Laurent Giraid brings a profound understanding of how machine learning and agentic AI are reshaping the backbone of our digital world. With extensive experience navigating the intersection of telecom infrastructure and cloud-native logic, he offers a unique vantage point on the recent pilot programs between Nokia and AWS. This conversation explores how “agentic AI” is moving from a speculative concept to a functional operational controller that manages the complex nuances of 5G network slicing for global operators like Orange and du.
Traditional 5G network slicing often relies on manual planning and fixed configurations. How do AI agents fundamentally change this workflow when processing real-time variables like weather or local event schedules? Please walk us through the step-by-step logic these agents use to maintain performance levels.
The shift from manual planning to agentic AI feels like moving from a rigid, printed map to a live, breathing GPS system that anticipates every roadblock before you even see it. Traditionally, engineers had to sit down and pre-configure slices based on static guesses, which meant that if a sudden thunderstorm rolled in or an unscheduled crowd gathered, the network simply couldn’t breathe or expand to meet that pressure. These new AI agents, powered by platforms like Amazon Bedrock, function by constantly ingesting a stream of telemetry that includes latency metrics and congestion levels while overlaying external data such as local event calendars. The logic begins with continuous observation where the agent identifies a performance dip; it then reasons through the available resources and autonomously executes an adjustment to the network settings to maintain the agreed service level. This creates a closed-loop system where the network is essentially self-healing, ensuring that a high-bandwidth slice for a broadcaster stays crystal clear even when the surrounding environment becomes chaotic.
Enterprise customers are increasingly seeking connectivity that scales on demand, similar to cloud computing. What are the specific technical hurdles in moving telecom services toward this model, and how could this shift finally unlock new revenue streams for operators? Please provide specific examples or metrics.
The primary technical hurdle has always been the sheer operational complexity of 5G infrastructure, which GSMA Intelligence notes has slowed down the adoption of slicing despite its massive potential for enterprise income. Unlike the cloud, where you spin up a virtual machine with a click, telecom resources are tied to physical and virtualized radio layers that are notoriously difficult to automate in real time without breaking something else. By solving this through automation, operators can finally move away from selling “best effort” data plans and start selling guaranteed outcomes, such as temporary high-priority slices for emergency responders entering a disaster area. We are looking at a future where an enterprise can request a massive burst of low-latency connectivity for a three-hour factory audit and then release those resources immediately, mirroring the “pay-as-you-go” cloud model. This shift transforms the network from a dumb pipe into a flexible service platform, allowing operators to capture revenue from high-value, short-term events that were previously too expensive or slow to set up manually.
Integrating AI-driven control loops into critical communication infrastructure introduces significant concerns regarding accountability and reliability. How are operators currently validating system behavior under real-world conditions, and what specific oversight mechanisms must remain in place? Please share any anecdotes regarding the testing phase.
When you are dealing with critical communication infrastructure, you cannot simply “move fast and break things,” because the stakes involve public safety and essential commerce. Operators like Orange are currently navigating this by introducing AI-driven automation in a very granular, graduated fashion during these pilot rollouts to ensure the system doesn’t hallucinate or make erratic resource shifts. The current oversight mechanisms involve “human-in-the-loop” configurations where the AI proposes a change or operates within very strict guardrails, and every autonomous decision is logged for rigorous auditability. During the testing phases, there is a palpable sense of caution; engineers are watching these agents handle simulated spikes in traffic to see if the AI prioritizes emergency services correctly over standard consumer traffic. The goal is to prove that the AI can be more reliable than a tired human operator at 3:00 AM, but until the data consistently shows 100% adherence to safety protocols, regulators and operators will keep a firm hand on the “manual override” switch.
The move toward hosting core network functions on public cloud platforms is accelerating. What are the operational implications of layering “agentic AI” on top of these cloud-based systems for high-density environments like stadiums? Please detail the impact on latency and resource allocation.
Layering agentic AI on top of public cloud platforms like AWS fundamentally changes how we handle the vibrant energy and unpredictable surges of a stadium environment. When the core network functions are cloud-native, the AI can scale virtual resources up or down in seconds rather than minutes, which is crucial when 50,000 people simultaneously try to upload high-definition video during a goal. This setup minimizes the physical distance data must travel and uses Nokia’s slicing tools to ensure that critical stadium operations—like security feeds or point-of-sale systems—don’t suffer from the latency spikes caused by the screaming fans in the stands. The AI acts as a sophisticated traffic cop, sensing the onset of congestion and instantly reallocating “lanes” on the digital highway to keep the most important traffic moving smoothly. It creates a much more efficient use of hardware, as resources aren’t sitting idle during the off-season but are instead dynamically provisioned exactly when and where the density demands it.
What is your forecast for AI-driven autonomous network slicing?
I forecast that within the next three to five years, we will see the total disappearance of manual configuration for enterprise 5G services as autonomous slicing becomes the industry standard. As operators move more of their operations into the software-defined realm, the sheer volume of data will make it impossible for humans to manage these networks without AI agents acting as the primary controllers. We will transition from a world of “static” connectivity to “intent-based” networking, where a business simply tells the network what it needs to accomplish, and the AI handles the invisible complexity of provisioning and maintaining that slice. This evolution will finally allow 5G to live up to its original promise, turning global telecommunications into a fluid, responsive utility that is as easy to scale and manage as any modern web application.
