Guy Daniels, TelecomTV (00:27):
Hello, you're watching TelecomTV and our special programme on how rApps are accelerating the journey to autonomous RAN operations. I'm Guy Daniels and today's discussion looks at how rApps are rapidly becoming the foundation for advanced RAN automation and progress towards higher levels of autonomous networks. A 2026 global survey by Analysys Mason of Tier 1 operators revealed the growing role of service management and orchestration hosted rApps, Agentic AI and intent-based operations. And joining me now to discuss the report's findings and the enablement of autonomous RAN deployments are Caroline Gabriel, research partner technology with Analysys Mason and Claudia Muñiz, Head of Market Development in Cognitive Network Solutions at Ericsson. Hello, it's good to see both of you and thanks so much for taking part in our programme today. Now, the Analysys Mason report highlights a strong ambition amongst operators to reach autonomous networks level four in 2030.

(01:44):
So Caroline, if I could ask you first from the conversations you've had with operators, what are the strategic motivations behind this push?

Caroline Gabriel, Analysys Mason (01:54):
Yes, Guy, it is extremely strategic and we shouldn't underestimate what a challenging ambition this is to get to it by 2030. So it shows the importance to operators of getting to this new level of autonomy. I think like most efforts to achieve greater automation, the starting point is efficiency. We know operators are under huge pressure to reduce costs, especially operating costs, and their network operations have become steadily more efficient over the years in terms of cost, energy, just general responsiveness. But those gains are often outrun by the increasing complexity of the network. 5G introduced huge numbers of new elements into the network, new functions, new processes. So really we need a step change now rather than just a gradual improvement and level four promises that, but we shouldn't think this is just about efficiency. Most operators tell us that to justify all the investment of time and resource and the disruption that it takes to get to this very advanced level, they need other commercial justifications.

(02:57):
So they're looking for an autonomous network to support new services that perhaps can't be effectively delivered with manual or lower level autonomous processes, certainly looking for quicker launch of new services. And perhaps most importantly, they see this level four as a way to significantly improve the quality of experience across the network in a greater responsiveness aided by AI, which we'll come on to later. So in the end, this is about remaining competitive, being able to deliver a quality service at affordable cost, efficiency in other areas such as energy. And that's really foundational for an operator to compete against its competitors.

Guy Daniels, TelecomTV (03:47):
Thank you, Caroline. And as you said, it's a challenging ambition. And from your perspective, Claudia, what are the challenges that CSPs are facing during this transformation to autonomous networking?

Claudia Muñiz, Ericsson (04:00):
Yes. So to me, the TM Forum autonomous network standard is like a map. So you have all the processes mapped as high-value scenarios, you have the different levels of maturity. So you have mapped everything that there is to do, but where do you get started? What's the icebreaker use case that's going to give them maximum return so you can build on it and how you go from A to B? That's where I think sets different CSPs apart now. So some CSPs have a very clear plan and very realistic one and others are a little bit still figuring out. So actually this week we welcome a Tier 1 operator group here in Málaga where we have the headquarters for our AI-native rApp development and actually their responsible for autonomous network automation transformation shared some key insights. So actually they started in 2021 with a very comprehensive approach across all the domains and all the processes and the average maturity across different countries.

(04:58):
It was slightly above one. And then last year they were already at 3.4, with some countries around two. And just like Caroline said, the main point of all this is actually growth. The efficiencies come along the way, but they don't do it only for that, right? And of course the major challenge is operational. So we talk about AI technology because it's very cool and because maybe it's something that is easy to point a finger on. I think reality is about the process transformation, it's about what we stop doing, it's about the people trusting the technology and it's about uncovering values that we didn't think about. So they spoke with us about having automated fibre design and of course they saved an enormous amount of time in doing so. But what was interesting is that they also reused a lot more fibre.

(05:48):
So then the overall CapEx investments reduced significantly. And then another important point that really caught my attention was the law of diminishing gains. So he was very clear that, okay, over the last four years we came across three levels, but now to go from 3.4 to four is going to be very difficult because here is when we need to introduce intent and it's when we change the paradigm from being with human in the loop and to some extent guided by human into fully guided by the machine. So that's why as all of our AI-native rApps, they are equipped with explainability so that all the engineers trust and understand the recommendations. And then the last thing we also debated is like which path you take now. So there are some propositions in the market that are about a vertically integrated stack with a very clear promise that you will reach level four here, here, here.

(06:42):
And then other vendors, including Ericsson, we believe more in a horizontal layer approach where the different layers for resource, service and business are interfacing through open standards. So this, and we acknowledge together with him that this might appear as slower because you are integrating potentially solutions from different companies, but over time, in our point of view, this is more future-proof and lower risk. But in our conclusion, I think everybody realises that it's not going to be easy, it's going to take time, but also we cannot continue like we are today.

Guy Daniels, TelecomTV (07:17):
Yeah. Thanks very much, Claudia. A lot of steps for operators to get through. And Caroline, your study suggests that operators are moving beyond traditional SON towards SMO and rApps. So can I ask you what you think are the main operational pain points driving that transition and where do you see operators get the biggest and most immediate value?

Caroline Gabriel, Analysys Mason (07:41):
Yeah, we have to remember that when SON was first developed, a lot of the technology enablers we now have, they were absent or very immature. We talked about intent, we'll talk about AI. But basically I think we ended up with SON that lacks flexibility. It didn't support machine learning, so there were limits to the level of its intelligence and its automation. And generally these were not open systems. So those three things are what we found in the survey are really driving the move to the next phase, which is largely encapsulated in SMO and the rApps. So those three things, flexibility, machine learning, openness—and that's really driving a shift in the whole approach to RAN automation. rApps are very important here. There's a strong link in the survey between rApps-based automation and the ability to achieve those higher autonomous network levels, to move to three to four, one day to five.

(08:43):
So really I think rApps have moved quite rapidly in the last couple of years from being something that companies were interested in, were testing, to the primary vehicle for supporting new levels of automation, and the ecosystem around them is maturing very quickly. I mean, by our calculation, the volume of automation that we think will be delivered by rApps in the RAN will more than triple in just two years and triple again if you look at the next five years. So there's a huge drive to move beyond SON to something that addresses a lot of those pain points. And we have to remember, it's a moving target. There'll be new pain points, networks getting more complex when we're thinking about what 6G may look like. So we think the rApps platform is a lot more future-proof as well because of its flexibility and openness.

(09:41):
I mean, in the survey, it was very striking that the largest percentage of the respondents, when they're asked why they would adopt rApps and why move ahead from SON, the top response was that they need end-to-end automation and orchestration. And two-thirds of operators said that that was a top three concern and driver for them. And 38% said it was the number one reason to move from SON to rApps. And there were other big drivers and a lot of these are about filling the gaps that there are in SON. So the second most important driver was the ability just to support more complex automation use cases. This is all about being able to deal with complexity, and that complexity increasing by the day, and openness also important. The third most important driver was that rApps and the whole SMO system provides open and standardised interfaces and control rather than being siloed into what may be a very good automation system that closes you into one supplier.

Guy Daniels, TelecomTV (10:54):
Well, it's so encouraging to hear about this real rapid increase of interest in rApps there. And one of the clearest messages from the study is that operators increasingly see rApps on SMO as their preferred path to more advanced RAN automation. So Claudia, why do you think this shift is happening now and how closely does it match what you're hearing directly from your operator partners?

Claudia Muñiz, Ericsson (11:21):
I think now everybody has come to the point of accepting that you can't keep doing the same thing and expect a different outcome, right? So we have differentiated connectivity coming in. We have 5G standalone, active antennas, different performance requirements for different slices. This creates a level of complexity that you cannot handle manually, you cannot do with rule-based tools and especially you cannot do with things that operate in silos like SON or traditional optimisation tools, right? So think about active antenna systems. You take a cluster of 500 sites. If you want to configure the optimum cell shape, meaning in horizontal, that is about 96 to the power of 6,000 different possibilities. If you do the optimal per site, you will have a great return on your network investment on your spectrum, but this can only be solved by machine learning. So that, I think for customers, typically is three things that, like Caroline said, SMO is a unified data layer that allows to materialise AI at scale, network-wide scale, closed-loop, fully autonomous.

(12:28):
It's a single point of data in the network activation. Think about who controls the automation layer controls the network. So it's very important to have security, to have coordination, to have guardrails, all these kinds of things. And then also because I think for the first time in the telco industry, we have very solid standardisation around the OSS space that really enables a standard ecosystem of innovation on a global level where anyone can develop rApps based on R1. So we come from a world where we have these use cases from SON that are coming down from 2009, right? Then that were complemented by scripts that the operators developed on their own, but they were all in their own little silos, like Caroline explained. We didn't have a global app store for the RAN, right? So now with SMO, they see it as a possibility. And I think now it's not only SMO and rApps, but also TM Forum that, like we were saying before, is bringing a roadmap for a transformation.

(13:29):
So a question that we receive a lot is, okay, how do these rApps help me reach network autonomy? So that's why we have leaned on the TM Forum online methodology to assess what is the level of maturity that our applications provide in different high-value scenarios. So for example, for RAN quality optimisation, we are now between 3.4 and 3.9 depending on the use case, and then when we introduce intent, we will reach level four, right? So it's about having this kind of tangible proof that, okay, with this, I can do something that with SON I couldn't do at all and how is it bringing me closer in this journey of automation?

Guy Daniels, TelecomTV (14:10):
Thanks very much, Claudia. Now you've both spoken about AI and intent. So let's focus on this now because the report also shows real strong momentum behind Agentic AI and intent-based networking with many operators seeing them as critical to reaching levels four and five. Caroline, how are operators describing the practical role of Agentic AI in real network operations beyond the usual sort of marketing buzz and hype we hear about?

Caroline Gabriel, Analysys Mason (14:42):
Yes, interesting question because Agentic AI obviously is sort of at peak hype now, but operators are also really trying to look beyond that and think quite hard and deep about exactly how Agentic AI can help in a practical sense, where it's strong, where it needs help from other technologies. But amazingly in the survey, a whole 90% of respondents said that they saw Agentic AI as either significant or absolutely critical as an enabler of getting to level four, future level five autonomy and intent-based networking. Even allowing for some hype and some overenthusiasm perhaps at this stage, this is a really high percentage. And as I say, operators are starting to think, well, what does that mean practically? Why will it make such a difference? And critically, Agentic AI follows the closed-loop automation chain, which we've talked about throughout the history of RAN automation, but it really helps to get to that.

(15:45):
So the loop of perception, analysis, predictions, then it does decision-making and that leads to an executive action, and all that can be done automatically and intelligently with the help of Agentic AI. It doesn't magically do it all by itself. It has to be playing a complementary role to GenAI. So GenAI is very good at interpreting high-level intent, for instance, and translating that into tasks that agents can then go away and execute. So there's a lot of interchange between different types of AI and of course, other types of tools, but together these can really enhance the autonomy of the RAN within the rApp framework, particularly coordinating those rApps, helping them to optimise operations under very dynamic network conditions. The network changes by the second, as we know, applying more and more intelligence to those real-time decisions. And as a result, we're seeing from the survey that 94% of operators plan to deploy Agentic AI.

(16:57):
So even more than the 90% that said how critical it was, just about all operators aim to deploy Agentic AI, particularly and initially to support orchestrating multiple rApps. That particular use case they're going to look at within the next two years and then add other functions and processes that Agentic AI can really enhance going forward.

Guy Daniels, TelecomTV (17:19):
Wow, those are really high percentages there. That's very encouraging. And Claudia, if I can follow up with you, what lessons has Ericsson learned in its current engagements with CSPs?

Claudia Muñiz, Ericsson (17:30):
Yes. I mean, first of all, without the intent, there is no Level 4. So I think this is clear. We have run the initial test of our intent management function. And I think the key is to understand that the value of all this Agentic AI and so on will not be to replace the network experts, but to scale up the impact. So the key question that CSPs always ask is like, okay, so now we start with, let's say, five rApps. What happens when I have 30? How is this all going to work? And that's precisely where the intent management function plays. So you have the service layer or a human providing an intent: let's maximise the energy saving while ensuring a certain SLA on these network slices. So then you have this function that understands this intent, understands the overall purpose and orchestrates this growing number of applications into coordinated autonomy.

(18:21):
And just like Caroline explained it very well, with closed-loop rApps, we start open-loop. So we look at what recommendation the application is providing, make sure that the engineer understands and then trusts the action. So step by step, we go implementing, step by step, we close the loop. So now intent and Agentic AI is the next level, right? So with networks being the backbone of society, to have this banner that we see sometimes in the LLMs, 'bear in mind that AI outputs are not fully correct', it's not good enough. We need to have transparency, we need to have guardrails, we need to have explainability. These are mandated by the customers as prerequisites to introduce this technology. And that's why together with Agentic AI, Intent, rApps, and SMO, we also introduce adoption services. So this is very similar to this concept of the forward-deployed AI engineer.

(19:16):
So we will have our rApp experts work with the top experts to work with the CSP every step of the way to understand not only how this technology works and why it provides recommendations, but also how to change the process and trust so that we can fully unlock the value of AI.

Guy Daniels, TelecomTV (19:32):
Okay, that's good. Thanks very much, Claudia. And some of these Agentic AI cases seem to make use of natural language interfaces in a kind of "talk to your network" fashion. Caroline, has this been addressed in your recent report?

Caroline Gabriel, Analysys Mason (19:47):
Yeah, definitely. I think this is really important. This need to simplify how people interface with automation. When we look at intent-driven networking, that starts with very high-level objectives, which often may be set by very non-technical people and teams. And also we're not looking to exclude humans with this autonomy. The idea is for them to be adding their value, but there's no real reason why that value should only come from technical people with very specialised skills on the RAN. There's a contribution that non-technical people can make, but of course they have to be able to interface in a non-technical way just to be able to ask questions or express their objectives, their intents in normal language, then that can be translated into much more specific technical targets and language as required. Operators in general are really viewing natural language interfaces as a really important enabler for everyone to interact with the RAN, but also to improve how the specialists, how the network engineers and the operations teams interact with these networks.

(21:00):
Because as I said before, they're increasingly complex even for specially trained people. So they need to be able to access the data, solve the issues much, much faster, and that's quicker and easier when you can do it in the form of a conversation or direct instructions without having to learn very, very specific languages. So the benefit of all that, of course, is that it improves efficiency. Another really high percentage here, 90% of operators, they said that that would be one of the top two benefits of using a natural language interface would just be overall efficiency, much quicker response to problems and challenges in the network and enabling existing capabilities to be extended much more quickly. And this is important. As I said, it's always a moving target here with automation. So if you can talk in your own language, it enables new functions, new processes to be introduced much more quickly because people don't need to learn lots of extra specialised programming.

(22:14):
NLIs, they also reduce the complexity of integration. They can automatically handle things like translating tasks between different network platforms or different data models. So you don't have to develop a custom API every time for every aspect of the integration. And generally, I think this is also part of a much bigger process of democratising AI, of everybody in an organisation being able to use it, to interface it so that all their skills and expertise and objectives are reflected in the end result.

Guy Daniels, TelecomTV (22:50):
Yeah. Thanks very much, Caroline. And another important theme that we've already touched on in this discussion is openness. Operators want access to a broader rApp ecosystem, including third-party and in-house development, not just a single-vendor model. Claudia, from the operator conversations you've had, how real is this demand and what does it mean for how the ecosystem must evolve?

Claudia Muñiz, Ericsson (23:14):
So I think from the Mobile World Congress announcement that we did, where Ericsson and Nokia joined each other's ecosystem of rApps, right? We made a very clear stance that when it comes to SMO, openness is the foundation and like I always say, the best man wins. So I think this is one of the tricky reasons to move to SMO for the CSPs: freedom, right? They have freedom to develop, they have even freedom to commercialise if they would like their own applications and to tap into this global ecosystem of innovation. We've seen it with the smartphones, right? The acceleration and the speed and the value that we generate when we truly open up is massive. But now in the real world, you might have seen recent press releases with AT&T, MasOrange in Spain and KDDI in Japan where in field activities we have full coexistence.

(24:06):
So it's not either-or, it's both. We have CSP rApps co-existing with third-party rApps, co-existing with Ericsson SON. So we position our AI-native applications in very differentiated use cases where the scale that we have in terms of operating across many networks in the world, the solid data and the investments on AI research and also the domain expertise when it comes to pure radio and core really provide a unique value that is then complemented by our developments. And I think that also has an important implication for the ecosystem. So earlier in my career, I worked in standardisation, right? But SMO is a lot more than just a standard interface or an SDK. It's about having a platform that supports development, that supports lifecycle management, that supports testing, that supports governance, that supports the whole operationalisation cycle, not only just a claim on a PR.

(25:05):
So practical openness. So that's why from the moment we started with Ericsson Intelligent Automation Platform, we put a massive investment also on the ecosystem and on the developer portal where the CSP, third parties and even ourselves—because believe it or not, we use exactly the same SDK and exactly the same set of tools as any other player—can innovate faster. So it's about providing a great developer experience with all the latest tools, a forum, an AI coding buddy, training, our templates, everything that is available for free to all our members because we really believe that together we can multiply the value so that one plus one is more than two, and this really makes this solution also the platform for automation in 6G.

Guy Daniels, TelecomTV (25:53):
Great. Thanks, Claudia. And Caroline, related to this question of openness and also the rApp ecosystem, you ask in your investigation about operator expectations for sourcing these rApps. So where do they plan to get them from and what do you think the answer's going to signal in terms of a longer strategic direction?

Caroline Gabriel, Analysys Mason (26:14):
Yes, and indeed our survey shows some quite striking results on this topic. It says that about 78% of operators currently are still buying their rApps directly from their existing RAN vendor, either the vendor's own apps or the vendor may be sourcing third-party partners. But in just two years we see a dramatic shift. Only 2% of operators think that their primary route for buying rApps will be their RAN vendor and 90%—that magic stat again—believe that their primary source of rApps will be from multiple third parties, and those could be different types of providers. They could be specialist rApp developers, they could be broader independent software vendors, but a huge effort towards buying usually from a marketplace from multiple third parties. And then if we look five years into the deadline that operators have set for getting to level four, we can see another shift because by then 76% of operators say that they believe they'll be developing at least some of the rApps themselves, using their own resources, using their own tools or open tools from the marketplace.

(27:36):
So again, another shift to operators seeking differentiation by developing at least some of the rApps for themselves.

Guy Daniels, TelecomTV (27:44):
And Claudia, do you want to add any insights from yourself about how Ericsson is helping develop this rApp ecosystem?

Claudia Muñiz, Ericsson (27:52):
Yes. So we are very proud because we have the leading ecosystem in the industry. So today it has more than 100 members because it's something that, like I said before, we nurtured from the beginning, right? But what excites us the most is really to see the practical examples of how this is alive. So in Mobile World Congress, and actually rAppCon that we had—a global developer conference two days ago with 53 companies across 23 countries—we are seeing real examples of rApps being developed, the value created and so on. So it shows that this kind of openness really has moved now from idea or theory into practice and becoming really this operating model in this global app store on how automation will be developed and consumed.

Guy Daniels, TelecomTV (28:40):
Great. Thanks very much, Claudia. And I've got a final question I'd like to ask both of you. The study also points to a major commercial and architectural shift towards SaaS delivery and greater use of public cloud platforms to support AI-driven automation. Caroline, when you spoke with operators, do they see this mainly as a business model decision or is there something else going on here?

Caroline Gabriel, Analysys Mason (29:03):
Yes, I think it's more than that. I mean, there's been a sort of steady move towards software-as-a-service delivery and OSS software for a while and generally that's been primarily a financial decision, operators wanting to spread their payments to move from CapEx to OpEx. So really all about their own financial agility. We're seeing something different or bigger here related to autonomous networks, I think, and a lot of it is about speed. I talked before about how one of the really important things about these new processes and the levels of automation is that networks can react so much more quickly to problems, to challenges, or just to a change in the network or a user that requires something very specific. So I think the move to SaaS now has been very heavily driven by being able to react more quickly and, for instance, to get upgrades to software in a much more ongoing way.

(30:08):
So it's constantly upgraded and updated rather than just having a great big upgrade once a year or two. So definitely looking beyond financial restructuring towards the move to what some people call live software, really the software reacting as a living, breathing thing, responding to change almost in real time. So RAN automation, we know it's becoming increasingly dependent on real-time access to large language models and also to cloud computing resources. Putting it in the public cloud and having delivery with SaaS makes all that much more flexible, much more immediate. 72% of operators in the survey expect that pure SaaS will be their primary licensing model for rApps as they move towards a level four. And that's big because the move to SaaS has been gradual, not particularly rapid in other areas of OSS. So this is quite a significant change. And the operators told us that the top benefit is that real-time access to the latest innovations.

(31:21):
They're always up-to-date. They can access the latest models, the biggest computational resources that they need. And all of that reduces operational complexity and also CapEx enables them to get to level four with the cost more related to actual usage and business requirements rather than a huge great CapEx budget. So the financial side is still important, but really it's about the flexibility, the speed, the responsiveness that the SaaS model and of course access to public cloud resources can bring.

Guy Daniels, TelecomTV (32:04):
Great. And Claudia, a final word from you. What's your take on this from your experience of meeting CSPs on a regular basis?

Claudia Muñiz, Ericsson (32:13):
Yeah, so actually even before the hardware prices went significantly up because of this AI rush that we are living, we've been focusing on software-as-a-service. So in February, we launched rApp-as-a-Service at our AWS marketplace offerings where we combine AI models, infrastructure and operations all under one price, which is based on actual usage. And for CSPs, it's very exciting because it's a very fast way to evaluate the field value by using the rApp-as-a-Service together with the SMO, with the R1 standard. For me, having been in telco my whole career, it's very exciting because we bring the best of both worlds together now. So we have the telco standardisation, the domain expertise, the resilience, and then we bring the public cloud elasticity, all this AI massive potential and the speed like Caroline was saying. So even if the operating model that we have for our AI-native rApps is already DevSecOps with the fast ML model maintenance, security and stability,

(33:12):
we see that being on AWS amplifies really our capabilities with using the large language models to enable this talk-to-data capability and also Agentic AI at a whole different level. Some advanced operators have already moved all their data workloads into the public cloud. They are developing their own agents. So to have this Agentic AI interface all enabled in the public cloud is very, very attractive. So that's why for CSPs that are really into the public cloud journey, this is really seen as a fast track to materialise Level 4 autonomy on the resource layer and a very, very interesting space that is developing, by the way.

Guy Daniels, TelecomTV (33:49):
Great. Well, we must leave it there for now though. That's all the time we have on this programme. Claudia and Caroline, thank you both very much for taking part today. And for further information on the topics covered in this programme and to watch other videos in our OneVision Multiple Views series, please click and follow the links in the text below. And you can also download the Analysys Mason report on how rApps are accelerating the journey to autonomous RAN operations. Just click on the link. For now though, thank you for watching and goodbye.