theCUBE + NYSE Wired: Mixture of Experts Series | Spiros Xanthos, Resolve AI

Clips
More from theCUBE + NYSE Wired: Mixture of Experts Series

Spiros Xanthos

Founder & CEO

Resolve AI

play_circle_outline SRE burnout motivating Resolve AI founding

play_circle_outline Integration with OpenTelemetry and existing observability tools

play_circle_outline Resolve AI Agents: Autonomous Debugging, Human-Approved Execution, and Proactive Incident Prevention in High-Change Environments

play_circle_outline Company trajectory, funding, valuation, potential IPO aspirations

Info

Spiros Xanthos, Resolve AI

Spiros Xanthos

Founder & CEO Resolve AI

search

>> Palo Alto Studio connection, Silicon Valley and Wall Street. I'm John Furrier, host of theCUBE, here with Dave Vellante, my cohost.

Gemma Allen

>> Welcome back to theCUBE Studio here at New York Stock Exchange. I'm Gemma Allen with NYSC Wired mixture of experts. And today we are going to have a conversation about what is happening when the world of agentic AI meets production from the perspective of SRE and incident management. Joining me now is a man who has built a career in this space, Spiros Xanthos, founder and CEO of Resolve AI. Welcome, Spiros.

Spiros Xanthos

>> Thanks. Thank you for having me. I'm glad to be here.

Gemma Allen

>> So I love the founder journey, and when I looked at yours. It is quite interesting because you have already, I guess, built and exited two pretty impressive companies, both in the world of observability, I would say, to Resolve, to VMware, to Splunk, one of which is now somewhat of a competitor for this new venture you're on. Talk to me a little bit about Resolve AI and why you didn't just retire and go live out your life in Greece and .

Spiros Xanthos

>> That's where I'm from, right? Makes sense. So I am actually passionate about technology, right? The fact that I even started these two companies before indicates what I enjoy and I'm passionate about. Right? I truly love working with people that are motivated, solve hard technical problems, have impact. And I would say also throughout my career, all the products I helped create, including by the way, Open Telemetry, which is now the foundation of all modern observability, we're actually towards actually helping software engineers and site reliability engineers actually improve their life. Right? It's not just about having some financial impact, but also having impact in the lives of people. And the motivation for starting Resolve AI was our own experience at Splunk. So when I was running Splunk Observability, we had a period of six months where 90% of our site reliability engineering team resigned because of burnout.

Gemma Allen

>> Oh, wow.

Spiros Xanthos

>> It was insane how hard it was for us to keep our systems reliable given the scale and the complexity. So Mayank, my co-founder and I, thought there must be a better way and that better way for us didn't look like a 10th tool on top of the nine already have in monitoring observability infrastructure. We decided to build autonomous agents that essentially debug and troubleshoot production alerts and incidents with the goal of actually running all of production eventually. But we started with a problem that we felt was the most impactful and painful in that domain.

Gemma Allen

>> So let's talk about what has changed even since you've sold your last company and built Resolve AI, right? Because the world of agentic from the perspective of production environments is obviously quite different, right? We know that there has always been a lot of noise, a lot of signal. These are messy, they're multi-distributed. There has been somewhat challenge in terms of even actionable and being able to action some of the very intelligent insights you garner, right? What is shifting in this agentic world and how quickly is it shifting? How real is that theory versus the reality?

Spiros Xanthos

>> So I described it was hard before. Now it has become impossible for humans because think about the adoption of coding agents that made code creation 10X, and that adoption was faster than anything else we've seen in the world of technology before. I don't see any company out there, right, whether it's a technology company or more traditional enterprise that hasn't adopted coding tools. So now we're at the point where we produce code 10X, sometimes 1X faster, humans are less intimately familiar to push all of that on top of the existing complex systems. So that's why we believe that having agents that can help you debug and run production is essential. We cannot really move forward or adopt all this speed unless we can also run software as effectively as we produce it.

Gemma Allen

>> So let's say you are an enterprise customer, you have data log and Open Telemetry already in place. How would a customer work with Resolve AI? Is this another layer of the stock? Is it a replacement? How does the world of observability converge or not converge?

Spiros Xanthos

>> First of all, adopting Open Telemetry is a great thing for everybody, in my opinion, because it gives customers control over the data. It doesn't belong to anymore. It belongs to them. And maybe you have monitoring observability tools on top of that, but in reality, Resolve does not replace the tools. Resolve can use these tools as effectively or actually more effectively than a human. What the result replaces, it replaces the human effort for these very, very stressful, tedious and hard tasks so we can move a lot faster and humans maybe can focus on what they love or what is more important, which is maybe building and architecting and changing the future of their business. So no, Resolve is an intelligence layer on top of, not just observability, right, but off top of all the tools that you have in software operations, which is application, infrastructure, observability tools, even documentation, anything else a human has to use in running a software system.

Gemma Allen

>> So let's talk about this shift to full autonomous agents in this space. So you create signal, you know something is not working as it should in any specific workflow or environment. Does Resolve actually take action on behalf of the human, and where are those lines drawn?

Spiros Xanthos

>> Yes. So the starting point with Resolve is you essentially connect Resolve, and Resolve starts doing the work of humans the way humans do it today because also the tools you have in place were designed to be used by humans. So Resolve can be on call as we call it, meaning every time something goes wrong, Resolve is going to pick up a customer complaint or an alert. It will do a full investigation. It will give you the root cause, and it will give you the remediation. And Resolve actually can take the action as well, can fix the problem. Now that depends a bit on the willingness of the customer to accept an AI taking action, but the ability is there. What we see in practice is the more advanced customers probably are getting now to a point where they want Resolve to take action for most of the, let's say, low risk changes, and then maybe they want Resolve to produce the action by the human to approve it for the high risk changes. But that's not the end, right? Resolve so now can, for example, monitor every change, every, let's say, code change that lands the production proactively. So not only it responds to problems, but it starts actually preventing things from escalating into a problem way earlier. Something that obviously it's impossible for humans to do, right? We have customers like Coinbase, DoorDash, Salesforce, Zscaler, with thousands of engineers, producing probably hundreds of thousands of changes a day. It's impossible for humans to keep up with that, right? Where agents like Resolve can actually monitor everything and prevent problems as well.

Gemma Allen

>> Let's talk about some of the new tech that's hitting production environment from things like vibe coding from this world where suddenly everyone is a developer and everyone's a technologist, which we know creates a lot of opportunity, but also a lot of risk, right, if we're being pragmatic about it. You said you're seeing hundreds of thousands of incidents, millions of incidents in any given day or moment. What sorts of new threats or new, I guess, naivetes are you seeing in environments?

Spiros Xanthos

>> I would say there are many citizen developers, and that's very desirable. And I would say in my opinion, AI is not going to reduce the number of professional developers either. I think we're going to have a lot more developers because we're going to produce a lot more technology, which is beneficial to the world. Now, I would say there are two things that are happening, right? Even professional developers, because they produce a lot more code through agents, they're probably less intimately familiar with all that code that hits production. So that's one area where regardless of whether you're professional or not, you need agents to help you run that software because you're not as familiar as before and it's a lot more. But even for citizen developers, let's say, if things they produce end up actually becoming critical and production-ready, the same rules supply and obviously you need tools for that as well. Oftentimes what I observe is non-professional developers produce things that maybe run on the laptop, right? Maybe they're not as critical to the end customers. So then maybe that part of software is less critical but still very desirable to do.

Gemma Allen

>> So let's talk about DevOps for a second, a space that also is undergoing some change, right? It's having a moment for sure. You're at AWS Summit this week, so are we had some conversations with some folks who are using DevOps agent in production. And it struck me that there's definitely a mixed sentiment amongst some of the DevOps community broadly about how far we should go with the world of agentic, especially from the perspective of actually implementing changes, right? Really having control over environments. Some folks love the idea of not having to be woken up on a Saturday morning with an incident, and some folks are worried about what it means in the longer term and how the ownership changes, right? What are your thoughts from the perspective of, again, where human value starts to creep or potentially wane?

Spiros Xanthos

>> Yeah. Like I said, I don't think we're going to have fewer software engineers, but the job is changing. The job is not writing the code or even debugging the incidents and production use anymore. The job is managing the agents that do that and do a lot more of it. But I'll tell you, this is a very hard part of the job. I just came before this, but from meeting the head of SRE from a major consumer technology company. And the thing that he told me that was the most important to him is actually Resolve fixing maybe the majority of issues and not having to wake somebody in the middle of the night. He said the human aspect of running production maybe is the most important to him, more than even improving reliability or more than the productivity gains you have. So I think people are ready for agents to do this well, but of course we are developing the right guard lays, the right policies, all the right mechanisms to do this safely and compliant in a way that aligns with the way folks want to run businesses.

Gemma Allen

>> Let's talk about the world of LLMs for a second, specifically as it relates to your business, right? So how much of the future do you think will be built on agents being very, very neat, knowing open to them too really, really well versus just better LLMs also absorbing some of that input and output and actually also delivering that work? Where do you see that split or that convergence happening?

Spiros Xanthos

>> So I see obviously the improvement in general reasoning continuing. We see this with the recent models that dropped, and that generally helps you build more effective agents because you can essentially chain together many more tool calls and the models don't lose track of what's happening. But if we step back, I would say for domains who are deep like software, like customer service, like coding, what we see is that I think that you can advance the state-of-the-art by training specialized models and building agents together with these models beyond what you can do with just frontier models. This doesn't mean the general purpose frontier models are getting out of the picture. It just means that you can build a model for your domain and build essentially a more effective full stack solution across models and agents to do this better. And that's our thesis. Resolve has established a lab whose goal is to essentially advance the state-of-the-art in production operations. That doesn't mean we'll fully replace or compete even with the large LLMs, but it means that we're going to models that we expect to be better and cheaper for the domain.

Gemma Allen

>> And in that lab, that oriented split, the division of time and energy spent, how much of it is spent on complete proprietary new, unique builds for Resolve AI versus even just keeping a pace with some of what's happening in the world of LLMs and frontier models broadly, is it?

Spiros Xanthos

>> Yeah.

Gemma Allen

>> I imagine there's somewhat of a divide there.

Spiros Xanthos

>> Yes. R&D team has two goals, right? To build the best product, best solution at the end of the day for production operations, for running software. But also to do that, we need to stay at the frontier of the technology, right? Our lab, I would say broadly speaking, has two goals, right, to develop models that are, let's say, state-of-the-art in the domain, but also to actually experiment constantly in figuring out how to have agents that are the most effective given all the technology, right? Not just our own models, but the frontier models. And of course, given the pace of change, I would say that's a very hard task and probably impossible for our end users to do it on their own. Every week there is new model with new capabilities, you need to change the whole stack oftentimes, right? You need to improve your eval sets, as we call them, to be able to test the new model abilities. You need to modify your harnesses to take advantage of it, right? So our lab, let's say broadly speaking, has the responsibility for both.

Gemma Allen

>> What are you seeing from the perspective of open weight models and how do you see them playing a role in the future of enterprise tech and corporate America?

Spiros Xanthos

>> Obviously the open weight models are the way open source in the past generation were important are equally important in my opinion. I think that we should not get into the world ... I don't think that's sustainable world, right? Where all the value gets extracted by the LLMs, right? And I think the world that actually, in my opinion, is going to be the future is where obviously the LLMs are an important part of the stack, but alongside them, there are open weight models that enterprises and vendors can modify and produce a better solution or a more specialized solution to their needs. And I think what we see now is that open weight models are probably six months behind the frontier models. And if that continues and it improves, I think we're heading to a very good future and that's my belief, right? My belief is that open-wide models will stay close and will even get closer to the closed frontier models. And one more point I want to make there is that I think like the big labs want to make everybody believe that it's impossible to train a state-of-the-art model, but I don't think that's the case, and I think we're seeing it from our own efforts and we're seeing it from others in the industry that you can actually ... Having, let's say the right domain knowledge, having the right data, you can actually produce a model using open weight models as a starting point maybe that is more effective in your domain.

Gemma Allen

>> The race is on. So we hear a lot about bottlenecks and constraints in the infrastructure space, right? We hear about compute, about energy, about all of those things. From your perspective, what are the biggest constraints right now for you building out this business?

Spiros Xanthos

>> Yes, I guess as the build of the infrastructure happens in a way, and we've seen this with the stock of all the hardware companies and the memory companies, it's moving gradually up the stack, right? In our case, there is constraint in GPU availability, right? And because we're training our own models, we need a lot of infrastructure for that, but it is a constraint for the industry. Now, the good news in our case is our domain, we're essentially trying to automate software engineering work, which is extremely valuable, right? I think we can afford to invest, have the right capital investment, let's say, to acquire the infrastructure that we need, but still it's very, very constrained. And I think the world is going to be constrained for a while for the next two three years at least.

Gemma Allen

>> Let's stay on investment and finish with an easy question. So Spiros, you have built and exited two companies. This is your third company. Very impressive financials too, right? You guys have valuation of 1.5 billion. You had an extension to a Series A. I mean, it's an impressive pedigree and resume here you have. Are we going to see you ring the bell here at the New York Stock Exchange someday or what's the plan with this one?

Spiros Xanthos

>> I hope so. I think the difference between this company and my other two companies is that this is such a big technology shift that essentially I think the whole software stack eventually is going to be replaced, and it's going to be agentic and we see the unbundling of SaaS vendors and all of that. So I think this creates obviously the opportunity to build something very impactful, right? What we're solving and mapping is universal. Anybody who runs production software that sells to customers wants to do that better, wants to do it more reliably. And the way for that is AI. So our aspiration is to be the most impactful companies in, let's say, the production software operations domain.

Gemma Allen

>> Well, we certainly hope we see you ring the bell too, Spiros. Thank you so much for joining us on theCUBE.

Spiros Xanthos

>> Thank you. Thanks for having me.

Gemma Allen

>> I'm Gemma Allen at theCUBE Studio here at the New York Stock Exchange. This is NYSC Wired's mixture of experts. Thanks for watching.

Spiros Xanthos, Resolve AI

search

>> Palo Alto Studio connection, Silicon Valley and Wall Street. I'm John Furrier, host of theCUBE, here with Dave Vellante, my cohost.

Gemma Allen

>> Welcome back to theCUBE Studio here at New York Stock Exchange. I'm Gemma Allen with NYSC Wired mixture of experts. And today we are going to have a conversation about what is happening when the world of agentic AI meets production from the perspective of SRE and incident management. Joining me now is a man who has built a career in this space, Spiros Xanthos, founder and CEO of Resolve AI. Welcome, Spiros.

Spiros Xanthos

>> Thanks. Thank you for having me. I'm glad to be here.

Gemma Allen

>> So I love the founder journey, and when I looked at yours. It is quite interesting because you have already, I guess, built and exited two pretty impressive companies, both in the world of observability, I would say, to Resolve, to VMware, to Splunk, one of which is now somewhat of a competitor for this new venture you're on. Talk to me a little bit about Resolve AI and why you didn't just retire and go live out your life in Greece and .

Spiros Xanthos

>> That's where I'm from, right? Makes sense. So I am actually passionate about technology, right? The fact that I even started these two companies before indicates what I enjoy and I'm passionate about. Right? I truly love working with people that are motivated, solve hard technical problems, have impact. And I would say also throughout my career, all the products I helped create, including by the way, Open Telemetry, which is now the foundation of all modern observability, we're actually towards actually helping software engineers and site reliability engineers actually improve their life. Right? It's not just about having some financial impact, but also having impact in the lives of people. And the motivation for starting Resolve AI was our own experience at Splunk. So when I was running Splunk Observability, we had a period of six months where 90% of our site reliability engineering team resigned because of burnout.

Gemma Allen

>> Oh, wow.

Spiros Xanthos

>> It was insane how hard it was for us to keep our systems reliable given the scale and the complexity. So Mayank, my co-founder and I, thought there must be a better way and that better way for us didn't look like a 10th tool on top of the nine already have in monitoring observability infrastructure. We decided to build autonomous agents that essentially debug and troubleshoot production alerts and incidents with the goal of actually running all of production eventually. But we started with a problem that we felt was the most impactful and painful in that domain.

Gemma Allen

>> So let's talk about what has changed even since you've sold your last company and built Resolve AI, right? Because the world of agentic from the perspective of production environments is obviously quite different, right? We know that there has always been a lot of noise, a lot of signal. These are messy, they're multi-distributed. There has been somewhat challenge in terms of even actionable and being able to action some of the very intelligent insights you garner, right? What is shifting in this agentic world and how quickly is it shifting? How real is that theory versus the reality?

Spiros Xanthos

>> So I described it was hard before. Now it has become impossible for humans because think about the adoption of coding agents that made code creation 10X, and that adoption was faster than anything else we've seen in the world of technology before. I don't see any company out there, right, whether it's a technology company or more traditional enterprise that hasn't adopted coding tools. So now we're at the point where we produce code 10X, sometimes 1X faster, humans are less intimately familiar to push all of that on top of the existing complex systems. So that's why we believe that having agents that can help you debug and run production is essential. We cannot really move forward or adopt all this speed unless we can also run software as effectively as we produce it.

Gemma Allen

>> So let's say you are an enterprise customer, you have data log and Open Telemetry already in place. How would a customer work with Resolve AI? Is this another layer of the stock? Is it a replacement? How does the world of observability converge or not converge?

Spiros Xanthos

>> First of all, adopting Open Telemetry is a great thing for everybody, in my opinion, because it gives customers control over the data. It doesn't belong to anymore. It belongs to them. And maybe you have monitoring observability tools on top of that, but in reality, Resolve does not replace the tools. Resolve can use these tools as effectively or actually more effectively than a human. What the result replaces, it replaces the human effort for these very, very stressful, tedious and hard tasks so we can move a lot faster and humans maybe can focus on what they love or what is more important, which is maybe building and architecting and changing the future of their business. So no, Resolve is an intelligence layer on top of, not just observability, right, but off top of all the tools that you have in software operations, which is application, infrastructure, observability tools, even documentation, anything else a human has to use in running a software system.

Gemma Allen

>> So let's talk about this shift to full autonomous agents in this space. So you create signal, you know something is not working as it should in any specific workflow or environment. Does Resolve actually take action on behalf of the human, and where are those lines drawn?

Spiros Xanthos

>> Yes. So the starting point with Resolve is you essentially connect Resolve, and Resolve starts doing the work of humans the way humans do it today because also the tools you have in place were designed to be used by humans. So Resolve can be on call as we call it, meaning every time something goes wrong, Resolve is going to pick up a customer complaint or an alert. It will do a full investigation. It will give you the root cause, and it will give you the remediation. And Resolve actually can take the action as well, can fix the problem. Now that depends a bit on the willingness of the customer to accept an AI taking action, but the ability is there. What we see in practice is the more advanced customers probably are getting now to a point where they want Resolve to take action for most of the, let's say, low risk changes, and then maybe they want Resolve to produce the action by the human to approve it for the high risk changes. But that's not the end, right? Resolve so now can, for example, monitor every change, every, let's say, code change that lands the production proactively. So not only it responds to problems, but it starts actually preventing things from escalating into a problem way earlier. Something that obviously it's impossible for humans to do, right? We have customers like Coinbase, DoorDash, Salesforce, Zscaler, with thousands of engineers, producing probably hundreds of thousands of changes a day. It's impossible for humans to keep up with that, right? Where agents like Resolve can actually monitor everything and prevent problems as well.

Gemma Allen

>> Let's talk about some of the new tech that's hitting production environment from things like vibe coding from this world where suddenly everyone is a developer and everyone's a technologist, which we know creates a lot of opportunity, but also a lot of risk, right, if we're being pragmatic about it. You said you're seeing hundreds of thousands of incidents, millions of incidents in any given day or moment. What sorts of new threats or new, I guess, naivetes are you seeing in environments?

Spiros Xanthos

>> I would say there are many citizen developers, and that's very desirable. And I would say in my opinion, AI is not going to reduce the number of professional developers either. I think we're going to have a lot more developers because we're going to produce a lot more technology, which is beneficial to the world. Now, I would say there are two things that are happening, right? Even professional developers, because they produce a lot more code through agents, they're probably less intimately familiar with all that code that hits production. So that's one area where regardless of whether you're professional or not, you need agents to help you run that software because you're not as familiar as before and it's a lot more. But even for citizen developers, let's say, if things they produce end up actually becoming critical and production-ready, the same rules supply and obviously you need tools for that as well. Oftentimes what I observe is non-professional developers produce things that maybe run on the laptop, right? Maybe they're not as critical to the end customers. So then maybe that part of software is less critical but still very desirable to do.

Gemma Allen

>> So let's talk about DevOps for a second, a space that also is undergoing some change, right? It's having a moment for sure. You're at AWS Summit this week, so are we had some conversations with some folks who are using DevOps agent in production. And it struck me that there's definitely a mixed sentiment amongst some of the DevOps community broadly about how far we should go with the world of agentic, especially from the perspective of actually implementing changes, right? Really having control over environments. Some folks love the idea of not having to be woken up on a Saturday morning with an incident, and some folks are worried about what it means in the longer term and how the ownership changes, right? What are your thoughts from the perspective of, again, where human value starts to creep or potentially wane?

Spiros Xanthos

>> Yeah. Like I said, I don't think we're going to have fewer software engineers, but the job is changing. The job is not writing the code or even debugging the incidents and production use anymore. The job is managing the agents that do that and do a lot more of it. But I'll tell you, this is a very hard part of the job. I just came before this, but from meeting the head of SRE from a major consumer technology company. And the thing that he told me that was the most important to him is actually Resolve fixing maybe the majority of issues and not having to wake somebody in the middle of the night. He said the human aspect of running production maybe is the most important to him, more than even improving reliability or more than the productivity gains you have. So I think people are ready for agents to do this well, but of course we are developing the right guard lays, the right policies, all the right mechanisms to do this safely and compliant in a way that aligns with the way folks want to run businesses.

Gemma Allen

>> Let's talk about the world of LLMs for a second, specifically as it relates to your business, right? So how much of the future do you think will be built on agents being very, very neat, knowing open to them too really, really well versus just better LLMs also absorbing some of that input and output and actually also delivering that work? Where do you see that split or that convergence happening?

Spiros Xanthos

>> So I see obviously the improvement in general reasoning continuing. We see this with the recent models that dropped, and that generally helps you build more effective agents because you can essentially chain together many more tool calls and the models don't lose track of what's happening. But if we step back, I would say for domains who are deep like software, like customer service, like coding, what we see is that I think that you can advance the state-of-the-art by training specialized models and building agents together with these models beyond what you can do with just frontier models. This doesn't mean the general purpose frontier models are getting out of the picture. It just means that you can build a model for your domain and build essentially a more effective full stack solution across models and agents to do this better. And that's our thesis. Resolve has established a lab whose goal is to essentially advance the state-of-the-art in production operations. That doesn't mean we'll fully replace or compete even with the large LLMs, but it means that we're going to models that we expect to be better and cheaper for the domain.

Gemma Allen

>> And in that lab, that oriented split, the division of time and energy spent, how much of it is spent on complete proprietary new, unique builds for Resolve AI versus even just keeping a pace with some of what's happening in the world of LLMs and frontier models broadly, is it?

Spiros Xanthos

>> Yeah.

Gemma Allen

>> I imagine there's somewhat of a divide there.

Spiros Xanthos

>> Yes. R&D team has two goals, right? To build the best product, best solution at the end of the day for production operations, for running software. But also to do that, we need to stay at the frontier of the technology, right? Our lab, I would say broadly speaking, has two goals, right, to develop models that are, let's say, state-of-the-art in the domain, but also to actually experiment constantly in figuring out how to have agents that are the most effective given all the technology, right? Not just our own models, but the frontier models. And of course, given the pace of change, I would say that's a very hard task and probably impossible for our end users to do it on their own. Every week there is new model with new capabilities, you need to change the whole stack oftentimes, right? You need to improve your eval sets, as we call them, to be able to test the new model abilities. You need to modify your harnesses to take advantage of it, right? So our lab, let's say broadly speaking, has the responsibility for both.

Gemma Allen

>> What are you seeing from the perspective of open weight models and how do you see them playing a role in the future of enterprise tech and corporate America?

Spiros Xanthos

>> Obviously the open weight models are the way open source in the past generation were important are equally important in my opinion. I think that we should not get into the world ... I don't think that's sustainable world, right? Where all the value gets extracted by the LLMs, right? And I think the world that actually, in my opinion, is going to be the future is where obviously the LLMs are an important part of the stack, but alongside them, there are open weight models that enterprises and vendors can modify and produce a better solution or a more specialized solution to their needs. And I think what we see now is that open weight models are probably six months behind the frontier models. And if that continues and it improves, I think we're heading to a very good future and that's my belief, right? My belief is that open-wide models will stay close and will even get closer to the closed frontier models. And one more point I want to make there is that I think like the big labs want to make everybody believe that it's impossible to train a state-of-the-art model, but I don't think that's the case, and I think we're seeing it from our own efforts and we're seeing it from others in the industry that you can actually ... Having, let's say the right domain knowledge, having the right data, you can actually produce a model using open weight models as a starting point maybe that is more effective in your domain.

Gemma Allen

>> The race is on. So we hear a lot about bottlenecks and constraints in the infrastructure space, right? We hear about compute, about energy, about all of those things. From your perspective, what are the biggest constraints right now for you building out this business?

Spiros Xanthos

>> Yes, I guess as the build of the infrastructure happens in a way, and we've seen this with the stock of all the hardware companies and the memory companies, it's moving gradually up the stack, right? In our case, there is constraint in GPU availability, right? And because we're training our own models, we need a lot of infrastructure for that, but it is a constraint for the industry. Now, the good news in our case is our domain, we're essentially trying to automate software engineering work, which is extremely valuable, right? I think we can afford to invest, have the right capital investment, let's say, to acquire the infrastructure that we need, but still it's very, very constrained. And I think the world is going to be constrained for a while for the next two three years at least.

Gemma Allen

>> Let's stay on investment and finish with an easy question. So Spiros, you have built and exited two companies. This is your third company. Very impressive financials too, right? You guys have valuation of 1.5 billion. You had an extension to a Series A. I mean, it's an impressive pedigree and resume here you have. Are we going to see you ring the bell here at the New York Stock Exchange someday or what's the plan with this one?

Spiros Xanthos

>> I hope so. I think the difference between this company and my other two companies is that this is such a big technology shift that essentially I think the whole software stack eventually is going to be replaced, and it's going to be agentic and we see the unbundling of SaaS vendors and all of that. So I think this creates obviously the opportunity to build something very impactful, right? What we're solving and mapping is universal. Anybody who runs production software that sells to customers wants to do that better, wants to do it more reliably. And the way for that is AI. So our aspiration is to be the most impactful companies in, let's say, the production software operations domain.

Gemma Allen

>> Well, we certainly hope we see you ring the bell too, Spiros. Thank you so much for joining us on theCUBE.

Spiros Xanthos

>> Thank you. Thanks for having me.

Gemma Allen

>> I'm Gemma Allen at theCUBE Studio here at the New York Stock Exchange. This is NYSC Wired's mixture of experts. Thanks for watching.