How can network orchestration turn the cloud into new value for telcos?

Amy Cameron0:02
… Partners webinar titled orchestration from cloud to new value. So I'm delighted to have you all here, delighted to have a great panel of guests with us today, where we'll be looking at network orchestration, and how that with the cloud foundation can enable telcos to achieve new services, new revenues, new opportunities. So, before we kick off, the usual housekeeping, which I'm sure you've all heard many times before. One, this webinar is being recorded, so you'll be able to catch the on-demand version later if you want to.

Amy Cameron0:02
All attendees are in listen mode, but we will be taking questions when we get to the panel discussion, so please do put your questions in. We'll be very happy to take them as part of the discussion. And you can see here, you follow the little question box, and we'll take them. And then we'll share the slides and the recording after the webinar. So we have here today, absolutely wonderful guests, myself. I'm your moderator for the day. My colleague Belinda will be presenting some of STL's thoughts on the topic.

Amy Cameron0:02
Ignacio Gonzalez from Red Hat will be presenting some of Red Hat's views on this. And then we will open up the discussion with Alexis Salas from Orange Spain and Terje Jensen from Telenor, who will give their thoughts on what we've shared in our presentations and also how they are seeing orchestration within their organizations and impacting their customers and their networks. So thank you very much everybody for joining us today. This is the agenda. So as I mentioned, Belinda will start and then we'll hear from Nacho and then I will moderate the discussion.

Amy Cameron0:02
So without further ado, we will hand over to Belinda.

Belinda Martin1:53
Thank you, Amy. So, yes, so we're looking at what orchestration is, what new value it can generate for operators, and setting out our view as a primer to the discussion that is fundamental to realizing new value, and that the cloud is a fundamental underlay of that. So here is a slide that's drawn from our report, the telecom industry in the AI era, which is quite a sort of high level speculative forecast. But we believe that orchestration has the potential by 2035 with the advent of AI services, both delivered by the network and within the network to create more value for telcos than simple connectivity, standard commodity connectivity, and physical network assets on their own.

Belinda Martin1:53
Now we're talking about cloud native orchestration, but orchestration across multiple areas as well, but cloud native orchestration is the foundation in our view. And it's been rather overshadowed in discussion of cloud native, generally cloud native adoption by telcos, by the understandable focused on the RAN and the core, but it's a fundamental part of the cloud native journey, and it does have the potential to release a lot of new value. So if we go to the next slide. So here is a very schematic diagram of the orchestration opportunity in our view.

Belinda Martin1:53
And we believe that telcos that combine edge compute, for example, APIs, private networks, and of course, satellite coming into increasing adoption, including for private networks, and cloud-native cores. In this instance, the 5G standalone cores into modular platforms that are programmable that enterprises and other clients can use to run their own applications, industry specific services, will have the potential to capture 57 to 70 billion US dollars by 2030. And that's three and a half to four and a half times what those infrastructure assets in isolation would be able to deliver.

Belinda Martin1:53
So orchestration is what enables that platform model. Orchestration across a common cloud architecture is fundamental to that.

Belinda Martin4:26
Otherwise, every use case becomes a bespoke network integration project. Next slide, please. Now, here's a sort of crash course in what that involves. Kubernetes is what makes it work at the leaders and other distributions of Kubernetes are available, such as OpenShift from Red Hat. And Deutsche Telekom has developed its own version of Kubernetes-based cloud-native orchestration called the Operator-Led Cloud Automation Layer, OCA. It's built on their own Kubernetes cloud that they've developed over the years. and it provides the platform to combine network assets across via APIs and across the edge and private networks.

Belinda Martin4:26
And this is something that they're very sort of keen on for generating new value. We could go into the detail of the Kubernetes platform there, but I'm sure it's sort of familiar to a lot of you. We have the microservices containing the pods, Kubernetes resources and managing things like network state and data persistence and so on. So these are all cloud native network functions and together, by being orchestrated together, that's what delivers a new network service, essentially, put very simply. And Deutsche Telekom itself argues and claims that it believes that the cost of running their platform like that will be 30% lower per network function than if they used commercial orchestration platform.

Belinda Martin4:26
So that sort of shows, you know, stark relief, what are the potential cost benefits as well as the service benefits and the flexibility to run these things that can be delivered by cloud native orchestration. So we go to the next slide. And cloud native orchestration also creates sellable IP that operators can market to other operators and to vendors and make money from it as a product. So it's not just about monetizing your 5G as it's about productizing them too. And the most well-known example of that probably in the industry is Singtel Paragon, which orchestrates Edge and APIs and network cloud-native infrastructure, to enable enterprises to self-serve their connectivity via the APIs and run an ecosystem of applications that are supported across that and across any cloud platform.

Belinda Martin7:13
So multi-cloud management is an integral part of that, and that's enabled by cloud-native network functions, the ability to port those services across different Kubernetes-based clouds, essentially, using infrastructure exposure, so the programmability of that underlying infrastructure via APIs. And they've already sold that platform to a number of other operators, AIS in Thailand, Chunghwa Telecom, Telefónica in Spain, and Telkomsel being among those. But OCA platform we've just discussed and has been in talks with Nokia and […] to market that as a product. So it's not just about releasing value from your existing network assets by orchestrating them as part of a platform model, but it's actually creating new IP that can be sold to other operators and vendors.

Belinda Martin7:13
So we go to the next slide. And finally, this is kind of a schematic diagram of how that works in terms of orchestration across network domains and vendors, different server vendors, but also down through the layers of the network. And we're talking about here really about agentic, intent-driven orchestration being the sort of the gold star of that. So some of the orchestration historically has been very vertical, limited to a particular network domain, rather siloed. But the goal here for the industry to release that potential is be able to deliver intent flexibly, dynamically across network domains via AI agents under the control of the operator.

Belinda Martin7:13
So you have intent and policy orchestration going drilling down through the network, through the domains, through the standards, the RAN APIs, the 3GPP and essentially standards based virtualized and cloud-native network functions with a cross-domain orchestration layer for validating all of that, as well as closed-loop assurance and automation, but only if the underlying infrastructure is cloud-based. So here that's present in terms of telemetry and network state, which is obviously also extremely important, but that's enabled through the cloud-native substrate of the network to be able to deliver those network functions, scale them, port them across different environments and the like.

Belinda Martin7:13
So that's the end of our presentation.

Belinda Martin9:47
I'll hand over now to Nacho to present it from Red Hat's point of view.

Ignacio Gonzalez9:54
Thank you so much, Belinda. Amy, I'm going to need your flicker and a bit of synchronization because we have a couple of extra clicks per slide in a couple of slides. But again, thank you so much for having me here. I'm Ignacio Gonzalez or Ignacio Gonzalez. Both are fine. I am part of the Red Hat Telco Center of Excellence, and I am looking after the mobile network business across […]. And today I'm gonna talk about, I'm gonna follow up a little bit of what Belinda was mentioned, but I want to go back to the basics because one of the things that we are seeing more and more in the different conversation that we're having with customers is, if you want to really master on the AIOps and all of the […], Before that, you need to make sure that you are able to master elements from previous steps.

Ignacio Gonzalez9:54
So, Amy, I'll go to the next one, please. This is the current situation today for many operators. They have been deploying 5G networks, 5G standalone, and in the last two years, they are trying to figure out what to do with AI. What I want to say with AI is with these new United AI, agentics AI capabilities that telcos are trying to embed it in their network. What we are seeing is that those telcos that they are working on trying to operate both on over the same, let's say telco cloud platform, which has to be a cloud native platform, the traditional 5G standalone and all the AI apps on top of these are the agents over the same platform.

Ignacio Gonzalez9:54
They are getting, I'm gonna say a higher leverage of these platform. So if we can go to the next slide, please. Yes. When we talk about Cloud Native again, and let me go back to the basics. We talk about, when we talk about the CNFs, we are talking about microservices and the service-based architecture. We're talking about APIs. We're talking about containers. But now we are also talking about VMs. We are able, at least at Red Hat, we are able to manage over these Cloud Native environment because this is what's a request coming from Telco.

Ignacio Gonzalez9:54
By the way, it would not only to manage containers, but also VMs, and also this term, what we call at Red Hat at least the platform engineering. What is the platform engineering? If we go to the next slide, I will explain to you. Platform engineering, at the end of the day, it's two concepts.

Ignacio Gonzalez12:24
One concept is the GitOps, which has been very well known in the last four or five years, and now has been deployed and established by different telcos. in the last years. And it is a mix of two different components. The first component is the repository, the key where you have all the different code for the different elements of your infrastructure, the CaaS platform, and also the CNF. So at the end of the day, you're managing the platform based on code. And then you also have your pipeline in order to manage the life cycle management of the solution in the daily daily.

Ignacio Gonzalez12:24
So this is, as I said, the two different components of the GitOps. Then what we're going to have is the observability layer. So collecting all the telemetry and putting it all together in one single layer is what we're going to get here. This is the minimum volume proposition. If you are able to manage your infrastructure and your network as a code, and you have your pipeline ready, and you're able to see what's going on Thanks to the observability, you have this NDP ready for addressing this declarative framework for a telco cloud.

Ignacio Gonzalez12:24
Let's go to the next slide. I'm trying to select vendor diagnostic orchestrator or SMO just to avoid any bias. And here we have two different layers. One is more the actualization layer, and the other is the derivative layer. In a nutshell, I want to try to explain to you that you can see at the bottom all the different domains for the network functions, for the cast platform, and also for the infrastructure. You can also see there's a body layer at the bottom right. And at the left, you can find all of the gear and inventory, et cetera, et cetera, that I was mentioning.

Ignacio Gonzalez12:24
So if you click again, Amy, this is in a nutshell the footprint of this platform engineering. And at the top, what we have is the brain. Let me put it that way. And here is how, what is the level of sophistication that you want to have in your orchestration here. If you click again, at the end of the day, this is what we're going to have is the AIOps. Click again to the next slide. When we talk about AIOps, we are talking about three main components, which is something that is getting more and more common today, which is this telco language model.

Ignacio Gonzalez14:54
It's dedicated for telco, has been trained and tuned with synthetic data coming from telco networks. And if you go to Hugging Face, you can find on the benchmark for this type of language model, different language models. I think that the number one today is still […] from AT&T. We have been working on this with them, and as I said, that's the first state that you need to have. The second element is going to be the outcome of these models, the increasing and how you this in terms of, I don't know, grading your chats for help or if you need to get some information on what's going on in your network.

Ignacio Gonzalez14:54
Again, you can rely on them. And the third element here is the AI agent, where is, when things are getting, let's say, more interesting in terms of what you can get and what you can extract. And again, this level of sophistication is getting higher. So if we go to the next slide, please, we back again to this slide, And let's click one, two, another one, yes. And here is you have the six elements that I was mentioning. Something very important, the observability layer is crucial for this telco language model.

Ignacio Gonzalez14:54
So that's why, and we have been talking about this for a while, data, the quality of the data, and this quality telco data you can get is gonna be crucial for feeding your telco language model in the best way. And if we click again, another thing that is, or another element that is very important and I want to highlight it is this automation platform. Why? Because I believe that was mentioned, this is not just only a cloud native, the majority of this is cloud native, but sometimes we need to go beyond the cloud nativeness or Kubernetes footprint.

Ignacio Gonzalez14:54
And for that, an automation platform is gonna help us a lot. We can try to automate end-to-end different process, which are not affecting, which are affecting let's say the overall automation when we talk about a cloud native. and also because in the execution layer, if you think about the AI agent capabilities, the first thought was, okay, let's try it also to put the agents in the execution mode. Well, it could be possible, but also we have seen that we are still having certain limitations in terms of guardrails or control of these AI agents from time to time.

Ignacio Gonzalez14:54
So again, if you have already implemented and deployed your automation platform like Ansible, as we have at Red Hat, you can start leveraging on this automation platform for your execution layer.

Ignacio Gonzalez17:33
Let's go to the next slide, please. So all the sentences here are not mine. All of them are coming from our customers. And the most important thing is what you have at the title, which is automation is the foundation of these AI-driven operations. You need to master these automation ops, if you want to call it like this, before start moving toward these AIOps approach. As I said, you cannot train reliable models on chaotic manual ecosystem process. You need to create these quality telco data that I was mentioning before and process and standards if you really want to embrace the AIOps in a much better way.

Ignacio Gonzalez17:33
Moving from automated ops to AI-needed ops, it has to be done step-by-step. Once, as I mentioned, you master the first, you can move to the next one. The two sentence that we have at the bottom, they are for me crucial. Every manual step is a gap in your model. For us, even on Red Hat, we say that every model step is a back in when we talk about a Cloud-native network, because the level of complexity is getting higher and higher. Again, automation is going to help you to mitigate as much as possible.

Ignacio Gonzalez17:33
Then these manual ecosystem operations create noisy and unreliable training and data. Fantastic. We have started this journey. We started with automation and now we are embracing AIOps. What is going to be the benefit of doing all of these? If we go to the next slide, please. I'm going to try to explain you very quick examples of customers. Again, the majority of these outcome is coming from end-to-end automation. They are adopting now the first, let's say, let me put it that way, the first level of AI.

Ignacio Gonzalez17:33
But just only with automation, you can achieve fantastic things like a reduction in your FTE when you are building your network or 10 times reduction in having a faster execution when you're building your telco cloud platform from this, while your servers are deployed, install until you have your telco network platform ready for onboarding CNFs, or you also can reduce your organizational cost of your team, thanks to this automation. Again, these are examples of different customers that we have seen in the past. And as I said, we are expecting here even more operational efficiency once AI is adopted widely.

Ignacio Gonzalez17:33
All right, this is all what I have to share with you. So the next slide is just to say thank you. And I'm now really looking forward to joining the panel.

Ignacio Gonzalez20:03
Thank you.

Amy Cameron20:10
Excellent. All right, cool. I have stopped sharing the presentation and now hopefully all of our guests can now join us in the discussion. Thank you both Belinda and Nacho for your presentations and I think it'll be really interesting to hear from Alexis and from Terje a little bit what resonates with them sort of where they are on this journey as you say I mean obviously STL has presented quite a forward-looking view of what might be possible in the future and Nacho you brought us a little bit back down to earth but finished with some quite punchy statistics about what might be possible in the future.

Amy Cameron20:10
So maybe let's turn first to Alexis Salas. And if you could introduce yourself first, so everyone knows your background, but obviously coming from Orange, Spain, there's been a lot of movement there. So I think that, you know, you're in a unique position.

Alexis Salas21:03
Okay, so I'm Alexis Salas in Orange, Spain, in charge of the core network and automation. So in the middle of the topic of the today meeting, and we started this journey years ago with the virtualization, because all these promises was also in the virtualization before moving to the containers. And one of the difficulties that we are facing today is also the quality of the data. I was mentioning a little at the beginning of the meeting, but also the quality, because we were used to use this data in order to have a view, but not to use it to execute things in the network.

Alexis Salas21:03
So the quality of this data is very important and now becomes something that has to be managed in 24 hours per seven days, not only in our best effort approach, and also the reliability of the score of this data. That means that we can guarantee that this data is always correct every time because we have issues in the network, maybe in the props or bugs or whatever. So we have introduced this score in each data that we are using in order to have the usage of the machine learning or even the artificial intelligence to use this score in order to mitigate the impacts of some KPIs when they are not reliable to be used without having to move everything to the garbage.

Alexis Salas21:03
And for us, what is important in this and we discover is if we put garbage in, we have garbage out in the all the artificial intelligence movement. So this quality of the data is very important and not to have black and white. So to have this score in order to take in account this score in the process of this. So it was, and we have to collect all these data in order to be processed after by the artificial intelligence. And the second point that we are starting is what you mentioned for the coding as a service because we have a lot of manualities that is very difficult to manage in a network that is becoming more and more complex.

Alexis Salas21:03
As mentioned before, it's not only core network approach. Now we have APIs in order to open to the third parties. we have a lot of interactions with edge computing, with other platforms. So at the end, this trustability of what we are doing is very important and it becomes a journey that is not so easy because we have to move people that are today in a continuous upgrades of the core network to something that will be more to continuous monetization. But today, the reality is we are having a lot of people working during the night in the maintenance windows in order to upgrade the network.

Alexis Salas23:39
And we need to focus all this energy to provide new services thanks to the automation. But we are in the middle, so we need the effort and time for both parts. And we have to split this effort in the both journeys.

Amy Cameron23:53
Yeah, I mean, that's great, great recap of how you're approaching this. And I really liked, Nacho, your final presentation of sort of needing to get the deterministic capabilities right before being able to move to a probabilistic environment, which is obviously quite uncomfortable for telecoms operators. So I guess, Alexis, all of the scoring and what you're doing here is making sure that you really, really nailed down that deterministic piece before moving into the probabilistic. Are there any, like, does that resonate with you, that change? Can you see that probabilistic on the horizon soon, or like how far away do you think it is for you guys?

Alexis Salas24:33
So, you know, the journey that we're doing. Yeah, so we are trying to paint the boat on the same time we're sailing in the boat. So we are also using these first use cases in the IP change management, we have started because it's the easiest best in order to do this part before moving to the special core network. And we can see these benefits. So the challenge for us is to avoid to have this black process and to wait for years having these benefits. We need to create the steps and to take benefits in order to go to the following one, because the business requires also to have these benefits that can finance the second step and the following steps.

Amy Cameron25:18
Great, excellent. So Terje, let's turn to you. And the same, you can give a little bit of an introduction to yourself to those who are listening so they know your background and where you're coming from. And, yeah, keen to get your perspective on, you know, where Telenor is and whether you buy into this perspective that orchestration can create new value for your customers and the services that you're delivering.

Terje Jensen25:43
Yeah, well, thanks, Amy. So happy to be here. So I'm Terje Jensen. So I'm working in Telenor Group. Telenor Group is an operator, so we are in six different markets, 114 million customers. and there I'm in the technology strategy. I'm just leaving a position as head of security, but it's like once in security, you never leave security. That's another, but we can probably have another discussion about that. Then back to the question, I think, Amy, so not to repeat what Alexis was saying, but of course we are into the orchestration.

Terje Jensen25:43
We see the value orchestration increases from an operator point of view. Increases, it strengthens overall as we see it, both in terms of importance to be a quality, almost a hub in a partnership ecosystem. It brings value, and I think we've seen a couple of examples in the presentation already. And we see it strengthens the relevance, both for the customers and the partners. And in their way, I think the presentation worked through some kind of improvements or examples of improvements, for example, when it comes to speed to be most effective.

Terje Jensen25:43
we see also the opportunity when the customer can self-serve and also select on-demand which component they want to have and then even try something and then leave and change and so on. So it's a bit of giving the customer flexibility, it's also a value, it's a kind of differentiation in that sense and in that way you're also delivering resilience actually from their perspective. So there are a broader set of value components into this. And this is one way then for us to take a step what we call beyond connectivity.

Terje Jensen25:43
Of course, we are doing orchestration when it comes to connectivity assets, what Belinda walked through, for example, on base station, transport, mobile core and so forth. But we're also taking position, for example, say in an AI factory. And then when you do that, you typically start with, call it a green sheet, you know, the green, what you call the blank sheets approach. So you can actually start with a cloud native approach from the very beginning. So it's a bit more exciting than working with 30, 20, 30 years old systems and try to orchestrate that.

Terje Jensen25:43
So in the in the factory, we are, we are offering quality staircase of services, starting, for example, with the data center, GPU infrastructure, AI platform, AI models, applications, and so on.

Terje Jensen28:18
And then we said that, okay, customers can choose maybe three levels of this, like either a GPU as a service, you can do as an AI development as a service, or you can do AI modeling as a service, where we are putting in also a lot of different partners into this. And that's just one small example of how we then need to orchestrate this together, I want to dive in on that and understand a little bit to what degree the AI factory solution that you're offering and the network orchestration that you're doing, how much interplay is there between that in terms of you can deliver a better quality AI service because you can improve the routing on the network or whatever it is?

Terje Jensen28:18
Yeah, so right now, that's many of these orchestration approaches. We have a bit domain-wise, if I may say, which it probably is. So we are doing an orchestrator, for example, on the AI factory and we're also doing an orchestrator for, for example, the IP connectivity. Actually, the IP connectivity we have, it's not, of course, not AI native, but we have a self-serve orchestrator for the IP network, IP VPNs, for example, for many years. So that means that the customer or we on behalf of the customer needs to go into the different domains and connect them But there are different orchestrator if that makes sense.

Terje Jensen28:18
I mean, yeah But we are looking into but we are looking into exactly connecting that I think that's what you're opening up So we need to connect all these kind of things not only connectivity and AI factory but also potential security packages on top of that So there are quite a lot of this is this is then taking a role into this ecosystem where we go by own connectivity and getting partners delivering different components.

Amy Cameron30:06
Absolutely. So we have a question from the audience, thank you. And I'm going to give it first to Nacho and then maybe others will want to, you know, Terje and Alexis or Belinda, you'll have views on it. But it's building on your topic around observability, Nacho. And we have a question which is asking, can you build an autonomous system that delivers that observability without leveraging digital twins? Or in your view, you know, where do digital twins fit into that requirement for observability?

Ignacio Gonzalez30:38
From a point of view, a digital twin can help you in certain cases. So if you are trying to, but it's not something that is compulsory in a nutshell. So it's something that it can help you a little. For instance, if you want to check, okay, I wanna do an upgrade of my network And I want to see, with this digital twin, what is going to look like, the steps. And if I follow these steps, how is going to be the final outcome? Yeah, probably you're going to need to, you can rely on observability and probably you're not going to spend the middle of the night doing the upgrade expecting any kind of nightmare because you more or less can guarantee that it's going to happen.

Ignacio Gonzalez30:38
Everything is going to happen well. So in terms of, I will try to say in the other way around, if you really want to go for digital twins and you want to rely on digital twins to help you in your day one, day two activities, probably observability is a master thing that you need to have, rather than in the way around. So, but to me, digital twin is something that, I'm gonna say it's nice to have, and probably is gonna be able to reduce the nightmares during your operations, but observability is something that you have to have.

Amy Cameron31:53
Yeah, I guess I mean if you don't have any data to put into the digital twin then there's nothing in it. That's right. That's the challenge.

Belinda Martin32:03
Digital twins are not, digital twins at the moment are not really comprehensive enough and they they don't factor in all of the complexities and moving states of the networks adequately enough. I mean I think they're more, as Nacho said, And nice to have to reassure worried engineers that their proposed changes in the network are going to be safe. But it's not really a guarantee or a failsafe. And you do need the observability beforehand.

Amy Cameron32:36
Alexis, what's your view?

Alexis Salas32:37
Yeah, totally agree. We are focused today, our priorities, more the observability part. We have started also this digital twin journey, because it's quite long also to implement. And we are doing the step-by-step, layer-by-layer, but the priority for us is really the data and observability part, because all the decisions will be made on this data. And it's not only to have to close loop, even to have an open loop, we need this observability part.

Amy Cameron33:03
Yeah, yeah. And I wanna come back actually to one of the points that you were making before, Alexis, which was that you're hoping that you could reduce some of the business's usual work that your team is having to do, what are the things that you foresee being able to get off their plates sooner?

Alexis Salas33:23
Yeah. In the core network, what we are doing is to do an upgrade because in the past, when we have more metal solutions, it was only the software upgrade. Now that we are in a virtualized world, we have to upgrade the software of the network function but also the infra. We have two journeys to do. And the complexity is we have to remove all the traffic of the site, working on the site in a manual way in order to do upgrade in weeks or days. It depends on the network layer.

Alexis Salas33:23
And our first step is to have it during the day and to have what you call ISSU upgrade and to have the ability to remove the traffic for this site. we see that this site is having KPIs degradation, so it means that we have to have these KPIs with good quality in order to make the decision very quickly if this upgrade that is done during the day can have an impact also in the customer KPIs. So really it's this information about how is the behavior of the customer experience and to have this real-time view about the customer experience is very important in order to make these decisions very quickly in the network.

Alexis Salas33:23
And the challenge is really to have this upgrade of the network, virtualization layer and the network function layer in a way that is not having all our engineers during the night working on it in order to make the things work well. And we see operators.

Amy Cameron35:02
The ever-holy grail that when I've had a bad experience as an individual customer, or you could do something about it quickly.

Ignacio Gonzalez35:09
I have to say, sorry, I mean, just one point here. One thing that in order to reduce Alexis's nightmares, or at least working through the night, something that we are seeing more and more and working with the different NEPs, the outgoing NEPs is the in-service upgrades. We are working more and more to make sure that you don't need to rely for 100% of your upgrades during a night shift that probably you can rely more on Yeah, maybe some critical elements can have to be done by night, not just reduce the risk, even complete them during the day.

Ignacio Gonzalez35:09
And maybe we need to wait slightly longer to be 100% sure that this is gonna happen. But in the last two years, we have been working with the majority of the NEPs to make sure that we are able to offer to the, I mean, to Orange or to Telenor or other telcos teaching service as a way to reduce the, but like Alexis was mentioning before, having people working during the night is a waste sometimes of resources and also super expensive.

Amy Cameron36:12
Yeah. So I have a question, which is that, you know, thinking this day and age, like you, I don't get around it. I'm gonna put it to you first, Nacho, but given that you're playing in the AI world, maybe Terje, you'll have a view as well. So I think that they're trying to understand what is the cost benefit of analysis of using an agentic system versus a traditional sort of automation system. Obviously we all know that using the Frontier models and generative AI is more expensive. We're all trying to understand the cost benefit analysis there. I wonder if this is something that you've looked at at Red Hat that you have some views on already.

Ignacio Gonzalez36:54
Okay, so first of all, you don't need a Frontier model or checking your network. Right now we are working with a small language model. The one that I was mentioning before from AT&T, […], I think they have four billion parameters. So it's quite reasonable. So you can install in a laptop and it works perfectly well. And probably you're gonna need to have other models put on top of these in order to complement each other. Probably this question is coming from the cost of tokens.

Ignacio Gonzalez36:54
If you think about, I want to use AI AMs for doing the execution part. It's going to cost me so many tokens, versus if I have already an automation platform, maybe you should use automation platform itself. That's one of the things that we are also trying to see. You don't need to reinvent the wheel. There are many of the execution part of your network that can be done with your system tools. It sounds very great having 25 agents doing that job, but probably you don't need that.

Ignacio Gonzalez36:54
You don't need it first in terms of cost, as you was mentioned. And also maybe in terms of trustability of unguarded rails and make sure that they're not going to be very, they're not going to be very happy in terms of doing things that you're not happy with. Let me put it that way.

Amy Cameron38:20
Yeah. Yeah. Cool. Excellent. Well, look, we have so many questions. I'm actually going to give you a different one, Terje, which is maybe, you know, Given that Telenor has made quite a lot of progress in terms of trying to look at the different services that you can develop and it has obviously meant that you've had to change how you focus your teams from being purely focused on some of the day-to-day of managing networks and creating opportunity to innovate. So how has Telenor gone about that kind of operational and skills transformation within your organization?

Terje Jensen38:56
Yeah, just to be, you know, setting the conditions a bit right, so of course we are not, you know, 100% AI-ified already, so it's back to, you know, that we have quite a lot of legacy systems as well. I think Alexis pointed to some of the challenges on the data quality and quite a few things that need to be, you know, to get right first. And then, of course, focusing on some of the domains, you know, you can move forward. Like we say that this biome connectivity, then we have this AI factory, for example, as one example.

Terje Jensen38:56
We have several B2B offerings in the same. So, you know, focusing on some of these product areas or domains or whatever, how we want to slice the task. And then, you know, give the team the flexibility and the support to innovate and then use AI agent or other ways to actually experiment and explore what could be these and then of course, also working with the customers. So that's also a customer led, you know, innovation is part of the approach as we are following here.

Amy Cameron40:04
Yeah, I mean, I think that's always the, you know, one of the challenges for operators is to be able to give employees enough flexibility and autonomy in decision making. Because if you want to have innovation, you have to let the people who have the ideas who might come from anywhere be able to run with those ideas. And that's not always easy to enable in a large organization.

Terje Jensen40:26
Yeah, just a small remark, I think, Amy, we actually had a survey among all the employees in Telenor. So there are roughly 10 ,000 employees, and then we asked them, you know, how are you using AI? And 90% were using AI for their personal productivity, which is an important element, but of course, it's not these big things we're talking about here, but so it's important not to forget about that element of the AI. So it's actually helping a lot of the employees for being more efficient.

Amy Cameron40:57
Yeah, but I think that this is one of the challenges and maybe Nacho or Alexis or Belinda, you'll have a view on this, is that it is good that it is helping with personal productivity. But I think to the point that Alexis is trying to make as well is that personal productivity doesn't necessarily translate into your core KPIs. It doesn't deliver lower cost. it doesn't deliver, you know, higher beta, it doesn't deliver higher revenues. And the trap that operators need to not fall into is personal productivity is enough. It's like, how do you make the whole, you know, deliver, in a sense? I don't know, Alexis, if you have comments from your perspective on…

Alexis Salas41:44
No, no, I totally agree with you that it's two different things. And we have also the fear of missing out, no? And it's a lot of things that use artificial intelligence. I don't care about what, but do something. So we have to avoid to have this type of approach and to refocus on what are the benefit of the operations for the business and to have something that is for the company. And other topics, it's how we can improve the personal usage of the artificial intelligence, but for the network point of view it's totally different, yes.

Amy Cameron42:21
So maybe coming back to sort of our more original topic and I think let me put this first to Nacho and I'm sure Belinda will have a view as well, and this is to again your point of the deterministic versus the probabilistic use cases. You know, Nacho, if you had to put money down, how many use cases do you think really need that agentic probabilistic approach versus the more deterministic one in a network environment?

Ignacio Gonzalez42:36
Probably the root cause analysis. Probably you cannot find something way more productive if you go for the probabilistic. Then you need to double check if it makes sense or not. But in terms of finding root cause analysis in order to do some kind of for telcos for many, many years, so it's not something new. But in terms of these root cause analysis, when the overall platform, the overall network is so complex, I think it's helping a lot. Then deterministic is still using us today in terms of how you're doing upgrades, how you're doing your daily job activities, how are you planning to upgrade your network, because everything is under control.

Ignacio Gonzalez42:36
And probabilistic, yes. If you want to find efficiency in your network, probably yes, try to use a couple of agents and figure out what they're able to show you in terms of insights and outcomes. But as I said, that's for me the two categories that I'm gonna find in your daily activities, execution mindset, and operation teams, deterministic 100%, probabilistic when you want to see, okay, what else we can achieve, what else we can do, what else we can see from our network.

Amy Cameron44:02
Yeah, so I guess paraphrasing is in in the areas like you use the probabilistic when you've like hit you've hit the full optimization you can with the deterministic and then you use the probabilistic sometimes to see what are the unknowns that might be unearthed.

Ignacio Gonzalez44:20
Yes and one thing that we are working at Red Hat right now is we have been joined to […] because we believe that with the agents, there's a room for improvement and their potential is so high there in terms of how we can make sure that thinking about this closed loop with no human integration, it could be something feasible, I don't know when, but in the near time future. And we believe that there's a lot of potential there, but as we speak today, we are not there. So we cannot control much and we need to have these guardrails that when you're putting guardrails at the end of the day, sometimes you're killing this probabilistic approach.

Amy Cameron45:05
Yeah, yeah. Belinda, I know you've been doing some work on this, so what's your view?

Belinda Martin45:08
Well, I mean, I agree with Nacho, and it's important to have the controls on the sort of probabilistic approach. But ultimately, as the technology improves, as the data improves, as the ability of the AI agents to factor in network contexts and changing network state and so on improves, There's going to reach a point where there's an extremely high probability that the changes promoted by or advanced by the AI agents are going to be more productive than what could be determined by a more deterministic approach. And I think that's more in the area of optimizations, root cause analysis, but also network evolution, defining service intent, actually doing things that network engineers and product development teams haven't thought of.

Belinda Martin45:08
So there comes a point where you have to sort of make an equation, you know, what point of risk and probability is acceptable, even for major changes. So we put it in the terms of, you know, five nines reliability versus five nines probability. So five nines probability means there's an extremely high chance that this fix or this optimization will be good. An extremely low probability that it will have negative consequences. So that's quite abstract. But that's the kind of risk equation that will probably come into play as the technology improves.

Terje Jensen46:40
So can I maybe offer a further complication just to be a.

Amy Cameron46:46
Of course.

Terje Jensen46:47
No, because I think it's important to remember in this AI phase where we are right now that, for me at least, autonomous means that you have automated things, which is probably a bit deterministic, but you're also aware of what's happening, and you also are able to influence the systems. And in many of these, say, fault cases, what we are seeing right now, there are so strong dependencies between different domains, maybe don't have insights into all the things. So instead of waiting until you understand the full picture, you start doing things.

Terje Jensen46:47
And then you start doing things by changing things, and you're measuring the impact, back to the observability. And that you can actually do quicker than in that deterministic way, so if you do that. But you need to have the guardrails in place, of course, like Ignacio was saying. Because we are, at least in connectivity, we are working with critical infrastructure. So you're not playing, you're not, you know, we are doing business here, but you could be quicker, and you could be on top of that by being a bit more, you know, dynamic, and that in many of these scenarios, I asked for being quicker.

Amy Cameron47:57
Yeah, it sounds like you're sort of bringing your mindset towards accepting the probabilistic versus the five nines reliability that we have today. So, I want to come back to sort of the orchestration piece of the story. We have, you know, a great question, which, you know, it's extremely valid, that efforts like ETSI MANO and ONAP and what have you have all been going on for years and years and years we've been talking about orchestrating across multiple network domains for a long time. You know, what do we think is different now than maybe two or three years ago that might to make us closer to that goal. You know, who wants to start?

Ignacio Gonzalez48:40
I can start. And the biggest difference that I have seen from the ETSI MANO all times to the new cloud native automation approach is not one big box coming from one vendor. That is the one who is controlling the automation here and there. And you need to, again, blind funding and trust on that. I think with these new models, with the nephew architecture and with this Cloud Native Automation architecture, what we are seeing is you're starting with something smaller that you can control and it's up to the telco, it's not up to a vendor putting in place these operational framework.

Ignacio Gonzalez48:40
Then as I was mentioned during my presentation, the level of complexity or how much you want to over-engineering the orchestration and being, it's what is gonna be maybe when you need to rely on other companies or on an AI, then you put it that way, but at the end of the day in other companies. But one of the things that we have seen a lot is now is not a bit about big boxes coming from somewhere else, it's about a smaller domains that you can control.

Ignacio Gonzalez48:40
By the way, it's standardized and it's based on open source projects and open standards that has helped a lot to again, having a common ground for many telcos.

Amy Cameron50:12
Alexis, Terje, I don't know if you guys also have views on this.

Alexis Salas50:15
Yes, I totally agree with Nacho. We are not now trying to have a single orchestrator that orchestrates everything. We have different orchestrators that are orchestrated at the high level in order to have this interface with the business. But the idea is we are using the orchestrations more linked to what we want to do that are managed after an orchestrator at a higher level. So an example is the NSO Cisco that we are using for the IP layer. And after, we have service director coming from HCL for other part of the network.

Alexis Salas50:15
So we are not launching the RFP to have the universal orchestrator that will bring everything with a high cost and very difficult to justify in a business case and so it's more an approach use case by use case but always in a framework and we try to have this framework in order to avoid to have silo solutions that are not working together.

Terje Jensen51:17
I think I also fully agree with both of them so I was worried that I think actually AI and AI agents have helped us and get started. Not taking it as a big step like Alexis is explaining. It's not the master of the master or the orchestrator, but it's actually slicing that into more focused and dealing with more focused issues. And then you can also safeguard the actions around that one. I think that's given a bit of a trust into the whole setup as well. So it's not like a big leap and then hopefully it delivers some returns. You can actually start doing something with a much smaller step and learn from that and build on that.

Amy Cameron51:58
Yeah, yeah. And I guess maybe underlying some of this sort of ability to iterate more and be less stuck to a monolithic system. You know, for me, the the implication is also that telcos are maybe bit by bit, bringing some of that development capability more in house and that enables them to be a little bit more diverse in their suppliers and do a little bit more of the iteration themselves. And, you know, maybe that's been the benefit of all of us are more nimble with technology, with AI at our fingertips.

Terje Jensen52:34
Yeah, and it is, but maybe I can go to challenge some of the guys. But it's like, you know, maybe maybe 10 years ago, we were discussing, you know, the cloud, is that something for us? Maybe 15 years ago. And then we realized that quite a few of these partners we want to collaborate with, They came with expectations that we already have a cloud platform. And that means that, okay, we actually have to embrace this. And now we are in this stage, I think, on orchestrator and AI agent. We probably need to be on that level. Otherwise, we don't have any partners who want to collaborate with you anymore. At least there is a risk.

Amy Cameron53:07
Yeah. So I want to finish on a question here, which sort of tailors to the creating new value piece of the title of our webinar. And maybe you can kick us off, Terje, because you mentioned that you're working with some customers in terms of what you're delivering and you're delivering more flexibility. So maybe you can tell us a little bit more, particularly on the network flexibility side, what are customers interested in? What are they looking for? What are you seeing that they are really using from a more flexible network?

Terje Jensen53:40
No, in short, I think, Amy, I think they won't see a menu which they can choose from. So they want to have the flexibility, but they also want it to be fast so you know they can try try get you know tick off this and then try is this something for me or should I change something else and then of course in many of the cases we are working on this the connectivity and this what we call […] is becoming a key component in their business or it could be in a mission critical for example on the defense and the emergency services but but so it needs to work so that means that we're coming back to the servietability and assurance needs to be part of it as well.

Terje Jensen53:40
But they need the confidence that, you know, this is working and they also want to be in control at least to be aware of if something is happening. So, working together with several customers, which are on lead on what they're doing there, I think it's very giving us a lot of insights, you know, in what are the challenges. And then to remember that, of course, it's not only technical systems, it's a lot about people and the processes and the business models definitely on doing this. But again, back to where we started, customers are basically looking for flexibility, becoming more efficient, having higher speed, and also being resilient in all aspects.

Terje Jensen53:40
Great. Excellent.

Amy Cameron55:03
I don't know, Nacho, Alexis, Belinda, whether you have any final comments for us because we're about to wrap up. I think we've done well answering everybody's questions. Thank you everybody for giving them. Maybe Alexis first?

Alexis Salas55:14
Oh, yeah. Coming back to what Terje said, we see also that the customer asking to have the network before was something that you asked something to the network and you have days or weeks in order to have the feedback, OK, it's implemented and now they want a portal and to have these things in real time to be able to create a slice or to associate a scene to a slice and things like that. And there's computing. So really, this automation makes up the network. And also, we have an open-gateway approach.

Alexis Salas55:14
CAMARA is a way also to propose APIs where the network is not a black box. It's something that interacts with the developers. And so we have features in the network that can be activated by this API and the information that is provided from the network to this developer. So it's not a black box anymore and we need to have this automation, basically an API wants to have something in real time and not in different places.

Amy Cameron56:15
Well, that is music to our ears. I'm sure that Belinda will join me in saying that, you know, we've been looking forward to this future from Telcos for a long time. So thank you to all panelists today for sharing your thoughts, sharing your experiences, answering everybody's questions. I'm sure if there are any that we didn't get to today, then I will pass them on to the panelists and we'll see if we can get back to you afterwards. So thank you, everybody, for joining us for the conversation. And once again, thanks to the panellists. Thank you. Thanks for coming through. Glad everyone enjoyed it. Bye bye. Bye bye.