KubeCon + CloudNativeCon NA 2025 | Amit Eyal Govrin, Kubiya

Clips
More from KubeCon + CloudNativeCon NA 2025

Amit Eyal Govrin

Co-Founder & CEO

Agentcy Labs

play_circle_outline Introduction of Amit Eyal Govrin, CEO of Kubiya at KubeCon + CloudNativeCon 2025.

play_circle_outline Challenges faced by platform engineers in managing AI complexities and project success rates.

play_circle_outline Leveraging Agentic Engineering to Optimize On-Demand Capacity and AI-Driven Project Planning for Efficient Resource Management

Info
Transcript

Amit Eyal Govrin, Kubiya

Amit Eyal Govrin

Co-Founder & CEO Agentcy Labs

In this KubeCon + CloudNativeCon North America 2025 interview, Amit Eyal Govrin, chief executive officer and co-founder of Kubiya, joins theCUBE’s Rob Strechay to unpack how platform engineering must evolve for AI to succeed in production. Govrin explains why Kubiya emphasizes abstracting away AI stack complexity so business stakeholders can prompt for outcomes and measurable ROI, rather than investing in over-engineered frameworks that never make it past proof of concept. The discussion covers the mounting mental load on platform teams, the need to treat AI ... Read more

explore Keep Exploring

What updates were shared by Amit Eyal Govrin regarding the industry since the last conversation? add

What challenges do platform engineers face when transitioning AI systems from proof of concept (POC) to production? add

What are the challenges and potential solutions for managing an unexpected increase in AWS costs within an engineering organization while maintaining existing project commitments? add

bolt Powered by CUBE AI

Amit Eyal Govrin, Kubiya

search

Rob Strechay

>> Hello and welcome back to KubeCon + CloudNativeCon North America 2025 from moderately-warm Atlanta. We're starting to warm up. We're definitely warmed up on stage here. We have some great stuff. We're continuing to train and talking about AI and making AI real. And I'm really excited to have Amit Eyal Govrin, who's the CEO of Kubiya and one of the co-founders as well. So, welcome onboard again. It's been a little bit, we've talked in Atlanta about six months ago, or not Atlanta.

Amit Govrin

>> London.

Rob Strechay

>> London. Where are we today?

Amit Govrin

>> Apparently Atlanta.

Rob Strechay

>> It feels like London. The weather's not cooperating yet, but great to have you on, Amit.

Amit Govrin

>> Thanks for having me back, Rob. It's always been a very interesting checkpoint in terms of where the industry... benchmarking, how fast we're running and seeing where everything's evolving.

Rob Strechay

>> So, give us a little bit of an update on what's been going on with Kubiya since London.

Amit Govrin

>> Absolutely. I think one of the things I'm going to take from this conversation, we're not going to talk about the enterprise stack, even though that exists, right? We can go down a very deep path and a rabbit hole around zero trust architecture with agents distributed compute, durability of the execution. We can go down that path, but that's not where I want to take it. We've actually understood from all of our engagements that the industry, in order to adapt AI needs to take it the opposite way. It needs to abstract the way these complexities and put it in the hands of stakeholders and business owners who actually can know and to prompt the AI to actually receive those outcomes and to drive positive ROI from their projects, not just to go into over-engineered projects, which we're accustomed to seeing with some frameworks out there.

Rob Strechay

>> Yeah, I mean, again, we were talking a little bit beforehand and I think everybody's trying to figure out... I was at the Infrastructure As Code conference, IaC, ConfConnect on Monday night doing a panel, and one of the things that came up... And it was a room full of platform engineers and they're like the complexity of running these AI systems and getting from POC to production was just such an immense amount of mental toil for a lot of these people. That, and how do you keep engineers productive and working in the right direction? But also, a big piece of it was how do they not become a statistic, like that MIT statistic that basically 95% are not getting from POC to production. I think a lot of that has to do with there's just no ROI done upfront. Help people understand where you guys are approaching this differently.

Amit Govrin

>> Absolutely. You definitely reference a few talking points we should cover. The first one around the platform engineering side. We really originated from platform engineering. What we understood that platform engineering should not be treated as its own unique segment. It should be treated as a necessary part of the process of deploying AI in production. And I think platform engineers oftentimes like to over-engineer and over-complex because everybody wants their own flavor, just like they did with the backstage approach. Everybody wants their own IDP. So, in this case, everybody wants to host their own models. Everybody wants to run their own inference, everybody wants to run their own frameworks, and that gets everybody in trouble because there's no standard of how we can do so where we understood and we had to take that step back, we looked at our customers, they were starting to get a little out of sorts in terms of where should we start? Can we go and introduce this framework? And of course you can, it's framework-agnostic, but let's solve the problem first. What are you actually looking to do? Step back. Engineering velocity. You're trying to go from point A to point B fast. You can do so on a microcosm of a workflow or an agent and get your engineers to code faster, but what happens with that code once it has to go into reach production, you move the bottleneck upstream, so you never actually solve the engineering velocity problem. So, now we said, let's take a step back. What are you looking to solve? You're looking to solve engineering velocity. Who's in charge of this? Your CIO, CTO, VP engineering. Let's give him a way to manifest this project to effectively go and to have an on-demand agentic engineering org that he can go and summon. He or she can summon and essentially go ahead and scale on-demand engineering capacity without needing to go into divert the attention of the existing capacity on hand.

Rob Strechay

>> Right, which I think is a key. I think when we were talking with all these platform engineers, a lot of it was, "Hey, we need some place where people can go and really self-service," but people want to interact in different ways and things like that. But funny enough, we're not getting 20% new engineering bodies brought in to go and solve for these AI complexities that we have here. How are you guys really approaching that where you can show people how they can get from POC to production with ROI?

Amit Govrin

>> I'll give even the most business-friendly attention. You're a head of an engineering org, and now your CFO office came to you and said, "Guess what? Your AWS bill is 20% more than your quota for this year." Now, you already have a roadmap, eight months figured out. You don't have the capacity... You may even be in a head count freeze, so you don't even have the extra capacity to deal with it. It's a top priority for the CIO on down, but it's not necessarily a top priority for the engineering board. How do you go and make that extra quota? So, typically, you'll have to go into hire, maybe even outsource some of that talent, onboard them, train them, get everybody up to speed. By that point, six months past, six months of burn happen. How do you actually do it without sacrificing that? Let's assume that same VP of engineering now had Kubiya as an example, and could go and prompt in the task in simple English, "Scan my entire AWS bill for the last 30 days. Find underutilized resources for the last 30 days that are candidates for retirement and terminate those." Immediately, there's going to be an agentic project planner. We're going to use a powerful model, of course, a thinking model. It's going to go and plan out the entire PRD. "Here are the agentic teammates or the agentic engineers that are required. You may need a CISO org to go to approve retiring some of these resources. You may want to have compliance involved. You may want to have the platform engineer and a FinOps team involved, and all of those are costed out. Now, you know exactly how long it's going to take this agentic engineering org to perform this end to end. It's going to know how much it's going to cost you in terms of token consumption. And if you want to approve it, at this point, you get full visibility. It's a kanban board that starts getting populated with tasks, and these tasks get solved with agents with the human in the loop. So, every time there's a necessary human in the loop, you can go and approve those. So, imagine a project that would've taken... This is a real project for one of our customers, a three month 20 headcount project and God knows how many hundreds of thousands of dollars it's taking them less than three hours.

Rob Strechay

>> Wow, that's-

Amit Govrin

>> $27, real stat.

Rob Strechay

>> Yeah. I think that type of improvement in velocity and efficiency is key because I think a lot of people are looking at this and going, okay, maybe with things like that and tooling like that with Kubiya, they can get to a better than 95% failure rate, 5%, or looking at possibly 5% success rate getting to production. But one of the things that's interesting to me is in the customers you're talking to is that, again, they're looking at how they can treat AI and treat these applications as products because you bring up kanban and PRDs and things of that nature. Are you seeing that organizations are starting to mature their platform engineering teams that way? Because we heard that from some of the other larger companies and we see it seeming to trickle down.

Amit Govrin

>> I think maturity is all relative. I think we're still trying to find a motion where a self-service organization is in fact self-service. You may have reduced the need for 20 more engineers for capacity because you've introduced self-service, but you introduced those same 20 bodies on the platform engineering side to attend to those. So, unless there's a true self-service way to automate not just middle-to-middle tasks, but end-to-end tasks, ones that require multiple data silos, multiple engineers and multiple individuals in the loop may require a full process automation of 200 plus steps, thousands of tool calls and agentic speak, and you could still perform those end-to-end. It's all because you knew how to break them down into tasks and how to introduce AI to expedite the velocities processes end-to-end.

Rob Strechay

>> And do you see that folks are trying to really drive their AI to, again, solve task-oriented problems? Because we see this right now where we started with generative AI and RAG and things of that nature and we've moved towards more agent-oriented types of deployments. And then, we're getting to agentic, which is several agents, maybe some real traditional AI type stuff or ML and tied with things. How are people looking at that entire end-to-end process as they... Because they're going through their journey. Like you said, the maturity levels definitely vary in there.

Amit Govrin

>> Sometimes it helps to be early, sometimes it hurts. It's usually both. But one of the things that we've discovered, because we were early in this space, is just how looking at it as an agent or an agentic workflow isn't enough. You have to look at it as an end-to-end process that may involve deterministic workflows, that may involve agents that are non-deterministic that can go and take action. It may require some BI and some analytics on top of that. It may require bring in a human in the loop and bring in a compliance team to actually go on to use infrastructure as code because we all need to be grounded with the sound practices and respect to SDLC. All of that has to be true. It's not either or. I think a lot of people are now starting to arrive at the bigger picture, and we're here to greet them in open arms, of course. To be a resource for them as well. To help them understand that when you look at this, you're not replacing engineers or the orgs. You're really introducing a new way to bring end-to-end engineer and velocity to the organization, to augment the existing teams and to really make everybody more successful.

Rob Strechay

>> So, as we look forward, again, we're going to be back in Salt Lake a year from now. It'll probably be the same temperature as it is here in Atlanta this week, probably with snow again. Yeah. What are you hoping that from here to a year from now, that you're going to be able to say that you can't say today about where Kubiya's at? Where your customer's at? Where platform engineering is at?

Amit Govrin

>> We're blessed to say that we're in the process of very lucrative projects and multi-year contracts that we're now signing out of Europe out of US.

Rob Strechay

>> Congratulations.

Amit Govrin

>> Thank you very much. And I think that's going to going to give a North Star to the industry that it's not too much to say that an autonomous or on-demand, agentic engineering org is possible. I was almost taken back to not have to say when we're introducing these concepts, because you always have to be sensitive to the political decisions in these organizations, but we're not ashamed to say it anymore because we're seeing the results in the field. We're having C-level board-level discussions, even with some of these companies, how this can be a real transformation for their competitive moat within their own respective industries and we're very happy about that.

Rob Strechay

>> Yeah. Well, I can't wait to have you and one of your customers who can come up here next year in Salt Lake.

Amit Govrin

>> Done.

Rob Strechay

>> Awesome. Sounds good. Well, thank you for coming onboard. Really enjoyed this.

Amit Govrin

>> Thank you very much.

Rob Strechay

>> And thank you for watching this episode of KubeCon + CloudNativeCon North America 2025 from moderately-warm Atlanta. We'll see you soon.