The Future of Software! When AI Becomes Your Reliability Team | Spiros Xanthos

Vit Lyoshin (00:01.124)
Welcome Spiros, nice to have you here today.

Spiros Xanthos (00:04.881)
Hi, thanks for hosting me.

Vit Lyoshin (00:07.684)
Yeah, yeah, absolutely. So let's start with your company, Resolve AI, and why don't we start with defining what the problem you're solving for your customers.

Spiros Xanthos (00:21.474)
Yeah. So Resolve AI is building agents that, you know, work like, let's say as an AI SRE, if you wish, right? The main goal of the product is to be on call on behalf of humans and help troubleshoot, you know, find the root cause of alerts and incidents and help remediate those. So if you are a software engineer or an SRE that, you know, has to wake up in the middle of the night, Resolve should help you maybe avoid that actually page altogether.

or if you have to wake up, get, let's say, to an answer and back to sleep as quickly as possible. But also, let's say for somebody who is maybe a manager or a CTO in an organization, the idea is that we want the result to be able to do all these, let's say, essential, but work that doesn't necessarily advance the product so that human software engineers can help, can focus on building new features or new products.

Vit Lyoshin (01:15.596)
I see, okay, so it's basically observability tool to help automatically resolve certain issues in the product.

Spiros Xanthos (01:24.566)
It's not observability in the sense that, you know, I, my own background is observability. I have to share that, but you know, traditionally observability tools, you know, collect data and, know, they provide the ability to create dashboards, create alerts so that we as humans, when something goes wrong, can let's say use these tools to figure it out, right? Resolve actually is not a destination for all this data instead is, can connect to all the tools that a human has. can connect to GitHub, can connect to observability, can connect to your cloud infrastructure.

Vit Lyoshin (01:30.318)
Yeah.

Vit Lyoshin (01:37.125)
Mm-hmm.

Spiros Xanthos (01:53.312)
and use all these tools the way a human engineer would, With the main goal of troubleshooting and getting to the root cause of alerts and incidents. So it replaces, let's say, human effort, especially this kind of tedious, stressful work, so that you can improve the time it takes to resolve incidents, improve reliability, and generally avoid escalations and avoid having humans do all this work instead of building.

Vit Lyoshin (02:21.168)
Got it, okay, so does it actually go and change the code and commit fixes or does it do some sort of configuration changes in the environment?

Spiros Xanthos (02:32.215)
So the first of all, the way resolve works is very simple. You can plug it into all the tools that you have as a human. And then it uses all these tools to build a pretty detailed understanding of the environment and does this constantly, right? It always runs in the background. Every time there is a new change and you've got your ration chains, it picks it up. And when it needs to investigate a problem, it has all this context available so that it can do it. Typically, when something goes wrong, the first thing resolve will do alongside humans is going to try to determine what happened and what is the root cause, right?

Vit Lyoshin (02:52.516)
Mm-hmm.

Spiros Xanthos (03:02.538)
Once you determine that, the next step is usually to remediate. Remediation oftentimes comes in the form of, let's say, maybe rolling back a change, maybe changing something in the infrastructure. But oftentimes, it's something that is wrong with the code. So in those scenarios, Resolve can generate actually PR or can generate the code that will do the fix.

Vit Lyoshin (03:20.908)
I see, okay, so it's basically like really like a human and it can analyze and figure out the solution for the issue. Okay, cool. So lately there's been a lot of AI coding and wipe coding tools on the market. How this one is different, if it is different.

Spiros Xanthos (03:41.487)
Yeah, it is different because our main goal at Resolve AI is code is obviously an input and deep understanding of code is important to the end goal we have here. But at the end of the day, our goal is to help humans run production systems, starting with obviously incidents of troubleshooting, but not stopping there. Our goal is to be able to understand the production system end to end, understand all the tools that humans use to run a production system, a software production system, and then assist with all sorts of tasks that...

again, today humans have to use. In that sense, the difference, let's say from coding agents is that we connect to many, many more tools, right? We'll use GitHub, but we will use also logs, we'll use metrics, we'll use traces, we'll use infrastructure knowledge, we'll connect to like your feature flag system. So the goal here for us is to be able to have this end-to-end understanding and to perform tasks that typically span multiple of these like categories of tools. And to be honest,

Vit Lyoshin (04:38.992)
Mm-hmm.

Spiros Xanthos (04:40.716)
We see ourselves more as a counterpart to the coding agents. I think we're headed into the future where the majority of the code is probably going to be generated from humans supervising, let's say, agents. And to me, that's a great thing. We can produce a lot more technology. We can solve harder problems, but also we cannot move faster unless we also address the bottleneck at the subsequent steps. Right. And maybe if you were to look at how much time a software engineer spends a day.

coding versus essentially dealing with all sorts of other things. The reality is that at scale, the majority of software engineering time is spent in dealing, let's say, with production related things, right? Not just the incidents, but compliance, security, changes. So our goal is obviously as a company to assist in making all this go faster in addition to code generation that happens at the front.

Vit Lyoshin (05:23.152)
Mm-hmm.

Vit Lyoshin (05:35.726)
Yeah, yeah, I think that that's like a key differentiator here because coding tools, they just generate a bunch of code and you still need engineers to analyze it, plug it into the right places, right? And I think you're doing it slightly differently. Okay, I see. And then also, do you have like human in the loop with this when the resolve AI suggests some changes, does it just apply them or somebody will have to like approve them for pull requests, for example?

Spiros Xanthos (05:46.648)
Correct.

Spiros Xanthos (06:01.134)
To be honest, that depends a bit on the tolerance of the humans. But generally speaking, let's say the most common pattern I see is that still today we're at the point where this is not different than self-driving cars. It's not enough that the car drives more safely than a human. It has to be like superhuman in its abilities for people to trust it. So for the most part, humans want us to like find the answer, want to suggest what needs to happen next. But usually

Vit Lyoshin (06:04.548)
Rock on.

Spiros Xanthos (06:29.996)
they want to be the last step in that chain, let's say. Now, but I think we're moving towards the next step now where it's, think the situation is going to be that results should be able to essentially take actions that are safe, that are reversible, that don't have like a lot of side effects potentially and report back to a human. And whenever there is a more risky change or a reversible change, then it's probably when humans are going to be involved. And of course,

Vit Lyoshin (06:32.495)
Mm-hmm.

Vit Lyoshin (06:58.573)
Mm-hmm.

Spiros Xanthos (06:59.426)
the more the quality of answers improves, the easier a lot of these next steps become.

Vit Lyoshin (07:04.536)
Yeah, yeah. And then also, sometimes we hear people don't really trust AI tools or agents yet. How do you convince your customers, like by being transparent or some sort of policies maybe you have, privacy, how you deal with privacy in the coding, in the repos?

Spiros Xanthos (07:21.742)
Yeah. First of all, when we started the company a year and a half ago, one of the questions, main questions we had is, was are people even going to trust AI in production? And what I observed over the past year and half is that, you know, as an industry, we're getting more and more comfortable having agents do work on our behalf in production. Now that said, it is very, very important to establish trust and not do anything that will also break the trust. So in our case.

We always try to ground any answer we give on real evidence. This evidence is presented to the human. Actually, the agent itself is using the evidence to always confirm and validate, let's say, that the answer is grounded on, let's say, real world data. We also try to, when we're not confident about an answer, we actually try to not give the answer. We try to share that our investigation did not yield any high confidence, let's say, theories.

So generally we try to follow all sorts of mechanisms to create trust with humans so that they trust the system more and more and they can feel like they can follow its suggestions. But beyond that, course, security, compliance, ensuring that it has minimal access to the data. doesn't essentially replicate data that it doesn't have to. It only looks at the data in runtime when it has to. All of these are very important things that we try to follow.

Vit Lyoshin (08:44.834)
I see. Were there any specific technical challenges building this AI tool that you can mention?

Spiros Xanthos (08:54.639)
So to give you my background briefly, this is my fourth startup. I previously built two other observability tools. I built a log analysis tool out of my PhD, a product called Loginsight that was acquired by VMware in 2012. And still today is VMware's log analytics platform. And then in 2018, together with Mayank, who's my co-founder at Resolve AI as well, we co-created OpenTelemetry and we built an observability platform around it. At the time our thesis was that

Vit Lyoshin (09:11.738)
Mm-hmm.

Spiros Xanthos (09:25.295)
our ability to let's say build powerful observability tools is limited by the fact that every vendor has their own proprietary agent and only sees a subset of the data. So open telemetry was our way of moving towards an open source, open standards way of collecting all this data. And so that we can build more powerful backends that have access to all the data. So I built essentially multiple generations of these tools. My last company was acquired by Splunk at Splunk and I managed like all of Splunk observability.

also there many new products. And I'm giving you all this history because Resolve is by far the hardest product I ever tried to build. We have a very strong team, which is a combination of, let's say, observability and systems experts and folks with deep AI expertise. We have a large number of people who joined us from deep mind, people who worked there on deep research and similar agentic frameworks or products. And it is a very hard problem. The reason is that

Of course, models make something like this feasible, but still they don't make it easy, especially because you're dealing with multiple different types of data. The data is not connected. Even a human has a hard time connecting the data or reasoning across code, logs, metrics, changes. So yes, the models maybe can move more quickly and can test a lot of hypotheses much faster, but still the reasoning that you need to do here is very, very hard. And the models are not really trained for this type of work, right?

So we to a lot of work on planning and reasoning and actually analysis of all this data that the models cannot do well. So it is a very hard problem in all the dimensions. Even the connectivity to the tools is a hard problem because many of these tools were not designed to be used by agents, right? So you have to be very creative in making the agent use the tools in a way that doesn't create more load than what a human would create.

Vit Lyoshin (11:21.54)
Yeah, I can imagine you have to analyze a bunch of things. And like you said, you have to build your own maybe prompts or maybe some analysis of different sources and then spin out some options how to resolve it. Yeah, that could be challenging. Sure.

Spiros Xanthos (11:38.832)
I'll give you a very simple example. Let's say you're dealing with a problem that requires to analyze logs, In practice, there is an unlimited number of logs that you know can contain or information about the problem, right? So you're very, very quickly going to run out of the context window in the model as well, right? So it is a fairly complicated problem even for the simple task, right? Just the task of, okay, go find some logs that might give us evidence about what's going on. Even that simple question is very, very hard to...

Vit Lyoshin (11:51.632)
Mm-hmm.

Vit Lyoshin (12:05.175)
Mm-hmm.

Spiros Xanthos (12:08.665)
to do and scale using models.

Vit Lyoshin (12:11.608)
Right, What do you think will happen in the nearest future? Will we still have engineers actually going through and troubleshooting and maintaining or agents will replace all of them?

Spiros Xanthos (12:26.223)
First of all, I think we're going to have more engineers than we have today because I think we're going to have a lot more technology and a lot more software in the future. I believe that software engineering is changing and is changing so that we will produce a lot more software, a lot more cheaply, maybe like a hundred times. Two or three orders of magnitude more software. As a result, I don't think we're going to have fewer people working in technology. If anything, we're going to have more people working in technology because I think the floor is going to be lower, but also the ceiling is going to be higher.

be able to produce, you know, solve harder problems using software. but I think the way we work is, is changing and it's going to change even more in that we have to be able to use all these tools. think that. Essentially the agents are going to provide a new layer of abstraction that, and humans will operate, you know, at one layer higher in the stack, let's say. And, for most of the tasks we have to manually do today, probably it's going to be a lot faster to direct an agent to do it.

Vit Lyoshin (13:18.8)
Mm-hmm.

Spiros Xanthos (13:24.463)
going to be faster, it's going to probably be more accurate. And then humans are going to be directing or organizing the agents. And the skills that we're going to need for that are going to be slightly different potentially. systems thinking, the technical fundamentals are going to be very important, but maybe knowledge of specific language is going to be less critical than it is today.

Vit Lyoshin (13:47.152)
Yeah, yeah, I can see that. So you think more people should be focusing on like computer science moving forward. It's like a hundred years ago people, some people couldn't read for example and pretty much today everybody can read. So we can safely say in 10-15 years it will be same as coding.

Spiros Xanthos (14:07.428)
mean, you know, nobody can predict what's going to happen in 10 years, but generally I'm an optimist. think that I don't want to minimize the challenges that come with AI, right? And the potential kind of automation of things that humans do today and, you know, the displacement of like maybe roles that AI is going to automate. But when it comes to technology specifically, I do believe that we're going to be producing a lot more technology and as long as humans

Vit Lyoshin (14:11.139)
Right.

Spiros Xanthos (14:36.77)
adapt, say, right, and learn how to use AI and be effective with it. I do think there is actually a pretty bright future there.

Vit Lyoshin (14:43.906)
Okay, yeah, okay. So in this space where you work and you're trying to solve this problem, help out people, do you see any other unsolved gaps that maybe somebody else can help or your company can help?

Spiros Xanthos (14:58.928)
There are many. I'll give you an example of something that is adjacent to what we do and it's not something we plan to address. I think a big limitation we have today in terms of how far we can go with agents. mean, there is limitation in terms of reasoning, accuracy, inference and all of that. Right. But as this becomes better and better, I think the bottleneck or a big bottleneck we have today is that a lot of the tools that we have at our disposal were designed for humans and humans operate at a certain speed, right. And they can do things serially.

I think there is going to be a need for the underlying tools, the systems of record or the tools that humans use today to perform various tasks to adapt, to be more friendly to agents, or they might be replaced by tools that are actually agent first, let's say, than human first. And I do think that there is a pretty big opportunity, pretty much in every sector of the world where agents have application. In addition, of course, to many other...

problems in the agentic or AI space that can be automated. But I find this actually very interesting that the underlying infrastructure has to evolve.

Vit Lyoshin (16:04.036)
Yeah. So you're talking with DevOps probably and CTOs and many companies. How is their general feeling towards AI and agents automating stuff for coding and maintaining code bases?

Spiros Xanthos (16:21.146)
So the, I think the interesting thing about what the resolve does is that it's great, let's say for AI to do creative work. It's great for AI to write code and accelerate that part. But what's even better in our view is for AI to do the tedious work and deal with the toil. And a lot of the problems around, let's say incidents and troubleshooting and production and on call are actually very stressful and tedious. So whether you're a non-call engineer and have to wake up in the middle of the night.

Or whether you're a CTO, say, right, whose team essentially spends a lot of time, oftentimes the majority of time, in maintenance, you really want AI to help accelerate these things. There's not much debate whether people want this type of work automated. And hopefully giving people back time to do more creative work. Now, when it comes to more creative work, maybe there is a bigger debate there. But in our case, I mean, what I really enjoy also, and I feel like...

excited about is that this is work that, you know, humans don't really want to do. And it's very stressful. And, it, it, it is not something that anybody really or rarely somebody enjoys doing.

Vit Lyoshin (17:32.655)
Yeah, yeah, I remember myself. I used to be that guy who wakes up in the middle of the night, troubleshooting something for customer. That's not very pleasant for sure. So yeah, a little bit going back to your entrepreneurial journey. As you mentioned, you founded a couple of other companies in the past and now with Resolve AI and you mentioned this is the hardest one by far. So...

But in general, like how your approach changed, maybe evolved as you moved from one company to another and some lessons, maybe you applied here to help make Yeah.

Spiros Xanthos (18:08.313)
Many lessons. definitely avoid the same mistakes. I make new mistakes every time, but I'll tell you like there are a few principles that, over the years I concluded are the right principles in building a company. I'm a huge believer in full transparency. think oftentimes founders make the mistake that thinking they will protect their team by not sharing all the hard problems or all the difficulties that happen at the startup.

Vit Lyoshin (18:13.206)
Hahaha

Spiros Xanthos (18:37.393)
And there are many, many problems at the startup. The problems are a lot more usually than the good things. And they do it from a good intention, right? Like I said, to protect people. But I think that doesn't help anyone because essentially if you don't have like the full understanding of what's going on, then you're not going to work towards solving the hardest problems. And also when things become really hard, know, it's going to, people are going to be surprised, right? Because suddenly finding out that there are maybe bigger challenges than they thought. So I'm a huge believer in transparency.

I'm a huge believer in ownership and extreme ownership. In fact, like I don't want to tell anybody what to do. Ideally, I wanted to agree on outcomes and I want to agree on like a few goals that are very important to the company. And I believe the best way to execute is for every individual to fully own their work aligned to these goals and have both the authority and the responsibility for, for executing those goals. I'm also a huge believer in, healthy risk taking. I try very hard to create a,

let's say an environment of psychological safety where it is okay to fail. Now it's not desirable to fail, but also if you never fail, it's a sign that you're not taking enough risk as well. So we try to experiment with a lot of things, especially in the area we're working. It is brand new, right? Like many of the problems we're solving are practically research problems that nobody has solved before. And the only way to solve these is to try new things, try different ideas. Many of them don't work.

Vit Lyoshin (20:00.497)
Mm-hmm.

Spiros Xanthos (20:06.139)
But oftentimes you have like a breakthrough and that's what advances, let's say the technology forward. So I'm a huge believer in like, like I said, in risk taking and environment of psychological safety. And maybe the final thing I'll share is that I became a huge believer in in-person work as well. I think that it's not like remote work cannot, cannot, is not doable, but I believe when you're solving hard problems and when you need to move at a high pace, the amount of trust you can have when you're working next to somebody, the level of communication.

the velocity you manage to achieve is way, higher than, let's say, if you have to also add the extra burden of remote coordination. So these are some of the things I'm following or were following in building Resolve AI.

Vit Lyoshin (20:51.557)
Yeah, yeah, okay. Yeah, it's interesting idea about the remote versus in person. Right now everybody's shifting towards remote actually. So is it creating any issues for you to find people who willing to do that? Or is it enough talent?

Spiros Xanthos (21:09.701)
So it actually it is a smaller pool where in San Francisco everybody works in the office every day, five days a week. So yes, the pool of people we can recruit from is much smaller. But at the same time, think there is, you know, in this smaller pool, there is quite large number of people that are very passionate about being in person and they prefer to be in person. In fact, they wouldn't join a common if it was in person. So yes, it is maybe the pool is smaller, but at the same time, for many people.

Vit Lyoshin (21:28.816)
Mm-hmm.

Vit Lyoshin (21:33.094)
Yeah.

Spiros Xanthos (21:39.195)
Being in person is a strong desire and it's much, much easier, let's say, to recruit folks who want to be in person.

Vit Lyoshin (21:45.189)
Yeah, I see. Okay. And then also, how is it different running AI startup versus let's say all day startup in terms of like finding talent, some specific issues maybe.

Spiros Xanthos (22:00.348)
So first of all, everything is moving maybe at the five times the pace of my last startup. And that includes the time it takes to build a product, the time it takes to find customers, the time it takes to grow your revenue. And that's actually a good thing. I feel grateful to be part of this and have the ability to experience this AI wave. But yes.

It is very competitive, especially for AI talent. It is very competitive in terms of also the opportunity itself, because AI made a lot of problems that were impossible before possible all at once. So there are probably a bunch of smart people in every sector of, let's say, the economy now, thinking how to automate or how to move to an agent approach to lot of these problems. At the same time, the way I see it though is that

The opportunity is so much bigger because in reality, yes, you're competing with maybe other AI companies and you're competing with those companies for AI talent as well. But in reality, I think like the addressable market of AI and agents is going to be a lot bigger than the prior generation of tools we have built. Because not only were now able to essentially get better outcomes in our case, let's say improve MTTR, improve reliability, but at the same time we're creating huge productivity gains and we're giving people time back to go focus on other things.

So I think the addressable market is going to be much bigger and it almost doesn't matter in these hard problems what your competitors do. The question is how well do you execute and how well can you help your customers get to better outcomes?

Vit Lyoshin (23:43.405)
I see. Okay. And then is there any difference in the approach that you're applying now based on what you did with other two companies and now in terms of like figuring out the problem, for example, and coming with solution?

Spiros Xanthos (23:53.317)
and then

Spiros Xanthos (23:58.877)
I'll tell you like one thing I learned over the years is that it is very, hard to build a valuable product unless you work very closely with customers. So there's all the AI we started working with design partners from, from the very beginning, before we even had a working product. and we iterated with them weekly, practically, and validated, let's say our ideas every step of the way. And I'm a huge believer in that because.

that essentially forces you to find the shortest path towards like a working solution or a valuable solution. And as you iterate and exhaust, say all the known solutions maybe in getting you to the next step every time, then you at some point you face like really hard problems. You go to the real, let's say research problems and by solving those, then you get to real breakthroughs, right? But by the time you get to those, you know that you're really creating value or solving a very valuable problem for somebody because you're working very closely with customers.

Vit Lyoshin (24:45.712)
Mm-hmm.

Spiros Xanthos (24:55.068)
So that's one of the principles we follow very closely. And we work very closely with all our customers, not just our design partners that we had from the beginning, but all the customers that trusted us subsequently. And every engineer in the company, everybody, but in particular, every engineer in the company has access to, let's say our customers and they can go and ask questions and that creates trust on both sides, right? Like also customers probably see value in working directly with our engineering team and discussing problems and brainstorming ideas.

So that's definitely one of the things that we did, you know, or was a principle on top of, or let's say it was a principle in how we work with our customers. And we try to follow it very closely.

Vit Lyoshin (25:35.889)
Yeah. Do you have any tips for people on how to talk to potential customer without having anything to show them?

Spiros Xanthos (25:46.033)
Yes. So the way we did it, and I believe it's probably the right way, especially in, if you're trying to build an enterprise product is that before we even started, we had like this high level idea that we wanted to use AI to let's say, automate the remediation, troubleshooting and remediation of problems. And probably interviewed 50 people who let's say are in this space, just even to validate the idea. And what happened organically and very quickly, first of all, you saw that, you know,

There was strong validation that this is an important problem. And you would see that there was a good subset of the people that were interested in a solution. If there was one you could build, right? And then maybe a smaller subset of these people were even willing to work with you right away, right? Like, you know, they offered to say, okay, if you're, if you're building something in this space, I'd be curious to use it. So for us, it was very natural to like essentially start from, you know, this high level validation of the idea and gradually move towards like having early design partners.

The other thing that I think is very important is to have even a prototype, even a demo, like when you're starting, right? Something that people can see, something that makes the idea real. And that's the other thing we did from the very beginning where we would try to essentially have a prototype that is maybe not fully functional, but at least conveys the idea fully. And I found that, you know, I would talk to somebody, they would say, yes, I definitely have this problem, but

how is AI actually going to really address it? Can it really do the work of humans? And then actually by giving a demo, you were able to connect the problem to a potential solution. And that made it a let's say, a lot more likely that somebody would raise their hand to use the product early on or to become a design partner.

Vit Lyoshin (27:16.261)
Mm-hmm.

Vit Lyoshin (27:32.559)
Yeah, yes, I agree. Sometimes people need to just feel in touch and see how it looks to give you more feedback and to get by and things like that. And then also, I believe that every entrepreneur is kind of an innovator as well. You have to, right? So as an entrepreneur, how do you figure out and decide which next idea should be your startup, basically? To go after?

Spiros Xanthos (27:59.711)
that's, for me. Okay. When I, I left my, my last rollout Splunk, I didn't have a specific idea, but I was very intentional about wanting to start a company again. said, this is going to be my last company. And I had a few principles in mind that I wanted to, you know, the new, this new idea to have. One was that I wanted it to be a big problem and, or a big market. And the reason for that is it's not like I cared about the money per se, but I wanted.

to be something that if we solved it well, many other Spant people would be interested to join and work with us to solve it. And usually that's hard to happen unless it's like a big market that has a big reward on the other side. I also cared about a problem that would allow to scale quickly to build a solution, let's say that was scalable, right? And it didn't require, let's say, building a custom solution every time a new customer showed up, right? There is a lot of similarity across customers. And of course, the third one for me was

Vit Lyoshin (28:34.886)
Mm-hmm.

Vit Lyoshin (28:40.593)
Hmm.

Spiros Xanthos (28:57.36)
I wanted to work on a problem that I understand well. didn't like that didn't mean that the solution is something that I have done before because that's the whole point of essentially innovating. But you know, this let's say observability and reliability and software is something that I understand very well. And I worked on it for many, years. And, you know, at least I had like, I was confident that I understand, how a new solution might address the problem better than what, whatever existed out there.

And honestly, my advice to people who want to start the company is to join a startup if they can, like join a startup that is early, maybe a Ced or a Series A. And then you can actually see a lot of this methodology of how you discover a problem, how you iterate, how you fundraise, how you find early customers, which is usually not something that you see in a bigger company that has already been established, that all of these things were figured out years ago. And probably the people who figure them out are not even around maybe anymore. And because the types of problems we have there are

Vit Lyoshin (29:48.432)
Right.

Vit Lyoshin (29:53.681)
Mm-hmm.

Spiros Xanthos (29:55.572)
about scaling, managing labs teams. So anyway, that's kind of how I ended up in Resolve, Like a combination of some desires I had about the problem and something that I wanted to really understand well.

Vit Lyoshin (30:09.829)
Yeah, I see. Okay. And then also maybe going back to your lessons from entrepreneurship, what was some of the biggest lessons you learned and you want to kind of caution people when they start something new?

Spiros Xanthos (30:26.055)
I think that, you know, as a founder, your main goal should be to figure out something that creates value for customers. And I think you should spend, you know, a lot of your time in focusing on that and not focusing on fundraising. In fact, as a first time founder, it's probably a lot harder to raise money than it was, let's say in our case, you know, being a fourth time founder. But regardless, you know, the fundraising is a means to an end is never

the end. And a lot of I think founders treat that as kind of a big win. When in reality, if you think about it, raising money maybe sets you back in terms of like how much harder it becomes to be successful. Kind of that's one of the lessons. Second, I think there are many, many opportunities out there that can make you very successful without requiring like the VC path. I'm not saying there is anything wrong with that, right? And if you have a big idea and in a competitive market, you know, money from investors and especially high quality investors will help you a lot get an outcome.

But it doesn't mean that every idea has to go through, through busy funding. And the third one, which is maybe the most important is that honestly, what makes a company successful is people. And, you know, no matter how amazing you might be, let's say as a founder, you cannot do everything by yourself. So I spend most of my time at resolve actually on recruiting and trying to convince people who are smarter than me to join us and take a part of the problem and solve it better than I would. And,

This is often something that founders recognize, but they're not actually honest with themselves about it. And I mean, they don't spend the time, enough time, let's say on that. you know, the three does a side job. Recruiting is definitely a founder kind of task and you have to be able to do this very, very well to be successful.

Vit Lyoshin (32:11.983)
Yeah, yeah, I see. I could see that, that you need to find good people, smart people to help you out. Yeah.

Spiros Xanthos (32:18.931)
And you know, the smart people always have multiple options, right? So you have to be able to figure out, well, first of all, you have to be building something that is interesting and valuable, but also you have to be able to find the people whose, let's say, situation at the moment, like goals in life, et cetera, align with what your company does. I'll give you an example when it comes to like AI recruiting in our case, which is very hard. Many of the people we have worked at a big lab before, right?

Obviously they had their solving interesting problems that were getting paid well. think what's attractive about the result compared to that is that here. Each one of them has the ability to influence the outcome of the company and you their own efforts will make a huge difference to the outcome. most larger organizations or larger labs today, the work could be interesting, but the outcome might be the same regardless of what that person is there or not. Right. And

For some people that's very, very important. It's very important to me. Like when I was part of Splunk before, I was working hard and you know, my own efforts maybe had some impact, but it took like months of effort to get some outcome that I could recognize, right? Versus here where everything you do every day can see the impact it has, let's say to your customers tomorrow.

Vit Lyoshin (33:36.207)
Yeah, that's true. You can move faster right now, make decisions and things like that that can impact people or customers. How big is your team, by the way?

Spiros Xanthos (33:45.621)
So we're approaching 80 people now. vast majority is engineering. we have, like I said, in AI, everything's growing a lot faster, including the team, including the customers. And the majority of our team is engineering, but now we're building also a marketing and sales team as well.

Vit Lyoshin (33:47.985)
80 people.

Vit Lyoshin (34:02.843)
Got it. Okay. Well, that's great. And how long you've started? Like when did you start?

Spiros Xanthos (34:09.62)
So we started February of last year. Initially, Mayank and I decided to self-fund the company because of the reason I mentioned earlier, right? At the beginning, we were not sure if people would be willing to trust AI in production. And the second question we had was, is this problem solvable? So we decided to self-fund the company for the first year to get to high conviction that these two hypotheses will be true. But very quickly, as we started working with customers and iterating with them,

Vit Lyoshin (34:16.378)
Mm-hmm.

Vit Lyoshin (34:23.984)
Yeah.

Spiros Xanthos (34:39.634)
We realized that the market was more mature than we maybe assumed or imagined, or it was maturing a lot faster. So we ended up raising a big seed round. We raised $35 million about a year ago from Greylock, Unusual Ventures and a bunch of notable individual investors who have Jeff Dean, Fei-Fei Li, a few other folks like that. And we started growing basically more seriously about a year ago.

Vit Lyoshin (34:45.126)
Mm-hmm.

Spiros Xanthos (35:07.636)
The company launched for the first time in October of 2024.

Vit Lyoshin (35:10.929)
Okay, wow, so it's not even two years yet. That's great. Yeah, that's very fast. That's great. Yeah, okay. So looking ahead, you probably have big plans and I don't know if you have like a roadmap for the future or not, but a few years from now, where do you see your company?

Spiros Xanthos (35:15.124)
See you in a half.

Spiros Xanthos (35:31.166)
So I'll tell you, maybe I'll answer that from the problem, from this perspective of the problem or the area we're focused. know, software engineering is obviously part of it is coding, but the job has many other aspects, right? You have to produce something that is valuable. You have to design, have to run production systems. And I think we're still scratching the surface of what you can, how can you help humans do things a lot faster with AI.

Vit Lyoshin (35:45.637)
Mm-hmm.

Spiros Xanthos (35:56.681)
So our goal as a company is to build AI that understands software systems deeply. It actually kind of gets the knowledge that humans have around these systems and becomes more and more effective. And the more, say the quality of answers or the ability of reason or the ability of these agents to reason improves, the more of these tasks we can automate, right? So our goal is to essentially be the AI product that helps humans.

build and run production software systems a lot faster. And so far, essentially we're mostly assisting humans or working alongside humans and doing the work. And I see a future where we do more and more of the work and humans are supervising, let's say, resolve AI agents and getting to outcomes. the more, let's say, the coverage and the reasoning abilities of these agents improve, the more humans are going to trust them with more tasks.

Vit Lyoshin (36:34.374)
Mm-hmm.

Vit Lyoshin (36:49.837)
I see. Okay. Got it. All right. And then also, do you have any sort of... So there is always like, you know, this technology goes waves and waves with AI. Everybody's speaking about AGI now. It's been what, about three years since large language models were released. Do you have any sort of prediction?

or maybe advice to people who still want to jump on this wave, maybe where to innovate, maybe some great ideas, maybe something not related to your business but in general, where you see opportunities for AI to really improve human lives.

Spiros Xanthos (37:31.638)
First of all, I'll say that it's always a great time to start a company or, you know, a new idea. It doesn't matter like where we are in the wave because problems that were, let's say impossible three years ago, maybe became possible two years ago, but still many problems were still impossible two years ago until the further advancement of models and agents and all of that. Right. So there are definitely many, many interesting problems out there that one can solve. And of course, being part of, let's say a broader wave, it helps. Let's say if you take the cloud transition.

Obviously companies who build solutions, assume the cloud was going to be successful, you know, ended up benefiting a lot, right? Cause there was a wave of demand that came with it. And I believe the same to be true for AI. And I believe the wave is a lot bigger. You know, I think my belief is that every solution that existed before AI is going to be replaced with an AI approach. So really there is nothing out there that is going to remain the same in my opinion, either the existing products are going to evolve enough.

truly become AI first, or they're going to be replaced by something that is more effective. Because AI enables us not to build something that is just faster, but it's a lot better. And it's also usually a lot different, right? We can rethink problems in a way that was not possible before. And I think that is true for any aspect of our life, whether it's a consumer kind of product or an enterprise product like Resolve AI.

Vit Lyoshin (38:43.749)
Mm-hmm.

Vit Lyoshin (38:54.267)
Well, yeah, I actually haven't thought about this way because everything kind of become digital from analog, right? Like before from radio and TVs. And now I guess it's next step is to get it even smarter and faster and better. Yeah, that's great. All right.

Spiros Xanthos (39:12.097)
And I also think that we're still at the point of this new wave where mostly we're limited by our imagination. We're trying to essentially evolve things kind of linearly almost based on what we had. But I think soon we're going to start maybe being a little more creative in how an AI solution would look like. And it might look nothing like the prior generation of the tools we had at our disposal. And I do believe that that's going to gradually happen.

Vit Lyoshin (39:20.719)
Hmm.

Vit Lyoshin (39:41.201)
Yeah, yeah, okay, that's great. All right, so we're almost at the end here and finally I just wanted to ask you if you have any sort of advice or final words to entrepreneurs or people who are thinking about starting something, just maybe some motivation, a little motivation for them.

Spiros Xanthos (40:00.639)
Listen, starting a is a very hard thing, but it's also the surest way to change your life to the better. And one thing I always ask myself is at every step of my career is what's the hardest thing to do, right? And I usually want to face two problems or two questions or like two paths. I always try to follow the hardest and you know, starting the company is definitely has been the hardest in every step of the way. So I do think that it's a great thing to do. And honestly, even if things don't work out.

I think you end up on the other end, a lot more skilled. know, let's say your skills and your career will advance a lot faster, even if what you build doesn't succeed the way you want it. And I've seen this many, many times. In fact, when I'm trying to convince people to join Resolve, if somebody has been a founder or has been, let's say a founding engineer at a company, regardless of whether the company succeeded or failed, I'm actually usually very, very interested in getting these people to work with us.

Vit Lyoshin (40:58.693)
Yeah, yeah, that's great. Okay, well, thank you very much for your time, Spiros. It was great learning about what you do and your take on entrepreneurship and AI. That's always good. yeah, absolutely. All right, we'll stay in touch. Thank you. Bye.

Spiros Xanthos (41:09.033)
Thanks, Vitt. Thanks for hosting me. I really enjoyed the conversation too.

Spiros Xanthos (41:15.253)
Thank you.