Mark Düsener, CTIO of Swisscom, talks to Annie Turner, Contributing Editor, FutureNet World about his award-winning, domain-by-domain approach to autonomy, new services and revenues, the importance of changing habits in the implementation of AI, and the impact AI could have on networks

Mark Düsener became CTIO of Swisscom in September 2024, having previously been EVP Mobile Network and Services. At DTW Ignite in June 2025, Swisscom won an Excellence Award for Impact and Innovation – see Appledore Research’s case study, Swisscom Autonomy Blueprint. It describes the operator’s approach to achieving network autonomy (network automation and closed-loop control) which started with operational autonomy at the domain level, decomposing the OSS into 30 autonomous operational domains, based on TM Forum’s Open Digital Architecture (ODA) framework.

What inspired this approach? Düsener says, “I’ll start with the core beliefs that drive us as a company. We need the ability to adapt and go fast in certain areas which will become more important as innovation is happening at a much faster pace. To cope with innovation coming from outside as well as providing innovation yourself, you need to speed up. We need adaptability and autonomy for multiple dimensions and it starts with an architectural approach.”

Grand scheme architectures do not have a high success rate in telecoms, which is tacitly acknowledged by Düsener who explains, “We made sure the architecture isn’t a monster but component-based, using the principles of TM Forum’s ODA. The same applies to teams and organisations.” The goal is to ensure each architectural component is sufficiently decoupled to allow a dedicated team to work on it without impacting others or being impacted by changes to others.

“This means you don’t need to wait for another team to do something to move forward. That’s the principle approach that will help us become faster and more adaptable,” he continues.

The dream of autonomous networks

Düsener says looking forward, “We all dream of autonomous networks [AN], but most likely they will be the partially run by AI agents and agents need three components. The first is data, the second is context, which is a different form of data, and third is APIs – without APIs your agents cannot act. The decomposed architecture is based on APIs – everything you consume, you consume as a service via APIs which determine every service, every resource, every ODM [operational domain manager], needed to deliver certain data inventories. Assurance by APIs allows us to scale, otherwise you need to adapt to every single service, and you need holistic governance.”

Does Düsener see AN less as a technical environment and more as an organisational issue? It’s a matter of applying Conway’s Law he says: “Create an organisation that looks like the architecture should – its to-be state – and then you get the architecture. This is what we did in the end.”

The philosophy applies to all kinds of AI too. Düsener says, “AI is still at its beginning, so we are just starting to grasp what it can do. Change might be enabled by technology but it will be realised by people. Using AI is about applying technology, but it’s more about changing habits, changing processes.

“It’s not taking the same process and AI doing what you did before. You redesign your process. If you want to know something, do you use search or ask Perplexity or OpenAI? It takes time to change habits.”

Small steps, big progress

Despite what Düsener calls “massive” investment in AI, Düsener notes, “As yet, we’re not seeing the impact on a massive scale…but in small steps. We constantly do PoCs [proofs of concept]. For example, we just created a digital twin using AI with Cisco on our IP network. It is better at making sure that we don’t create issues through a change than testing it in our labs. This wasn’t feasible even two years ago…the constant small steps are as fast and impressive as a big leap.”

He shrugs off the suggestion that PoCs don’t always scale saying, “The context is obviously integrating it into our tool chain, integrating it into our processes. That takes time, as I’ve said, and we are taking our time because if we rushed into certain domains, we would create silos or we would fail because we need to work on the fundamental peak parts. The focus now is preparing for agentic AI – defining our structure, our governance and our platforms.”

Düsener says the first trials are imminent and that the first use case will be up and running this year, “but it is but it is mandatory that our data transformation is done by the end of 2026 then we can start.” Like all operators, Swisscom’s work on its data has been ongoing for a long time but over the last 12 months, it has worked on a data architecture to overcome the issue of siloes, replicated and incomplete data and so on.

“Now we’re aligning it so you can do cross correlations – real analysis, changing from a report-driven organisational business…that perhaps takes two or three weeks’ time to get an answer and being able to ask the data and it tells us what we want to know. Again I stress that progress is about changing habits, changing the way we operate our business, not only the platforms and technology.”

Designing agentic workflow

Will agentic AI help with cross-domain operations? “Cross-silo work will happen only if we think across silos in our agentic design,” Düsener cautions. “We have a simple framework in our heads about what designing an agentic workflow means: developing the agent is maybe 15% of the work; 35% is understanding the status quo, including identifying or creating the necessary APIs; then 50% is redesigning processes and even asking maybe if we have the right data? The last, small part is building the agent yet that’s where the public focus mostly is.”

In the long run, he thinks some kind of AI will be “a support for whatever we do by default” but that the real questions are about to what extent and in what capacity. He says, “We will maybe start by accessing information, then as copilot, then autonomous work for a person. In the future, job definitions will look different…it doesn’t mean we don’t need a software engineer, for example, but that the software engineer will work on a software engineering system as a whole rather than on particular lines of code.”

Impact on the network

Will the use of AI change the network itself? Düsener says this is something he and his teams look into constantly, but thinks it’s too early to say yet. “We envision you will have 10 or 100 agents residing on your phone, connecting to other agents somewhere in the network or in the cloud. All of a sudden, we have tenfold or a hundredfold the number of users in our network.

“How do you do identity and access management? Rights management? What kind of patterns will we see on the network?” He suggests that perhaps in five years’ time people won’t watch the same movie on Netflix, instead an AI-powered engine could create content for individuals, which “would really change network patterns. Today we have content distributed networks (CDNS) but that model won’t work if it’s not a streaming platform anymore.”

Likewise it is probable that millions of pairs of Meta AI Glasses or similar devices uploading a constant video stream will reverse today’s traffic pattern where downloads accounts for most traffic. At which point, “Latency might become an issue,” Düsener points out. “There are different parameters that don’t play that much of a role today but will play a role tomorrow. Then the big question is, must we do all this again for free – give away the extra value created by the agents for customers, or will they be willing to pay for it?”

New services and revenue

Will AI-powered services deliver new revenues? Will they change Swisscom’s business model? He says, “That’s the wrong way to ask. The question is, what value can we provide to our customers so that they want to pay more? 80% of revenues today are from services that didn’t exist 10 years ago. We are one of the biggest TV providers in Switzerland, for example, and provide cloud services on our own infrastructure and in partnership with hyperscalers.”

The operator also offers spam filters and security services for B2B as well as B2C, and in June launched beem. This is the world’s first secure access service edge (SASE) where the converged networking and cyber security service is operated entirely within a telco’s infrastructure. It is for businesses of all sizes.”

Düsener concludes, “I think this one of the most exciting industries to be in and one of the most exciting times to be in it because there is so much innovation. I love that. We’ve learned that we cannot just pile new technologies on top because it creates technical debt.

“We will have phased out 70% of our network applications and platforms by the end of 2026. We had 15 IP networks, now we have one. 2G has gone and 3G by the end of 2025. I’ve already asked my team when will we phase out 4G? That’s the habit: if I want something new, I must get rid of something old – a paradigm the industry didn’t follow in the past.”