At Network X Americas 2026, Ahmed Abdelaziz, VP, Automation & Transformation, Rakuten Symphony, joined a panel that pushed the AI-in-telecom conversation beyond theory and into execution: what does it take to apply AI in live networks, at scale, across systems, teams and operating environments?
Moderated by Kristian Toivo, Executive Director, Telecom Infra Project (TIP), the panel brought together:
The discussion closely reflected an overall theme: AIOps is not simply about adding AI to the network. It is about creating the operational model, governance and organizational alignment needed to make AI usable at scale.
What stood out across the panel was how far the industry has moved beyond experimentation. The focus is no longer whether AI can help telecom networks, but where it is already delivering value and what is required to scale it.
“You don’t start by trying to solve for everything, you start by solving for the hardest, most impactful problems,” says Abdelaziz. “By focusing on a specific use case, like energy reduction, you create a blueprint for data integration and governance. Once you prove that a model can reliably manage one domain, you have the architectural foundation to scale that trust and logic across the rest of the network.”
Abdelaziz, who noted active work in incident recovery assurance and predictive maintenance, added: “We are past the experimental phase. The breadth of activity now underway across RAN, automation, agentic frameworks, RAN AI and 6G, underscores the need for a policy management layer that can orchestrate AI use cases into all these domains consistently.”
Bureau highlighted similar practical outcomes in the radio network, including anomaly detection, network observability and the ability to push remedial actions into the network. He also pointed to a more customer-centric benefit: “We are able to predict how customers are perceiving network experience.”
A second theme was the relationship between data architecture and AI scale.
Mahdi described a more distributed model in which “the data stays where it is,” rather than always being centralized. “In telecom, valuable operational data is often trapped inside vendor-specific systems and interfaces. The challenge is not simply collecting more data but making different systems interoperable enough to use it effectively.”
That is where programmability becomes important. Open interfaces, O-RAN certification and common frameworks help expose those environments and make them easier to work across. Mahdi also pointed to network language models (NLMs) as a way to create a more usable language for the network itself, while simulation and synthetic data can help establish baseline models and refine them over time.
Taken together, those points reinforced one of Abdelaziz’s core themes: AI maturity is not just about model choice. It depends on whether operators can create the data and execution environment in which models can act reliably.
If the technical foundations are becoming clearer, the panel also showed why AIOps remains a leadership challenge.
Sankara described Reliance Jio’s ambition “to achieve fully automated agentic network,” but stressed that this requires change across “three layers” at once: networks, peopleand organizations. He reminded the audience that autonomy is not achieved by software deployment alone.
Bureau described a similar balancing act at TELUS, combining centralized data engineering with a federated AI model to scale use cases without losing control. That organizational discipline matters because AI in telecom is inherently cross-domain. Without alignment, even strong use cases risk remaining local successes rather than anenterprise capability.
Soni focused on the opportunity to “bridge the legacy with new,” using foundational models and LLMs to correlate data more effectively, improve predictive maintenance, enable autocorrection and support energy savings.
Taken together, the panel offered a practical view of where telecom is heading. AI in telecom is no longer just an analytics layer or a productivity tool. It is becoming part of the operating fabric of the network. But making that shift work in practice requires more than technical ambition. It requires interoperable data, programmable systems, policy-driven orchestration and organizations prepared to trust and govern more autonomous ways of operating.
The future of AIOps will be shaped not by how many AI tools operators deploy, but by how effectively they build the operating model around them.
As Abdelaziz concludes, “We have to shift our mindset from ‘AI as a tool’ to ‘AI as a teammate.’ Trusted autonomy isn't about removing the human, it’s about providing the guardrails so the human can focus on high-level strategy while the network handles the complexity of real-time optimization.”
