Orange is applying agentic AI across several operational domains, including 5G security operations, telco cloud operations, RAN energy optimization, lifecycle management, and incident management
In sum – what to know:
Operational outcomes — Orange said AI success should be measured by faster incident resolution, improved customer experience, and reduced manual coordination rather than AI activity metrics.
Business value — The operator said agentic AI use cases should be prioritized based on measurable business impact, not technical feasibility.
Trust first — Orange argued governance should enable AI adoption through risk-based controls, while operators should build trust progressively before expanding agent autonomy.
As telecom operators accelerate investment in agentic AI, success will depend less on the sophistication of the technology than on delivering measurable operational improvements through a gradual, business-driven adoption strategy, according to Orange.
Speaking during RCRTech’ Telco AI Forum, Philippe Insargue, vice president of cloud and software engineering at Orange, said operators should resist the temptation to pursue full autonomy too quickly and instead build trust progressively as AI systems mature.
“For me, the success is not really about AI activity in terms of how many agents are running, how many tokens are we having. It’s much more about operational outcomes end-to-end. How many incidents we resolve faster, how customers experience got fewer disruption, how operations team basically are spending less time coordinating manually,” Insargue said.
He described a maturity path that progresses from assistive AI, to supervised agentic AI, then bounded autonomy before eventually reaching orchestration across multiple domains.
“And here, honestly, I truly believe that the telcos do not succeed by jumping directly into the full autonomy. The maturity path honestly matters because you will learn, we will learn at every stages. And the trust that is earned at each stage, for me, is about the business, the value, and also how to level up the skill of the team,” he said.
According to Insargue, Orange is applying agentic AI across several operational domains, including 5G security operations, telco cloud operations, RAN energy optimization, lifecycle management, and incident management, while selecting projects based on expected business outcomes rather than technical novelty.
“It’s important to have use case. It’s even more important to have business case. And at Orange, we are really in a business value first selection. We are applying a methodology called high value scenario. And we don’t start from what is technically feasible, I would say. We start from what could create at the borders of our Orange affiliate the most business value,” he said.
The executive also addressed one of the industry’s emerging challenges: measuring AI’s return on investment as inference costs become an increasingly important operational expense.
“Honestly, I would treat tokens as unit cost, but measure them at a level of real outcome. Token cost per assisted decision, per resolved case, per completed workflow, not total token conception. The aggregate number for me is meaningless if we do not have the full context,” he said.
He added that organizations should establish measurement frameworks before deploying AI systems, rather than attempting to justify investments retrospectively.
“So for me, the causal chain to demonstrate is we need to track the AI that is used, what are the decision that is changed, what the impact on operational KPI that have been improved, and then what is the business value that we created here. And each link needs evidence,” the executive added.
Beyond performance metrics, Insargue argued that governance should be viewed as an enabler of adoption rather than a barrier. “The goal is not the governance over the trust. It’s the governance that is enabling the trust. And they are not in tension. A badly designed governance is in tension with adoption. A well-designed governance will speed all of this.”
He warned that governance frameworks that are either too restrictive or too permissive could undermine enterprise AI adoption, advocating instead for controls that are proportional to operational risk.
Looking ahead, Insargue said the telecommunications industry has already made significant progress with AI models, intent-based architectures and open-source ecosystems, but still faces important production challenges before agentic AI can scale across complex operational environments.
