulated, mission-critical setting. It also maps directly onto the industry’s autonomy levels. Today, raNora delivers multi-step, closed-loop engineering workflows with defined human checkpoints, which is the practical definition of L3. The architecture is built so that specific, well-understood actions can move to policy-controlled execution as confidence grows. That is the route to L4/L5 autonomy: expanding what the system is trusted to do on its own, one governed capability at a time.

Q6. Looking ahead, why is Aircom well positioned to lead the move toward autonomous RAN?

I see three foundations. First, deep RAN domain knowledge, built through our engineering teams, our methods and decades of planning and optimization work with operators around the world. Second, proven engineering capabilities and algorithms that are already exposed as services for autonomous workflows rather than sitting in standalone desktop tools. Third, and this is the piece most of the industry still lacks, a purpose-built agentic platform in raNora that brings data, engineering functions, machine learning and governance together in one framework, and is already being deployed with operators.

This combination matters because operators should not have to recreate radio-engineering intelligence from scratch to introduce autonomy, and they should not have to bet on an unproven general-purpose AI layer either. With raNora they can build progressively on capabilities they already trust: assisted analysis first, then governed recommendations, then policy-controlled execution where it is justified. Our roadmap extends the same framework across all RAN domains ensuring the network can be managed as one system rather than a set of isolated tasks.

The measure of success will not be the number of AI agents deployed. It will be whether operators complete engineering work faster, improve network experience and make better investment decisions, with outcomes that are traceable and measurable. That is what raNora is built to deliver, and it is why I believe Aircom is leading the practical implementation of agentic AI for autonomous networks rather than just talking about it.