James Pearce, TelecomTV:
I'm delighted to be joined today by Miguel Familiar-Cabero, who's going to be talking to us about next-level performance and resilience in the core. Miguel, nice to see you. Please talk us through it.

Miguel Familiar-Cabero, Ericsson:
Thank you, James. Thank you. So this is a story about autonomy in the intelligent core, in the core network. This is an evolution staircase approach where we are starting from automatic deployment like scripting, but now we are leaving that. And we are entering into agentic AI in the core network, then followed by intent to achieve full autonomy when it comes to autonomous network level four for selected use cases. And then the story also goes into how we are going to industrialise those selected use cases when it comes to cApps. And those cApps are applications to optimise the core network for preventive healing, high performance and capacity management, into the Ericsson Core Domain Manager that is going to allow cross-domain, end-to-end, open ecosystem for innovation and evolution into the intelligent core.

So in this demonstration, we show one use case in a live network, which is a 5G lab. And then basically we are running a real PCF and UPF in the core network — so the user plane and control plane for the core. We are collecting metrics. We are collecting also configuration parameters that model the network. And with these data points, we are calculating a baseline with generative AI to create a latent space. So we are able to simulate new scenarios with a digital twin capability. The combination of these use cases is the so-called capacity audit, but also a capacity prediction. So what you have here on the screen in the scatter plot, each one of these dots is a combination of many parameters and configuration metrics that comes into a billion-set of potential configurations in the network. With this, we create a baseline and from there we are able to simulate, for instance, traffic explosions, energy savings — so multiple scenarios to understand how the system should be configured to achieve, based on an intent, a target and a goal.

And you can see that this allows the customer to simulate and to play with the generative AI for specific models to understand how the network should look. Those are initial use cases, but also we are able to close the loop, and this is something that we also present at the demo. Again, with a live system, we should be able to monitor the system, but with an intent we should be able to simulate and execute scenarios like, for example, a traffic explosion to 70 million packets per second, and then close the loop into the network. So we can see that in action. The network is activating — so closing the loop. And then we can also see how the number of pods and the hardware core CPU load are dynamically reacting. The beauty of this is that it is not based on LLMs.

So this is not token prediction. We are not relying on GPUs. This is a commodity model that is really optimised for the network and is able to interact and act in real time. We combine also this solution with simulation to show thought leadership about preventive network healing, and how, using agentic AI with intent, we are able to let the network be optimised for a given intent and how the system is going to recommend, for selected KPIs, what the exact boundaries are. So the system is collecting metrics from the network and it is also recommending parameters to avoid incidents. But also we have use cases to simulate high signalling, some misbehaving UEs that are running a DDoS attack, or full configuration inter-frequency handover. And based on that, the system is able to give us an early warning that a problem is there. So what we have shown here are selected use cases.

This year we are innovating with tier-one customers for selected use cases in capacity planning and also network function anomaly detection. Next year we are going to have our initial general availability for those. So it is an exciting journey with our customers, and more will come.

James Pearce, TelecomTV:
Super exciting. Thank you so much for that, Miguel.

Miguel Familiar-Cabero, Ericsson:
Thank you.