In the Hot Seat with Akshay Kolte

Speaker A: Insurance Unplugged in the hot seat, where the complex world of insurance is laid bare. Hosted by Lisa Wardball, this podcast promises an unfiltered glimpse into the industry like never before. Each episode invites you to listen in on the candid conversations that usually happen behind closed boardroom doors. From deep dives with industry leaders and thought leaders to innovative discussions with minds shaping the future of insurance, we bring the most genuine talks directly to your ears. Our guests take the hot seat alongside me to explore the inner workings, challenges and triumphs of the insurance world. If you've ever wondered what goes on in the shadows of the insurance industry, from the boardroom banter to the behind the scenes strategies, this is your chance for a front row seat. Prepare for unguarded, enlightening and engaging discussions that cover every angle of insurance presented in a way that's both insightful and accessible. Welcome to the conversation. Welcome to Insurance Unplugged in the hot seat with Lisa Wardlaw. Welcome to another episode of Insurance Unplugged. I'm your host, Lisa Wardlaw and I've been so excited to have one of my dear colleagues, uh, I'll say partners in crime, actually, when I introduce you joining me for this session today, I'll let him introduce himself, but just for all my guests and listeners. We talked so much about Agenic AI and the stalls of Agenic AI. Really, the floor is collapsing and we're going to really go through today why Agenic AI keeps stalling. Akshay Coltay has joined us. He is the co founder of a new company called Run State. He also runs a wonderful IT consulting and uh, development engineering firm called Etlok. And he's somebody that I've just had the esteemed pleasure of building with. Gosh, I guess it's been for more than a year now actually. So welcome to the hot seat. If you don't mind introducing yourself to our guest. Yeah, welcome and thank you for your bravery and joining me in the hot seat about this, you know, very dicey topic today.

Speaker B: Thank you. Thank you Lisa and welcome everyone. Yeah, so as Lisa said, I am very much into the thick of things when it comes to sort of system architecture. I have background in sort of building, uh, gaming architecture as well as like a lot of infrastructure development. I've actually worked with, I think about 20 or so applications ranging from like consumer applications to enterprise applications. And it has given me a lot of insight into the juxtaposition that we find ourselves in now with respect to generic AI and really what is required from a underlying Infrastructure perspective. So definitely have that sort of foundational experience and would love to discuss that topic here today.

Speaker A: So excited. And you're too modest to tell people that you're one of the few technology savants that I actually love designing with and don't throw out of my office as soon as I start talking about, I don't know, maybe you're too afraid to carry that, um, badge into this conversation actually. But thank you so much for, for um, just your way of thinking, honestly. So you and I talk a ton actually, about every time there's an article, I'm like texting you or I'm calling you like smoke and mirrors. Smoke and mirrors. And you know, uh, the recent MIT report that was released I think just kind of proved that clearly, you know, AI came out, it's been around a while, generative AI. Now we're moving to a genic AI. Everyone in my opinion is really focused on what we would say is top layer orchestration, but underneath it, and you and I have experienced this directly, right? It's still brittle, synchronous, and in my mind, uh, incompatible with autonomy. You know, actually your depth as uh, systems architecture, all the work you've done with really decentralized infrastructure, as well as I think your gaming background in terms of speed and velocity and all these things that I think Agentic will look to emulate. How do we get beyond this collapsing floor for the AI ambition that I do believe CEOs and boards and investors are putting upon people, but I don't feel like people really want to admit it. What do you think is the initial thought of what's going on in the industry right now? How would you frame that up for our listeners?

Speaker B: So I think that in the industry right now, people are coming in and they're evaluating AI solutions, they're looking at the different agents and um, how they can help them drive solutions. But the existing infrastructure, the existing infrastructure is based typically on uh, sort of this request response, API style systems, which are great for when they were working and they were also great for doing integrations and things like that. It was a standardized way of communication. But request response essentially means that you're asking for something and you're getting something back in response. Um, whereas in the agentic flow, the flow is not request response, it's basically something that needs to happen in the background. So it's really more event driven. So you want to think about it as, hey, an event occurs, go do a bunch of things and then those trigger the next set of events that can do the next bunch of things. So it's not really this request response, uh, format. And lot of the existing systems reside on these monolithic, uh, infrastructure setups where all these things are API driven. Um, lots of people are not even into, uh, microservices yet, so it's not even distributed. So this, um, I think is a big challenge, um, in terms of implementing AI, because you need to plug in the AI somewhere and you don't have a system to be able to do it in an incremental way. So the step to take in order to implement AI is looking like a gigantic step where you have to overhaul your entire infrastructure. And that's really where the industry is hitting a wall.

Speaker A: Yeah. And, you know, it's like watching bad choreography, it just, like, it kills me. Right? Like, I'm like, oh my gosh, what are we doing? And I think that, like, what causes that? Like, for example, why can't CIOs and Chief AI C suite members, chief data officers? These are not people that do not fail to understand that they don't have an event native architecture. Right. Like, are they aware? They're aware, right?

Speaker B: Uh, they are aware. The thing is that there isn't, um, a solution in the market. And this is something that we're working on at Run State, and through this series of podcasts, we'll kind of bring these elements out in terms of solution. But the key is that Even though the CIOs understand, hey, we need to go even native, there aren't enough, um, systems out there that they can use right off the bat. If they have to do something event native, they have to architect it from scratch. And that's something that comes with a huge bill, which is the reason why it's like crossing a grand chasm here to try to get there. Uh, that's really the challenge that I see in the industry.

Speaker A: Um, do you mind just for a moment clicking in on that? Because I remember and I think this is why you and I got along immediately. When I first met you, I was like, ah, finally. But I remember thinking back, like, you know, 2015, 2016. I remember using concepts of dynamic. Like, dynamic was like my favorite word to use. And streaming. Right. Like, those are my, like, two favorite words. Like want everything to be real time. I want it to stream, meaning not at rest, and I want it to be dynamic. Now, clearly, actually, you know that I ended up going very heavily into, um, embedded accounting and kind of event native accounting, which maybe was just kind of a muscle flex I was trying to do but what surprised me, and I'm getting to the point is how many very, very, very robust big SaaS plumbing deeply enterprise grade solutions are out there that are in fact not event native.

Speaker B: That is true. And there are, it's kind of like the API layer sort of right now is just sort of hitting this uh, mainstream area where all the SaaS companies want to do like all these different APIs that can integrate into different things. And really it's driven by the industry because it's like the API hasn't become that standardized way of, of communicating. So a lot of times people are not thinking about this from a real time event native perspective. And so therein actually lies a challenge. And a lot of the technologies are out there. For example, you have websockets, you have Kafka, these are elements which some of the people have started to introduce. But still there's been a big gap in terms of understanding how to actually leverage that in lieu of um, a request response, type of, type of system. And predominantly it goes back down also to, I think predictability is a way that I uh, would like to describe it in a request response format. It's predictable. You send a request, if you didn't get a response, something went wrong, you just send the request again and you'll get it right. Whereas if you're thinking about sockets, you have to maintain the connection. There's, there's sort of practical elements and implications to it in terms of how um, this process should or should not work. Right. And um, as technology has evolved, especially like you see this um, across the board with respect to, like you said, streaming, a lot of times uh, you see, and you see this in gaming a lot where data streaming and sort of real time interactions is very commonplace and it is very stable and there are processes to ensure data integrity um, within those systems. So it's essentially taking that element and applying it to enterprise architecture is sort of where I look at the next step in terms of preparing this for the agentic generation.

Speaker A: Yeah, let's go into that, let's go into that, uh, I like to call it architecture theater. Which is as a non trained technologist I get really annoyed when people try to like hyperbole architecture on me or tell me they've created a system without a data model or you know, all these things. But I really think that in particular when we went into this BPMS orchestration layer and everything was like process flow centric, you know, everything was kind of like flowcharts and triggers and I'll say it was Kind of parading and masquerading as autonomous. Like, like, right, like we've got these things set up, low code, no code, you know, all the things. But where are the really constraints? Because I don't believe that your genic orchestration platform can operate in these worlds. And I think we're about to see this catalytic moment at scale, which is why I believe we're seeing such low. It's not human adoption of the agency AI, it's literally like there's an oil and water situation going here where it's incompatible. And then I want to ask you a few more questions, but let's start on that kind of determinism and what are the constraints that we're seeing in the architectural layers?

Speaker B: Yeah, um, so existing, like, you know, take um, existing sort of BPMS systems or orchestration systems. Most of these are based on sort of to look at again like in this sort of request response format there is the low code, no code scenarios where repeatable code is uh, packaged into modules which can be reused and all that. And that is sort of the structure of a lot of these components. But what is missing in those components I think is a um, distributed purpose built, um, and auto scaling um system. So the way that I look at it is when you think about sort of your flowcharts and your triggers, you're kind of saying, okay, hey, if I need this, I'm going to make this request, it's going to go to X module, it's going to execute, um, send stuff back. But in reality when you're thinking about this from a agentic standpoint, what you want is you want to say, hey, I want to do a calculation and therefore I'm going to send this to a distributor worker that is purpose built for calculation and it's going to do this and once it's done processing this, it's going to actually fire the next event. So it's really just a loop. So the way I think about it is, okay, you have a listener that can get like, listen to your event, figure out what are the different things that it needs to do. And here a lot of concurrency and parallelism is a big factor here because um, it's not that one event has just one process. It's essentially okay, the event fires and multiple things could be listening to it and say, hey, okay, this event fires, I need to do the next process. They complete the process. At the end of it they're going to send the next event saying hey, I'm done with this, what next? And there's a loop there. And one of the components in this loop, which is going to be vital for sort of the agent, um, um, execution is going to be the proof part of it. So we need to have the ability to say, okay, I did this, here's the proof of what I've done, right? So some sort of notarized way of keeping track and auditability of saying these are the things that I just did. And then you fire the next loop and so on and so forth. So if it was built out in this sort of series of loops, um, that system then um, can basically function. So you could have millions of agents running underneath, they're all distributed and you're just using this sort of looping system

Speaker A: to execute those which in essence having grown up on the more business side of it, kind of, I'll say partnering with my technology team to deliver what we needed to for the business. I was leading it as an executive sponsor, but with the tech team it's interesting because at the core when I would see things, I would feel like we were kind of like borderline event driven. I would feel like we had things that were listening, right? And in my mind if I were hearing this, I'd be like, yeah, like m, check, I've got that check, I've got that check. I've got this. And by the way, uh, everyone stays at this like three bullet point level, right? Like nobody wants to get deep, everyone wants to keep meetings at this level. Gosh, almost nobody even wants to read anymore. But when I come down a level actually and then I ask someone like you like, hey, go do this. And you're like, m, scratch your head, whoops. You know, it's really that substrate level that I almost think about it like the darkest part of the ocean that nobody really wants to go to, that light doesn't really persist. But that's where nothing really is being built to kind of funnel this. Which I would love your thoughts on Nanda and the MCP and ADA protocol layers as well. Why aren't people fixing the substrate? And then of course I want to know why do you think that's an area of opportunity? Clearly I want to know your thoughts about that.

Speaker B: Right, um, at the very substrate level, if you look at the current systems that are there at the very substrate level, you're still seeing things like queues and batch processes and uh, you're still seeing non event driven real time elements at this bottom layer. So um, even if you take, so take your example of A2A or MCP, you're still using existing sort of request response, type of format to communicate between different things. Like you might be saying, hey, I'm going to set up this thing agent to agent. But what you're really trying to think about is so agent to agent. I think there's a problem in terms of, well, how do you maintain traceability when you're basically just sending things out from agent to agent? I think that uh, if you start thinking about it as a loop then you uh, start to see that there's a way to actually keep that um, traceability and essentially have an intelligent router determine how you process. So rather than going agent to agent, you're kind of going to agent to sort of intelligent router that goes to the next agent. And that way you could actually keep track and keep that um, lineage of what are the different processes that happened. And if you sort of commit that out to a chain, you could actually have a auditability in terms of saying, hey, this is the things that. Even though, let's say There were like 10 different agents that went in and did things in this process, you have the ability to trace that and say what exactly happened. And these elements, if you kind of go to the very, very bottom layer of a lot of these systems, these are not in place. You're essentially just using things that you have which is, you know, like, okay, well currently we have a queue system, currently we have batch processing, currently we have schedulers, uh, timer nodes and whatever it is that you have underneath these layers and you're leveraging those. But in reality what you want is something that is real time, which is, you can think about it as sort of a pub sub type of model where it's like, okay, hey, as soon as the event is fired, I have something that's listening to it and it can, can pick it up and execute it right away. And there are technologies available to be able to build that. Uh, um, it's just that the industry hasn't stumbled upon it yet. And that's I think a really big area of opportunity because this is really where um, we're at run state. Essentially we're looking to come in and look at how can we develop that substrate so that we can actually provide that underlying base structure so that any new application can actually just. It's a starting point. Like, you know, people can come in and start there and they're already, even if they uh, were trying to do just prototypes and MVPs and things like that, if you had a place to start which is already enterprise grade Already, like, you know, you can, you can do your prototypes, but when you scale, you don't have to rip out and rebuild your architecture. I think that that is kind of where, um, it's like a big opportunity from my perspective.

Speaker A: Yeah. And I think of all the kind of back to some of the huge SaaS giants that are technology companies, big ones, right. Google, ServiceNow, Salesforce, et cetera. And interestingly to me, I think, or do you think they have people in the back room working on this substrate layer while they're working on MCP protocols? And a to a. Because come on, again, I'll go back to. Actually, they're not, not smart enough to understand what we're talking about. So it's either two things. One is the economic yield and if you will, the commoditization is only rewarding, like, kind of like sugar feeding sugar. Keep getting the sugar high, keep getting the sugar high and they know there's gonna be a crash. But it's like, oh, well. Or they're literally like too big to handle. We gotta feed the sugar high. And we've got a room of people, you know, like, like, I can't remember who said this to me. I think it was Microsoft. Um, says to me we got a room of people, like not literally locked in like a room, but we got like these brainiacs that we keep like hold away from everyone and they're working on all these really deep things. This was have been like, you know, like eight years ago actually. It was actually in Toronto, which I know is where you're from. But like, what do you think is really going on with the commercialization and commoditization? Or do you just think nobody wants to focus on this because demos and touchability rewards visibility and infra is just not something that VCs really reward.

Speaker B: There's sort of two lines of thought, right? So there's some companies that are thinking in a systemic point of view, right? So they're thinking, okay, if I was starting from scratch, how would I do this? Right? How would I build this? And a lot of the technologies, like Nanda is one of the good examples. It's like you're basically saying, okay, well if I was starting from scratch and I was building a whole protocol around this, how would I develop that? And so that's something which is um, being thought of as a, uh, a element which might come into play a few years from now. But it's essentially looking at the sort of systemic overhaul type of thing. On the other front, you have companies which are coming in and saying, okay, well, I'm going to go and build applications on top of, um, the things that already exist. And these are the companies which go in and say, okay, well, I'm going to do my mvp, I'm gonna do UI only. I'm gonna not really think about the substrate. I'm just gonna go build on top of it. And they're really driving sort of the applications over here. Right. So that middle ground though is where, uh, there's a gap. And this middle ground is hard to get into for a lot of companies because a, to get into that middle ground, you're essentially putting something out there which is gonna be dependent on, you know, the application, like the sort of the next generation of startups to actually pick up on and be able to like, you know, make it successful. So it's really like, you know, that's not really where the big VC focus is. And at the same time it's not sort of this new, shiny new attractive protocol level or, uh, you know, new concept level thing, which again can drive interest from VCs. So the middle ground here is where this, the substrate lies. And I think that like we saw with APIs as well, I feel like this area is where a lot of people will fumble. And if technology, like, if people don't come in and put technologies in or standards in a lot of things, that a lot of times what's going to happen is that people will just fumble around, come up with their own sort of cobbled solutions. The industry will have a lot of. You can think about that like a monopolistic competition where there's all these different products competing for each other with slightly different variations. But you see that there is a vast number of different companies that might go in there, each with a suboptimal solution to this. So I think that right now is like a very, very good time to jump into the space, build out a solution that has the potential to become a standard within this substrate layer.

Speaker A: Yeah. And I want to talk to you, like, I want to go a little bit more now where, like, where do we need to go and what's kind of, kind of, let's call it like capture the flag, like take the hill. Right. It's so interesting, right? Because in my mind, actually, I've always been kind of, honestly, I've never really been captivated by the digital app isms of the last five to 10 years. It never really interested me. I mean like, yeah, pretty is pretty and we need to go digital. And I thought that was Just like table stakes. But it was never where I was fixated. I always, always fixated on what doesn't hold state. Like when, when, uh, maybe it's because my audit background, but like when things weren't verifiable, when systems weren't trustless, I felt it, right? Like I lived that experience in CFO roles, in auditing roles, in M and A roles, when you're acquiring businesses and things aren't yielding. I also felt it when humans are working with technology. To me it was never human led. It was like the humans need to trust that the system is doing what the system needs to do. It was always a state of trust. I really think in my opinion, what we ultimately need is composable, verifiable event native execution. How do you think we get there and where, if you know, if there is an investor listening to this? Because I do think VCs will not invest in this layer. It's not in their DNA. I think private equity is out too. I think it's going to be maybe some deep infra people. Even though so far we've been disappointed with, I'll say, like where Sequoia and Horowitz are investing in AI. And there's been lots of papers written about this. This isn't my opinion, but fascinatingly, Akshay, I would love to know, do you feel this is a moment of inevitability? And of course, I'd love to record it on this podcast. I can look back at it like five years from now.

Speaker B: Um, absolutely. It is absolutely inevitable and it is a direction in which it's going to have to be established in order for agentic to succeed. I feel like the way that we're looking at things, and I think a great point that you made, is essentially the trustless architecture. So you want to be able to have agents be able to do things. And for you to be able to see saying, hey, okay, this is the things that happened, but have something running behind the scenes that actually maintains that auditability as well. So when you look at something and say, hey, okay, I'm running my business. And let's just take a real world example. Let's say that one of the clients that, um, I work with, they rent out Internet kits and they have a system underlying where if people don't return things, if things go late, or sometimes people need shipping labels and things like that, and if all this stuff, if you can imagine a world where all this stuff was handled by an AI agent, there's no human involvement and the CEO or the CIO can come in and look at data and reports and basically just say, okay, well the agent handled this and be able to just trust in it and say, well I know this is actually accurate because I have the proof that these are the things that took place in this order. Then that's like a huge win. And you can think about what the person can then start thinking about in terms of expanding their business and the new ideas and everything rather than being stuck into the day to day of these type of operations. Right? So I think that that's really where the world is going to and I think that in order to accomplish it, essentially having that loop of saying, okay, well something needs to be able to a listen to an event that occurs. It could be anywhere, right? Like it's not something which is like monolithic and built into one sort of worker or server. It basically like you have these distributed uh, agents which are listening to events. As soon as the event occurs they, they go do their performance, they do their actions. There is a provability in terms of all the things that they're doing. So all that stuff gets notarized, gets maybe added into, into a blockchain and when they're done they fire the next event. And it's not a response, it's basically the next event that can fire the next thing or you know, generate a report, could be one of the events, but uh, one of the actions that it does. But basically it's still all event driven and it's a loop. And I think that that substrate is basically what we need in order to have this be functioning and be able to handle that across the board.

Speaker A: Everything in my body feels that it's like the culmination too. So many interesting build ups to this moment. Like I think when we look back in time actually we'll think, you know, all that stuff about blockchain that you know, you know you've got a uh, divide. People either make fun of it or people like are hardcore, you know, crypto and coins. But if you think about the magic of a trustless architecture, I think about immutability and I think about cryptographically verified transacting, right? Like how do you get to trustless? And I think that's a thing. I think how do we go event native? I think all of that combing up to kind of almost the eliminating of uh, process orchestration. The way we used to think about it and think about it digitally. To your point, it's very similar to gaming. It's event response, rapid fire and uh, everything in my mind says like, all the signs are like pointing directly at it. Like I'm like, wait, how do people not see it? But I know that you and I share this. When you're working in these real life examples, you tend to hit walls faster than when you're in an academic or maybe even a strategic consulting role or position at an organization. Can you talk to me just about. I'd love it if you'd share with our listeners just some more examples and use cases you have of, uh, how do you know APIs won't work? Like, how have you felt that? Because I think it's different to say I have a theory versus like now I've already hit that wall. That's just like a different perspective.

Speaker B: Yeah. I think that there are several applications where we've seen this, where even if it was a UI driven application, I mean a lot of times already you're starting to see applications that use websockets for, for example. Right. It's essentially you're not having to wait for a request for something. Uh, you're in a space, you're doing something and you start to see, let's say you're talking about a screen. You can also start to see the screen update without you actually having to touch things. And I think that it's kind of like the starting area of, and this was something like a five years ago, seven years ago thing in terms of where I was, where I started to actually already start to see this in terms of saying, hey, wa want to do this as a request response. I want to actually be able to see this, uh, update in real time. And from there we started to see a lot of different applications where this started to fail in the backend as well. So we have, um, you talked about event native accounting. That is a great example. Usually accounting was always done in terms of, okay, there's a nightly job, so you have all these transactions goes into a batch process that runs in, in the night. But suddenly, uh, you start to see scenarios where that is actually one, it's a single point of failure. Like if something fails there, you have to have all these recovery processes. Second is you have all of these things that are happening and the amount of load that you're doing in sort of your nightly batch is going to be significant. So now suddenly you have said, oh, okay, well I have this process that has to run every night with all these certain volumes. And until that point I can't see, see information in my reports, in my data. So the requirements have started to drive, saying, hey, wait, I need to be able to see if something happened. I want to be able to see that information as quickly as possible, which is hilarious.

Speaker A: I was thinking about some of the times you and I have gone to again, very industry enterprise grade people. And we're like, so how do we see it update in real time? And they're like, what?

Speaker B: Yeah, uh, it's like there's countless examples of this happening in the industries and legacy systems right now. And the interesting part of it is that changing that sounds like this giant mountain that you have to climb. Because a lot of these legacy systems, they have been built over a lot of years and the reason that they're still there is because they work. And so for CTOs and CIOs to be able to say, okay, hey, I have this legacy system, I need to be able to change it, then that feels like this gigantic mountain of task to scale. And I think that that is really where, um, first thing is, uh, having AI. Second thing is having a substrate layer that we've been talking about is going to basically make this so much easier for these CIOs and CTOs to be able to transition. Because now they don't have to actually completely dismantle their legacy systems. Their legacy system just becomes one of the distributed nodes that's doing something. You're just pulling out one thing at a time and saying, hey, I can get this process and I can have it built, have it run on this particular purpose built system or agent, I can have an agent go and do this calculation or I can just have this agent go and do this XYZ process. And you're not actually fully dismantling your legacy process. You're doing this in an incremental fast.

Speaker A: Right? Which I think is so refreshing. Right? Because I do think we were very binary before. And part of the reason why we got ourselves into this trouble though is that we started wrapping legacy tech APIs. But I do think, and actually I'm so grateful and I hope my listeners love. I am just captivated by how like simply and like educationally you're able to break down very complex topics and explain them. Um, so I want to do like a whole series on this, which thank you for like agreeing to do part one of this series with me if you are listening and actually maybe like as your call to action as we start to wrap up because again, we're gonna keep going deeper on all this. And again, I'm so grateful to have access to you actually. What would the first of all, like, I'll get, I'll give it like if they're trying to do a smoke test and understand, like, do I have a problem? Like, like, you know, and you're a novice and you haven't been in the details on this or you're a deeply technical person and you just want to sift out like what are my vendors telling me? What is this? A, to a MCP agency, uh, project for my consultants that I just hired to tell me what are maybe two or three, I'll call it core, indicating questions they could ask. And then I want to ask you, like what your call to action would be for people going forward to say what should they just immediately start doing, what should they stop doing and what should they continue to do?

Speaker B: I think that the immediate thing, I think that, let me start with the question that they want to ask. Right, so the first thing, question that they want to ask is, is the processing that they have currently, is that actually distributed? Is it decentralized or is it running like in these monolithic, um, environments? Right, if it is decentralized then great, you've taken sort of the first step. You have your systems actually um, broken down. Now the portion that really I think that people should stop doing and especially with slightly larger companies, like with startups and other other things, like I understand it's harder but with larger companies or mid sized companies, I think that they should stop looking at processes in this sort of UI first, um, way. So if you're doing something new, if you're deciding, hey, you know what, I want to create this new system rather than thinking about it and saying, oh, I'm going to build out the ui, I'm going to do a demo. Rather than that, I think they need to start saying, okay, well what can I do? Like I need to basically take a step and understand my architecture before I actually layer something on top of it. Because you can do whatever ui, uh, you want. The architecture portion is sort of a decision that you have to take at this moment. And if you don't take that right decision, you can't change that very easily. It's going to stick there and stay there for years to come. So it's really, I think that one of the things that people should stop doing is just sort of building out that UI first way of saying, hey, here's what it's going to look like and here's what the experience looks like. Rather than that they should start focusing on saying, well, how am I making sure that this architecture is actually viable? And a great question to ask is if I wanted to do this, if I just wanted an AI agent to do this, will it work in my CART system? Like, if I were to just say tomorrow, say, I don't want to do this in an API, I just want to replace this and I want to put an, uh, agent in there, is it actually going to function? And I think that's a good question that can get people thinking about how they should lay out their architecture other than the portions. I think people should continue to really push the boundaries of what is possible because just like right now with Agendic, there's infinite possibilities of what you can do. And so I think I would really encourage people to keep thinking about new ideas, keep thinking about new things that they can do, and, uh, always think architecture first. Um, and that way they can start transitioning from their existing systems out to more, um, agentic, friendly architecture.

Speaker A: And, you know, I'm doubling down on that both in my LISA Zone series, my Innovating at the Edge series, and here we are, guests. We're going to be doing lots of breaking this down on insurance unplugged over the next couple of months. So, you know, I really, really, really don't know how I got drawn into infra. Actually, I genuinely don't. But I fundamentally believe with like, every fiber in my being that if we don't get the AI infra layers right, then all of the rest of it is genuinely theater. It cannot support the load because just think about the roi. Let's kind of flip to the CFO CEO level here and the innovation and all the things, right? You cannot yield ROI on things that do not scale. It's deterministic in that sense. Like, you have to scale in order to get ROI out of things. So if we can't drive scale, which in my mind demands infrastructure. So thank you so much, Akshay, for being a guest today on this initial part. Very excited to see, you know, what you are up to at Run State, how this, you know, continues to evolve for you. If people want to find you and follow you, where's the best place for them to connect with you on LinkedIn website? What's your best connect?

Speaker B: LinkedIn is always a great place, uh, that people, people can find me, uh, and connect. And, uh, you can also look up, uh, the Run State website. We're going to be getting that up and running in the next few weeks as well.

Speaker A: Awesome. So to all of our listeners, we will continue this journey. Until then, stay curious, stay informed, and stay plugged in. Thank you so much.