Romano Roth: AI-Augmented DevOps with Platform Engineering

In this talk, Romano Ross discusses the evolution of technology within organizations, shifting focus from AI and DevOps to what he calls the 'cybernetic enterprise.' He highlights his extensive experience in software engineering and DevOps, revealing how continuous value delivery can be achieved through automation and effective team organization. Romano critiques the current AI hype, emphasizing that automation cannot solve existing organizational issues. He introduces the concept of a cybernetic platform, which integrates various tools and promotes a self-service model for product teams, allowing them to develop and manage digital products efficiently. The speaker also shares insights into the development of a cybernetic delivery platform with a private banking client, illustrating how such platforms streamline operations and foster innovation in software development.

Good morning. AI is not the future and defops is also not the future. Cybernetic is the future. And what that means for you, that is something I will show you in a minute. Let's come with me. My name is Romano Ross and I'm the chief of DevOps and a partner at Tsulka. I work for Tulka now since 23 years. I joined Tulka directly after university as a

junior. Net engineer became then an expert software engineer, then an architect and finally a consultant. And one thing that was always very close to my heart is how can we continuously deliver value? How can we automate things and how can we ensure the quality of what we are building and how can we organize whole teams or whole organizations across the value stream so that we are able

to continuously deliver value. So when the whole DevOps AI um platform engineering movement started, I jumped right on top of that. Became one of the organizers of the DevOps meetup sur which is a monthly meetup we are doing in Zurich with over 2,500 subscribers. I'm one of the organizers of the DevOps days which is a two-day conference in Zurich and these DevOps days they are all around

the world and I'm the president of the DevOps day. Now you see all these topics are very close to my heart. This is also why I have my own YouTube channel with over 250 videos all around these topics. I'm currently also writing my own book about the cybernetic enterprise which will be released in August and I'm also a lecturer at the HSLU in Switzerland for the topics

of CAS uh enterprise architecture digital transformation and also obviously DevOps. I work in different projects with different clients and I'm currently the product manager of a huge platform that we have built together with a private bank which I will also showcase during this presentation. And of course every night I also sleep 8 hours just in case you are asking. Do you sometimes also feel that way? Completely

overwhelmed about these AI news, about all these new models that are that are coming, all these new techniques, all these things you need to know about AI. It's just mindblowing how fast it is. Just this week there were different conferences where new stuff was coming out. The latest thing is wipe coding. Of course, it was coined in February 2025 by Andre Kaparski and by end of February

2025 to be precise and we already have two books with zero experience behind it. managers are coming to me and are telling me, "Hey, Romano, we don't need any developers anymore in the future." What is your opinion? Hands up. Who believes that AI will replace developers and DevOps engineers? Who believes that? One, two people. Let's have a look. Who has already built something with AI? Hands up.

Yeah, of course you have. I mean, you need to try that. It's It's cool. It It works somehow for Hello World stuff. But who has let an AI develop a complex system on its own? Like for example, a air traffic control system? Hands up. Oh, no one. H it doesn't surprise me. But who of you believes that AI will once in the future develop and operate a

complex system like an air traffic control system? Who thinks that some people? We need to have a talk at the coffee machine. So latest trends that we see coming is of course a chantic AI but the newest stuff I hear is that company wants to be a AI first company whatever that means we are completely trapped in a hype and you can see that very clearly by

the numbers. When you look at the numbers, then you see here the investments that have been done over the years versus the revenues. And I just point out the 2025 $1,000 billion US will be estimated to be invested into AI only this year. In 2025, the revenue will be roughly 120 uh billion out of that. And you can see there is a small gap between investments and

revenue. And as always, someone will pay for that gap. And um well I mean when you look at that you can clearly see there is a little bit of a problem because that gap is of course funded by some investors because they believe that the efficiency gains will be sort of in there. So what you see in that gap is what people believe the efficiency gains could

be when we are doing AI. So quite big. Now I already have experienced the same thing and now you also recognize how old I am. In the year 2000, the internet came up and everyone wanted to go to the internet. Everybody needed a online thing like a online shop and startups were coming and were doing some online things sort of and of course um also the investments

skyrocket. the revenue were not so big and to be precise on Friday March the 10 to in the year 2000 the dotcom bubble burst and I don't know but I feel sort of the same I don't hope that the AI bubble will burst but in my opinion we are completely trapped in that hype the latest news uh you can See here it's uh Cla um last year

um Clara CEO said all of the support personal will be replaced AI. Now he says um we had some quality issues and he's again hiring humans. The interesting part is more in the bottom area where you can see that the return on investment on AI is failing and many companies are already abandoning their AI um prototypes that they are doing. In my opinion, we are living in

the area of the AI idiot where chat GPT cowboys, clueless polit politicians, and short-term hyped managers are burning billions on that buzzword. And just to be clear, I don't believe that you are AI idiots. just to make that clear and also there are wonderful people out there which are researching that topic and they are absolutely brilliant. Nevertheless, I think AI is absolutely awesome. It's I use it

every day for my work. It really accelerated me and it's a absolutely fantastic tool that we have But we as DevOps engineers or as CTO's, we need to cut through that fog that is out there um which has built up and see clearly what really matters in that thing. And we as DevOps engineers, we need to shape the foundation that supports our company so that our company

can reach their business goals. That's the important thing. It's not to jump on the next AI sync and build something just with AI because everybody is building something with AI. It's really about supporting our business goals. And to be very very clear about that, AI not fix your broken processes, organization, technology, and governance. It just will not fix it. And therefore, I say the future is not

AI. The future will be cybernetic. And that's a very important thing because cybernetic means that we are building an adaptive system where governance and yes and AI together with humans operate in constant feedback loops to continuously adapt, improve and support the company's business goals. That's the important thing and this is where the companies will move into the future. Now when we talk about such a cybernetic system

then this is a cybernetic enterprise and such a cybernetic enterprise needs to have the right foundation and this foundation is built by us the DevOps engineers. Now as I always say we are going into the area of software development industrialization. At the moment you can see that um many product teams are working with internal teams or with external teams or with mix teams and they are building

digital products um on different environments and they are bringing that into production. that usually in many company leads to quite a large tool landscape or even methodology landscape. Now where the whole industry is moving which is also a big trend and you will see that at that conference despite AI is platforms. So many companies are building nowadays platform where they are standardizing how software is built, how

their digital products are built in their company with a clear set of services and products um which show them how to build their digital products and with that you get this standardization over that platform. Now when you look at the target operating model of such a cybernetic enterprise then we are going clearly away from the product teams which had a huge technology stack and all the roles

in there because the cognitive load was too high and also the complexity to build these softwares were just too high. We are moving away from that and are moving into a direction where you have teams which support a small technology stack and where you have teams that work on a digital product and solely focus on this on that product on the feature development of that product or

even of that module. Below you have a selfservice platform that is built by uh a platform team which supports the large large amount of the technology stack. That's an important thing. And now because that's so important, we need to zoom in there because there is usually quite a lot of misunderstanding when it comes to platform engineering. So when we look at that picture then you can clearly

see that on the bottom line we have the platform team. The platform team builds a cybernetic platform. This is an internal product and the customer of this internal product are the product teams. They use this cybernetic platform. That's the first important thing. The second important thing is that this is a selfservice platform. There is no ticket ops. These product teams they are using directly this cybernetic platform.

This platform has all of the tools in there that the product teams need to develop their products. It has all the capabilities in that these product teams need to create products. Now the product teams they have the end to end responsibility about their application. It means that they are doing DevOps. They build it and they run it. That's the important thing. Now when we look at an

example because that's quite important. Observability for example, the platform team will give that capability of observability for example via um Grafana or via Prometos to the product teams. Um it can be by default dashboards or by Grafana that is already working. the product teams they will use capability on that cybernetic platform to monitor their application. So the monitoring is done by the product team because they have

the end to end responsibility of their product. That's the important thing and with that you can clearly see that the platform teams they are generating value for the product teams and the product teams are generating value for the customers. That's the important concept behind Now some of you might say okay Romano sounds good um in theory but does that work also in practice and yes of course

that works also in practice. Um we at Tulka we have built our own cybernetic delivery platform um together with uh a private bank from Likenstein and this should only be an example how that is so that you get sort of a glimpse how you can also build your own platform that's the important thing so we at Tulka we use that platform internally for our internal products but

we also use that for our customer projects. So when a customer comes and he has no uh other platform then we will use that platform but we also use it to give to customers so that they have their own platform or we use it as a blueprint or we use parts of it because it's quite modular as you will see in a minute. On top you have

the people who are working with the platform. You have a CLI and you have the um the internal developer portal which I will show you also in a minute and they are using directly the tools of that platform and of course in your area these tools can be completely different but that's completely okay. Then we have the internal developer portal and usually here we are talking about

backstage. Many people think that platform engineering is just setting up backstage but that's completely wrong. We have not used backstage. We have built our own. As you can see what we have done is we have done domain driven design proper architecture. We have identified the subdomains which we need for our platform and then we have created a processing and automation layer and that's the hard stuff and

that's where many companies are currently massively struggling. So what you want to do is you want to automate all of the infrastructure stuff but everything that is uh all around that because you don't want to do different um um ticket ops you want to give a self-service um self-service portal to your uh your customer which are the product teams and here um it's also important to recognize

sometimes you go to the cloud or to different cloud provider But sometimes you are also on prem. So for example in our case we really have the cloud providers and onrem. Then below that we have the unified integration blocks. And of course they are sort of the same. And then we have the adapters. And here is another very important concept. When you create your platform, you want

to integrate all of the tools, but you want don't want to integrate the tools together. You always want to integrate it over the platform because if you integrate them that way, you cannot get rid of these tools. One example here we have GitLab. I love GitLab. GitLab is awesome. But you also need to see that company makes roughly minus 50 million every uh quarter and um there

are rumors that the company should be sold. I mean if it's sold to Google everything is good thumbs up. If it's sold to Broadcom this might be a little bit difficult and then you would like to get rid of that tool. And that's the important thing about creating such a platform. The concept behind that is this concept of a so-called floating platform. This concept is from Gregor

Hopp out of his book platform strategy. Absolutely awesome book absolute must readad. So what we have is that floating platform which we want to build and we want to integrate all the tools, all the DevOps platforms and all the cloud providers that we have into our platform so that we can create that self-service platform. What we don't want to do is duplicate any of the features because

then our platform begins to sync. So we just integrate that. Now how does that work or how does that feel? Here I have one example how that can feel and how how that is implemented. Um we see this area where we have the cloud and here we have onrem in that case this is a private bank. Um where the developer are internal and external and they are

directly bringing their own devices. They can directly use the platform which is in the cloud and we have here a development environment and here we have the production environment. The developer cloud of course there you need to have API mocks you need to have synthetic test data so that they can work because the real data is protected in production. So here they are developing their um their

applications and now we have another instance of the same platform in onprem and in the production cloud and this is very important. Now we also see why many people are getting confused about platform engineering. They think this is a developer thing. But when you want to have a real cybernetic platform, then this platform spans from ideiation until production, you have a seamless integration of the same platform

of the same tools. And that's a very important concept, which means that these platforms they are going outside and getting the release candidates into um the production directly. That's the important thing about that. Of course, this is just one case how you can do it. We have at the moment also another case where we are completely onrem. There are different cases how you can develop your own

platform. The important message for you is just don't focus only on the development part. Focus on the whole picture until production. Now some of you may say but Romano hey AI is not in this picture. Why is AI not there? Usually AI is everywhere. And here we are at the very important point in your cybernetic platform. AI is just a capability. It is there to support the

product teams in developing their application. That's the important thing. But not only that, over such a platform, you can provide AI in a uh secure and governed way to the whole um to the to the whole enterprise and that's the important thing about that. So some of you might say, but Romano, it's such a small box. AI is usually huge. So let's zoom in. It's not that

small and now you also feel a little bit the extent of such a platform that you can build. So when we look at that then of course on the top layer we have the application we have there for example chat bots synthetic test data AI coding assistance that you want to provide or knowledge management all of these systems you want to provide to the whole company but

also to your product teams that's an important layer the second layer is then the tools where you have a prompt engineering vector databases or even golden pass solution for AG um solutions that's also very important that you can provide that over the platform to your teams and then on the second uh on the on the lower level you have your model hub where you store the different

versions of the models that should be used where they are versioned and on the bottom layer you have then your Gen AI infrastructure which can be of course in the cloud but in some cases you also want to have that on prem and this is also some trend that I see very clearly that many uh people are also now building their own data centers again up and

hosting LLMs on in their own data center. So now you have seen quite a lot of theory and how that is but um I would like to do a quick demo so that you get a glimpse how that feels when you have such a platform in place and uh I hope everything works with the demo. What you see here is a platform um that we use at

Tulka. uh at Sulka we have that platform in in place for us for our internal pro projects but also for our customer projects. So this is how it looks like and down here you see some blurred out stuff. This is because of customer names. Um so I will go now into the administration area. And in the administration area, what we see is that at the moment 461

uh users are using that platform. We have 15 partners and 26 spaces and 13 Kubernetes clusters which are running. And down here we would see the costs of that. One important thing is the so-called partner concept um where we are able to onboard new partners. In our case it's new customers. In the case uh of an of another bank it's their vendors that they have with that

we can just um add a new partner. We enter their domain and over um Microsoft enter ID and bring your own identity. They are onboarded. Now the importance thing starts. You remember back the floating platform principle. All of the tools are integrated which means they have now access to all of these tools and to the space or the source code repository which also means we can onboard

someone within a second and of course we can offboard that person within a second which is already quite awesome because in many companies this takes two two weeks to a month. The other concept is the so-called space concept. A space is nothing else than a network zone. We use the hubspoke network uh for that concept. And in such a space, our uh customers can create their own

kubernetes cluster or they can uh use virtual machines in there. Let's go into one of the spaces. We have a platform space. So we built that thing with in inside of this platform. So eat your own dog food. As a developer, I directly see what tools that are in place in there. I can go to repositories. I see uh my repositories. I can click um on one

of the repositories and over single sign on I'm directly signed in. So I don't need to remember any of my passwords. Now the interesting thing starts when you have such a platform and all the repositories are in there suddenly what you can do is scan them. Um normally what you would do is you would have a CI/CD pipeline where you do the scans. Now you can do

that platformwide which means you can enforce security directly over the platform. Here you can see repository scans. We are using soft license scanning and um secret detection in here. I go um oh oh back we can also create new repositories here. In our case we can use GitHub or GitLab repositories but we can also uh create the repository from a template and these are these golden path

templates. Of course, in the case of of our company, uh we don't have that many, but many of our customers have massive amounts of of uh templates where they specify how um an application should look like, how a REST interface should look like including all of the infrastructure behind that. Then uh quickly to the uh registry. Here we have the container registry. I quickly go into one

um of these containers and again what we are doing here is container scanning that we can do where we see the violations and we can analyze that container image with an AI and these are new things that we are currently developing where we are bringing in AI into such a platform and that's an important thing as soon as you have such a platform you can use that

AI capability also inside of the platform to analyze um for example here that docker file. We also have uh the classical packages like Maven, Nougat and so on. But the interesting part is the Kubernetes cluster where the applications are running. Here we see for example a chat application where we see all of the uh the uh Kubernetes um resources. We see the public endpoints again. Application violations

are shown in here and um what we can do is we can open the log files in here. We see now all of the log files. Um I just quickly go over here because that takes some time. Now this is quite interesting because now when you have such a platform you can provide observability or monitoring directly to your customers which are the product development teams. They get

monitoring out of the box and they don't need to do anything because the platform provides that directly for them. The only thing that they need to do is um log out to standard out. Here um we see for example a tempo the um the graph for the microservices and of course we also have a bunch of um dashboards which are already pre-filled for um our customers and

with that um we go to the service catalog and that's also an important thing because sometimes you want to have a new database for example a database and then you can just click on that and the platform will provide that fully automatically for you. You don't need to care again about um passwords about backups. Everything is standardized in a governed way um for you. Uh and of

course our preferred way is via infrastructure as code but you can also create it directly. But the interesting thing is that open AI this is how you are providing to the development teams in a governed way AI use and we also did that over that platform and this helped us massively because before we had uh everybody was doing some AI stuff. It was absolute hell and our

zo was like whoa what are you doing? Um and with that platform we could bring that all in a secure and governed way. Um we based on that we have developed a chatbot which uh is running on our on-prem infrastructure so that we don't um put out any data um to to the cloud providers. We have a so-called reference finder. We are quite an old company over

50 years old. We have thousands of projects. So if I want to to ask have we done a project like that then I can type that in. We have a senai project where which is just a bunch of very interesting prompts that we use on a daily basis. And then we have a whole education platform for our people um with examples uh on how to develop their

AI stuff and all of these things that you just saw they are all running on that platform. This is um when you have such a platform this is a massive enabler in in the meantime people can create a new application within 15 minutes and they are directly in production with certificate with backups with everything fully standardized and also our CISO is absolutely happy with that. So that's

a huge enabler. The last thing I want to show you is um the last screen where you see the costs, the CO2 emissions and you can also add members here. Everything is selfservice and that's the important that. Let's go back to the presentation. So what you could clearly see is that we want to move away from that picture that we see here where you have ticket ops

where you need to wait for weeks until you get some infrastructure stuff. um where the services of cloud providers are just put there and you need to figure out how to use it. Where you want to go is into a cybernetic factory where the services are provided to you where you can just pick out of your um of of the shelf the services that you need or

even that you get a whole factory for your project directly delivered by that thing. That's where the industry is moving at the moment. And the concept behind that is the so-called cybernetic enterprise which is a continuously involving um regulated system with feedback loops in where humans and AI are working continuously together and adapting and continuously improving that system. That's the important thing. And in summary, we can

see we are trapped at the moment in an AI hype. Um, what we need to do is we need to cut through the fog and see what really matters. It is that you need to build the right foundation. This is something you do with platform engineering. What you need to do is build a floating cybernetic platform and AI is just a capability that you provide over that

platform to your customer and with that you can build your own cybernetic enterprise and in my opinion the future really belongs to those who master the symphony be between organization process technology governance and AI. And this is the future of DevOps. Thank you very [Applause] much. All right. Thank you so much. And uh let's go through your questions. I want to remind you that we have a

slido platform. So this platform is for you to write your questions. And Roman, here we go. Could a AI replace mention platform team in the future? If not, why? Very good point. At the moment, I don't believe that they can do it. What you need to see um what we are talking about when we talk about AI. At the moment, we are talking about large language models.

What is that? A large language model is nothing else than a a simulation of a neural network which is labeled by someone in an enterprise for example open AI. So it just mimics some answers some best cases. In my opinion a platform as you can see before is a complex system. Our people are also developing that platform. they are using um AI to develop that. But as

my expert software engineers usually say, when they need to create new stuff, they just disable um copilot because it just um puts rubbish in. Romano, we do have a question from Robert. Are you suggesting that companies should become uh cloud agnostic because that would make it harder to make use of the managed services these providers offer? Yeah. Um I'm not suggesting that. Um what what I see

um many companies they are sort of alarmed um due to the the developments uh in in the US and at the moment they want to become more resil resilient. That's why many companies want to be able to move their workloads if needed also back on prem or to another cloud provider. When is breaking point for a company in size when they should build platform teams as for

startups or small companies they need to move fast and create Roy? Yeah. Um in my opinion to build such a platform the breaking point is that you have at least five product teams. So at least 50 persons then it makes sense to have such a platform. If you are below that, it does not make sense.