KubeCon + CloudNativeCon EU 2025 | Amit Eyal Govrin & Shaked Askayo, Kubiya

Clips
More from KubeCon + CloudNativeCon EU 2025

Amit Eyal Govrin

Co-Founder & CEO

Kubiya.ai

Shaked Askayo

Co-Founder & CTO

Kubiya.ai

play_circle_outline Adoption of AI in enterprise and challenges in deploying AI experiments to production

play_circle_outline Importance of governance, predictability, and controllability in deploying AI applications

play_circle_outline Security concerns in exposing AI capabilities and the need for network-bounded resources

play_circle_outline Ensuring Enterprise AI Success: Guardrails, Hardening, Deployment Experiences, and Future Plans for Kubiya

Info
Transcript

Amit Eyal Govrin & Shaked Askayo, Kubiya

Amit Eyal Govrin

Co-Founder & CEO Kubiya.ai

Shaked Askayo

Co-Founder & CTO Kubiya.ai

Advancements in AI Adoption for Platform Engineering at KubeCon+CloudNativeCon EU 2025

In this insightful episode from KubeCon+CloudNativeCon EU 2025, Amit Eyal Govrin, CEO, and Shaked Askayo, CTO, co-founders of Kubiya.ai, join theCUBE Research in an engaging discussion on the evolving intersection of artificial intelligence (AI) and platform engineering. Hosted by industry experts, the video explores Kubiya's innovative approach and its implications for the enterprise environment.

Govrin and Askayo provide their expertise on emerging trends an... Read more

explore Keep Exploring

What was the previous conversation topic regarding agentic workflows and its relevance in the enterprise, and how has the conversation shifted since then? add

What is one of the challenges when it comes to controllability in operations, especially when it involves actions in production that could potentially bring down a website? add

What are some concerns related to exposing AI capabilities to the outside world, particularly in terms of security and model deployment? add

What are some of the challenges and concerns surrounding enterprise adoption of AI applications, and how has a particular platform addressed these issues? add

bolt Powered by CUBE AI

Amit Eyal Govrin & Shaked Askayo, Kubiya

search

Rob Strechay

>> Hello and welcome back to KubeCon CloudNativeCon London. We're really packing the day today. Actually got some extra people coming on to help us unpack what's going on. We're continuing down the path of AI today because today seems to be AI day and I could not really think of two gentlemen that I would love to have on. I want to say thank you. You had me at your dinner. I got to meet some of your end users the other night. Amit and Shaked, thank you for coming over here. You guys are the co-founders, CEO and CTO of Kubiya. And I got to really talk to some leaders in platform engineering and it was really cool because trying to understand some of the experiences that they're having dealing with this thing called AI and all of the complications that it's bringing to them. Kind of help us break it down for around the what's going on with AI and platform engineering and how it's changing things and why it's so important and what's Kubiya's perspective on that.

Amit Eyal Govrin

>> Sure. Thanks for having us again, Rob, and great seeing you the other day. That was great conversations all around. I'll start with maybe the genesis. Had we had this same conversation one year ago, you would have been asking me why agentic workflows and why should we use it in the enterprise? Because it was still in the market education phase. I think that's solved all of Sam Altman and Dario and all those folks have done their work. Now we're having a different type of conversations, how, how to deploy these agentic experiments, if you may, hackathons into a production grade environment that the enterprise can consume that you can go to your CISO and they can sign off on it without laughing out the room. That's really kind of where we're taking it.

Rob Strechay

>> Yeah, I mean because we see it from a perspective that it's exactly what you see in the data. About 75% to 80% of organizations still haven't gotten the production yet with their AI and they're spending a lot of money. And in fact in this year what we see is that budgets are only going to grow by like three to three and a half percent this year, which is down even over last year because of all the global uncertainty. But let's kind of take it a little bit deeper because you and I were talking and catching up on how there's just a lot of parts pieces. MCP has had a moment this week and has been talked about. I've already been quoted as saying it's a science project because there's so many other things that need to go around it. It doesn't solve the end-to-end problem. And you had a very interesting perspective on that.

Shaked Askayo

>> Absolutely. So I'll take into consideration when somebody wants, when an organization wants to adopt agents, so we can throw the word agents to the air, but a lot of things are popping up to our heads. What exactly does it mean and what exactly does it require to get an end-to-end solution, a real problem, a real world problem, really solved using those agents, right? And the agents by themselves are nice. I mean, it could be very cool for an hackathon project, right? Maybe you can take some open source orchestration platform and try to get it to run some nice python function for you. But in the real world, especially in organizations, a lot of questions are popping up around how am I going now to scale that? How am I going now to adapt it as part of the organizational processes? MCP for example, which is as you're saying, a lot of trend now, in the end is just a standard. Okay, just a standard that came in. It's not a solution. It wouldn't save the way that you're going to consume those agents or suddenly get agents into production. Something important about, for example, MCP, the word that comes near MCP is usually the word server. So MCP server, MCP servers. And when we throw the word server to the air, what comes up is wait, FLDC deployment, maintenance. How am I going to replicate it across environments? How I am going to scale it to where it's going to be hosted? All of those are part of what the ecosystem is now trying to face. And we take a different approach to how we are looking on it. Unlike how people are talking about agents which are still on the application layer, we prefer to look on agents like infrastructure because they're not any different from the way you would manage your cloud, right? How do you replicate it now across environments? How are you going now to assign different permissions? How are you going to extend those agents with tools that are bounded to your VPC, for example? That's where things are getting very, very complicated. The aspect of how we see it is essentially you shouldn't manage it any different from how you will do infrastructure as code, right? So why not look on those agents as resources, right? And everything that is combined into this end-to-end workflow will be pure infrastructure resources that will give the application layer what it needs to really get you into production.

Rob Strechay

>> And like you said, and I think this is one of the places that it always... It's why I called it a science project because I said again, when we were in the data on Kubernetes thing, we polled the audience, just 600 people in there, 10 and 15 were actually deploying production AI applications and I mean by that gen AI and doing inference and/or training. And it was even less on the training side, which I think is a good thing because I think if you're training a model and trying to... Unless you're Anthropic, OpenAI, meta or some of these big guys, I don't know why you're building your own model. Maybe you're fine-tuning it and doing that stuff and putting guardrails. But I think to your point, like you said, a big piece of what I look at it is there's a governance that has to happen. And people are looking at platform engineering and saying, "Hey, how do you help me control AI?" Is that what you're seeing as well?

Amit Eyal Govrin

>> Controllability is a major driver and I think one of the things you hit on the head very nicely, it's one thing to do lightweight operations, update the CRM record for me, go query a database that's public or some frame of mind. But what happens when you actually have to take action in production? That may take your entire web front down or website down. Now that's a whole nother level of operation that requires a lot more governance, a lot more control, or predictability nonetheless, in order to get that. And I think that's where we separate these type of operations from nice to have to some that requires predictability and controllability all the way through.

Rob Strechay

>> Yeah. Because I mean, again, and I just, I'm trying to pull the signal out of the noise of MCP. MCP Is a protocol. It's not like a thing. It's like a protocol that it's part of the solution. And do you see it that way? Is it like, Hey, it's good because like you said, it's a standard, but it's not everything and this is why you need to be able to build around it?

Shaked Askayo

>> Absolutely. So MCP allows basically those AI applications to consume tools and external data in a more standardized way, but it's not that suddenly because MCP is declared now as a protocol, it's going now to save the fact that those are servers on the other side that are running those tools. And essentially in an organization, those servers will need credentials to your tools. They need predictability like Amit was saying. And essentially someone will need to manage them and watch over them. And because like I said, server, then your DevOps team, your platform engineering team will need to manage those. They will crash. It's not that the AI suddenly can intelligently know how to consume those tools. And I think that it's a good start to build the ecosystem around. But as you were saying, it's just one piece and what's going on behind the scenes in order to get it really to do meaningful stuff, solve real world problems, especially in let's say air-gapped environments, things are getting very, very tight.

Rob Strechay

>> But that's where I was going to go because that was the interesting conversation I had with said customer at your event, was that the fact that they had the requirement for being able to do stuff in an air-gapped environment. And we know that sovereign, sovereign data, data sovereignty. We also know that 85% of data that organizations want to use in AI is actually on-prem and maybe it's not necessarily air-gapped, but they don't want it leaking and they're very conscious of that. Are you seeing that being a push for you guys and where you've actually taken your product in that direction?

Amit Eyal Govrin

>> It goes without saying financial services, automotive, there's a lot of those frameworks of customers who are talking to us about this. Whether we want to or not, that's where the customers are taking us. And of course a hybrid infrastructure or a hybrid SaaS where the control plane lives on our prem and then they could have it on their side. That's also a nice little in-between for the SMEs, SMBs that talk to us that don't have the same regulations.

Rob Strechay

>> Yeah, we were actually talking about that with some of the hyperscalers earlier this week about the fact that on-prem is in vogue again. And so it was very amusing because I was like, "Okay, we're going to be talking on-prem and I know their solutions very well." And again, there's all of them have some ability to go on-prem. Are you seeing that... Also, it's the fact that it's not only on-prem versus a cloud. It's on-prem versus multi-cloud. Are you seeing a lot of that with your organizations? Not necessarily stretching AI across there, but they're trying to get control and doing their platform engineering in a standardized way across those.

Shaked Askayo

>> Yeah. So what really happens is that as part of this new kind of standard, right, part of the questions that pops up is they need to expose now their kind of superpowers to those AI applications. So they have the protocol and the capability of doing so, but they are very afraid of let's say exposing the server now. So cloud or cursor can now consume it, but you need now to sacrifice the exposure to the internet, right? Because those apps are essentially on the cloud. And what really happens is that in the real world, they don't want those servers exposed in such a way. They would want them to maybe talk with their inner tools which are network-bounded. So multi-cloud environment, like you're saying, I could have a thousand Kubernetes clusters and each of those clusters has different resources that are bounded to the network of this cluster. And then what I'm going now to expose all of those capabilities to the outside world just so an AI can call it, but now also humans can get into the traction. That's part of what we're hearing, the security concerns about that, which is a significant part. But also the fact that they want to run without public models. They want to take leverage of let's say a fine-tuned LLaMA or even function calling model open source that they want to host on their infrastructure. Because they don't want to expose all of that, the conversation history and the context with their tools to the outside world or those public models. So a lot of the concerns that we're hearing are exactly about this topic. It goes through security, but it varies to model deployment and management of those tools and aspects.

Rob Strechay

>> Oh, absolutely. So last question here, the time's flying by. So let me ask you what you're excited about now. Why Kubiya now? Why Kubiya when we get together maybe in Amsterdam in a year and we're getting an update from you guys? Help me understand that.

Amit Eyal Govrin

>> We've seen a lot of enterprise adoption strictly because a lot of people are looking for people who've done that, who've deployed enterprise AI applications in production with all the guardrails that enterprises want. I think there's a lot of misnomer, a lot of noise out there, a lot of frameworks out there claiming to be AI native and able to do so. But when you dig under the hood, you're seeing where all of these things fall between the cracks and it's really the enterprise use cases, the hardening, not exposing new attack surfaces, non-human identities, giving an agent God mode to go mock in your infrastructure. These are the hard questions and concerns, rightfully so, that people are trying to solve for. And by blood, sweat, and tears, we've had to go through this for the last three years internally for ourselves, for pre-gated use cases. And now we've opened up this platform really for others to consume and use as a usage base, which really is come build on top of us, not just consume end user use cases with us, which has been our claim to fame up until now, but also come build on top of us. Even ISVs can build on top of us at their enterprise ready applications tomorrow.

Rob Strechay

>> That's great. I think we'll leave it there, but I'm looking forward to the update when we get to either Atlanta or in Amsterdam. I think it's great. You guys, again, I thought some of the folks that you had there and the customers were asking all the right questions and I loved the answers. So thank you for coming on board and we'll see you guys soon.

Amit Eyal Govrin

>> Thank you.

Shaked Askayo

>> Thank you so much for having us.

Rob Strechay

>> And thank you for watching this episode of KubeCon CloudNativeCon London in 2025. I can't believe it's already 2025, and we're cruising through day two. Stay tuned. We'll be right back.