If you're not using AI, you can't compete with other people. What you were able to do in a year, now you can do in a couple months. So is AI gonna get to a point where it's just much better at producing code without errors? You still need human interaction. You still need knowledgeable developers to verify what was written. AI will get better, and human will better understand how to prompt AI, what to tell it to do, and how to describe logic to the AI better. Do you think it gets solved by AI getting better at writing code or humans getting better at understanding the logic that AI used? I think it's kind of Welcome to the twenty thirty podcast. I'm Keith Jensen, president at CADRE AI. We're here with Slava Krstic from Tubtida. Welcome. Thank you. Thanks for having me. Slava, you are the founder, CTO. You started the company thirteen years ago. Right? Tell us about it. Well, I did a lot of software development. I was a consultant, software engineer, architect for many years prior. Worked for really good, interesting companies before, and one day I was let go because company I worked for, they lost a big client. There was a lot of layoffs across the company, and they shut down our San Diego office. And pretty much I decided, what do I do? I don't want to be dependent on anybody else telling me, you know, what to do, how to do it when, and I started Tatita as a company. Very slowly with another partner and one developer. This is we I was doing still a lot of software development, supporting projects. And slowly, we started growing, hiring more people, expanding, and we kind of stepped away from working in the business and started working on the business. So that was And now you're focused on building software for health care businesses. Was it always the case, or did you pivot into that? Well, when you when you start up, especially in software, open consulting. You pick up any projects. It's just, you know, to pay your bills and keep people employed. The reason yes. We're concentrating about sixty five to seventy five percent of our business is health care vertical oriented. And reason for it, before even I started doing software engineering, I was working at the Institute. I was working at UCSD. After the case, I'm job at Pfizer. So all some other companies as well, all kind of near health care or health care related. And I do have a lot of knowledge on that. That's why I was we're still focusing on health care. And so, okay, being in the health care space, what I'm really interested in is because I come from health care background as well. I find that, in my experience, it has been more of a slow adopter type of industry. Obviously, you guys are in development. AI has taken the world by storm. My guess is that if all your developers are not using Claude code, they soon will be, or some version, of, you know, AI generated code. Would love to also dig into to what how you feel about that and the percentage. But I'm curious about the health care side, and do you feel like they have been slower to adopt some of the newer trends or not? Sorry to talk about all companies, but smaller companies, they're more adaptive. They want AI. They wanna move faster. Bigger companies, they, you know, they're protective themselves. They you know, there's a lot of bureaucracy still involved. And so they are duffing, but not as fast as and it's always been the case. When did it become evident to you that, uh-oh, we've gotta really get all of our developers using AI? And and do all of your developers use AI today? Yes. All of our developers using. If you're not using AI, I don't you can't compete with other people. Yeah. What you were able to do in a year, now you can do in a couple months. Whereas became evident to me, I would say two years ago, you know, especially GTP four came out. And you probably all know that right before I could just give you some snippets and, you know, help you with the code. Now you just do wipe recordings most of the time, and it's it's getting better and better. It's still not perfect. And in our company, we require to have at least two AIs, two models, and you challenge one model with another one. Interesting. Because cloud is great. What JetDeepKey especially came out, what, a couple days ago, five point five right now. You challenge those two, and you get great results. And is that a manual process? You're, like, you're having one write the code and then moving it over to the other one and say, judge this? You vary. It depends. Depends on the task complexity of what you're doing, actually. You still need human interaction. You still need knowledgeable developers to verify what was written because it still glucinates. And sometimes what you can write in, you know, one little function, you know, ten lines, AI will put, like, two hundred lines or just sometimes it's it likes to overcomplicate. So prompting is important. Experience is important. And what I found, and I'm passing and pushing to all our developers, verify with one AI what's other AI produced and challenge them. And then you get really good results. How has the adoption curve been of your developers? And how how many developers do you have with the business? Well, we have about, right now, around sixty five, sixty five engineers. Adoption, some people still kind of hesitant to use it, and I think it's AI is doesn't know what it's doing. What I see was tools getting better, models getting better, they kind of understand, well, what they they don't have to type anymore. You know? AIs type of them. They're verifying. They adjust. They correct. Sometimes they write the code, and it's asking AI to find any bugs or, you know, some security bridges or how you would improve performance. That's so they slow it up to it. But I see right now more and more of our developers saying, you know, we're using a lot of tokens. Help that. Yeah. Okay. So let's say they're I think my question comes from, like, if I'm a business owner and I've got developers working for me that that I trust are gonna put in the eight hours, right, that you you know, whatever it is, you're paying them, they're working for your clients. Now they can, you know, voice prompt if they want to or type a prompt in. They hit a button and it runs. No. Tell me. You have to babysit. You you can you can step away maybe for five minutes because it will still stop. You need to monitor. The good thing right now with a lot of changes in those models and tools, it actually spits out what it's doing. It's think aloud, I would say. You have to watch it. And sometimes you have to you see something is going off, and it's like you wanna stop right away and say, well, no. No. No. No. Don't do that. Do this. And this is where you're experiencing. It pretty much babysitting that AI. At least in our experience. You know, maybe other people do something else, but when we work in very complicated scenarios, you you still have to watch. You we don't do just run and you come back, you know, eight hours late now. Doesn't work this way, at least for us. Your client base in health care, do you have enterprise, or is it mostly mid market, small business? What are you typically working with? Right now, it's mid market, mostly. Yes. We had some enterprise big clients, but right now, it's mostly mid market. I would imagine that when you're building something, it's not always just in a silo. It is gonna be integrating with other systems. It's gonna be pulling from their APIs. It's going to be writing data to their core database and architecture. Are there ever concerns that you hear from the organizations that, hey. Well, we don't want AI having access to our, you know, structure? Always the concern. Data privacy, of course, your IP. Yes. People were working strong, but the good thing, it's more and more silos and where people can run their AI totally in a consult environment, which is now training of the general models. We can see that coming. If you could go back and and give yourself a piece of advice, you know, let's just say in twenty twenty two, whatever, twenty three, whatever it was when you said you said two years ago or so, so twenty twenty four. When you said, hey. We saw this coming. Right? Do you give yourself a piece of advice knowing what you know now about where it went? Anything you could think of? You didn't get a trade job. Just kidding. You just need to adapt. It it was like with any revolution, right, or technology spikes or COVID. Right? People panic first. They don't know what to do, then I adapt, and they're moving on. So I to just how fast you can adapt and what you can do. Still need to learn. You still need to know a lot. You need to know adapt to those tools and accept this as a it's it's it's just your tool. It's a better tool to use. I don't think oh, like, I heard a letter laid out a lot of people. And, yeah, yeah, I'm gonna pick up. But engineers, good engineers, I think they will stay, and they will be needed a lot because you still need good knowledge of architecture, you know, distributed environments. You need to know how to do prompts and human factor. Right? As we talked earlier, communication is very important. I know agents can talk to another agent, but you still monitor all this process. And it's you still need to watch and make sure it doesn't drift somewhere you don't want it to drift. I mean, models getting better and better every day. It's been thirteen years or so since you started the business. The trajectory we're in right now with AI last couple of years, how would you say that makes you feel compared to the other maybe stages of the of the journey? Like, is this an exciting phase where you're like, oh, wow. We can take on more clients, do it with less team members, or, you know, not have to scale as fast, but be able to do more work? Or is it more of a, uh-oh. I don't know what's around the corner, and it's, you know? Or maybe there's another feeling. It's kind of all that together because you don't know what's around the corner. We're still playing to scale. The thing is AI gave a lot of capabilities to smaller for smaller businesses to build bigger systems and applications. Right? I think it's getting less expensive for smaller businesses, build something bigger. They don't need huge teams anymore. You know, Two or three developers can do what before five, ten, or twenty people could do it, and it's faster, more productive. You still need to watch quality. I think quality and I think right now quality assurance people are getting more valuable because it's harder to validate. Another thing I see as a big problem, and it's probably this is I think it is issue with junior developers because they don't have that much experience. When you start doing white coding, even if you're senior developer engineer, you're losing the track of what's going on, and it's becoming extremely difficult. For senior developers, it's a little bit easier because they understand the, you know, architecture, how code written, what it's supposed to do. And if you generally don't know what it is, you just blindly trust AI, and it's it's not always a good idea. But with swipe coding, it's great. The problem is if you do have a bug, it's so hard to find that bug because you'd you know, when you write the code by hand, you pretty much even if you have thousand and ten thousands of lines, you mentally remember where you wrote that function, you know, what it was doing. And it's easy for you to go back and fix the problem. Right? With AI, because it generates so much code, you just you can't, and this is one of the big issues with quality. So okay. This is just you know, it's gonna continue to be an issue. Do you think it gets solved by AI getting better at writing code or humans getting better at understanding the logic that AI used to like, basically, the the whiteboarding. Right? Because when you're whiteboarding, you're trying to figure out the logic that it used to get to something. So if there's an error at the end, well, how do we backtrack to find that error? So is AI gonna get to a point where it's just much better at producing code without errors and we don't need that, or do we need humans to get better at understanding it? I think it's both. I think it's both. AI will get better, and it will have less bugs and errors in logic. And human will better understand how to prompt AI, what to tell it to do, and how to structure their or describe logic to the eye bear in their ways. It's kind of the eye will is growing, and people need to adapt and grow with it. So I think it's kind of both ways. This is a where you are today is a it's been a long journey from where you came as a a child. Right? Born in Ukraine. And do I did I have do I ever write that I think I read that when you came over here to the US your first time, it was, like, coincidentally, like, right when the Soviet Union was breaking apart? Yeah. It was pretty much two months. So I came from Soviet Union, and in two months, it's revolution happened and it broke apart. Yes. And did you stay after that? You stayed. Okay. So when you came here since you came here the first time, you have stay you are now you were then based here. Like, you you didn't go back and live No. In Ukraine. Okay. And you said you were at a couple companies before starting this. What companies were were those? I used to work at bigger ones like Pfizer. Just I've been, like, six years at Pfizer as a consultant. What's the big at at UCSD, Salk Institute, and did biology, actually, research for there. I used to work at NewDesik. It's a big consulting company in Orange County. And when the layoff happened and you decided, alright. I'm gonna start this thing, What was the pain that you saw in the market? Pain was actually I saw the pain well before I, you know, I started the company. And usually before I started the company for about fifteen years, I was doing software development engineering. And trend with offshoring outsourcing was, you know, picking up, you know, like, two thousands and, you know, started increasing, increasing. It was good money saving opportunity for companies. And it's not just money. It's it's a time saving too because pretty much company with outsourcing, offshoring can produce and generate code twenty four hours a day, pretty much. Right? So you're using all time zones. And that came at the cost, and, usually, cost was communication. And that was one of the biggest problem I saw personally, and we tried to address it many, many times, but wasn't that easy. So one of my goal when I started doing that is, well, I'm from Ukraine. We have great engineers there. I have good experience. I've been doing this for a long time. Why don't I try it? And, you know, the key is to have very good, knowledgeable people. You know, you just don't hire just whoever you find. No. It's like, in our company, old people are senior. We cannot afford to have junior people. It's too expensive to have junior people. Because one mistake they make, it can, you know, be a hundred thousands times more expensive than not to handle that person. So all of our engineers are seniors with a lot of knowledge. So we started doing that. And on the requirements, you have to speak pretty good English. So we don't need to have a translator, man in the middle. You know, our client counts directly, and they do communicate with developers. You know, you don't need to have well, we do have project managers who help just with organization and managing projects, processes, flows, blah blah blah, all that. But our developers in the meetings, they hear what clients say, and they can respond and talk to them directly. So it eliminates that gap in Like forward deployed engineers Yeah. Would yeah. Absolutely. That's great. That's that's very important. And doesn't matter what country you're in. We have people in sixteen countries. So but English is number one. Yeah. Well, technical skills and, of course, English. That's very important. So I fast forward twenty thirty. Right? Where do you think things are going? I still we will have a lot of engineers working with AI and producing a lot and a lot of software. And as I mentioned before, you know, I see some big products may go away. Because for even small company, it would be very easy to go and build your own CRM or, you know, inventory management system. Why would you need to depend on some huge company and pay huge fees and licenses fees? Right? If you can have your own, which will do what what you want exactly, you don't need those customization. And when you build generic applications like what we have, those huge monsters, it's really difficult to build generic application because you need to encounter for all possible combinations. It's very difficult. When you have your line of business, you know exactly what you want. You know, your workflow, your processes, you know what needs to be done, when it needs to be done. You just build with AI so fast, and you don't rely on anybody. So I see a lot of growth in companies who do hosting because there's gonna be a lot of hosting power supplies. Right? Because you need a lot of power for, you know, generating AI and, you know, energy, all of that stuff. Yeah. I couldn't agree more. It's it's funny. At at Cadre, we you know, with our clients, we always say, we're consulting. We're gonna go dig in, figure out the strategy of what you should build because we'll go figure out what all the manual workflows are across the, you know, every department of the business, and then say, alright. Here's where AI can help. And we've always had this this point of view that if there's something on the market that can do it, we're just gonna point you to it rather than decide to go build it. We don't want to be a dev shop. Right? But to your point, it is so easy now. And when we say, yeah, there's this product on the market, it's not built specifically for you, but it's got all these configuration settings and all this. And, like, we could set it up, and it's a hundred dollars per seat per month and and and before you know it, when you get down to the math, you're like, we could probably build this in a week and do it exactly as you need it and integrate with exactly the third party APIs and everything that you like, it's wild how far it's come in such a short period of time. Yeah. I absolutely agree with you. Yeah. It was APIs, connectivity was always good to products and connectors out there. It's it's easy. And so well, I'm not sure if it's gonna change for now, but it always was less expensive to buy than to build your own. I think it's changing right now. It is changing. And some of our clients, they're still hesitating to move from bigger companies because they have so much data for decades, and this is biggest concern. This is I think those companies, this is where they keep them. But the problem I think the problem for those big companies is that the data that the that the company the client has is stored with that company, and they wanna access it because they want that data to be talking to other pieces of data across their other tech stack. And when you build, you get to decide that architecture. You get to build all everything on the same frame with the same database, you know, whether it be Snowflake or, you know, whatever, Postgres. Like, you are building it in one place so that your data is easily accessible. So I think to your point, the there's that lock in factor of, man, we already are so far down the line with this big one. But as things keep shifting, I just wonder if, you know, the the ease of build and the ability to determine where the data sits and how it talks to other components is gonna be more valuable. Well, I think even with big companies and big products applications, you still own your data. You know, it's successful. It's just it's how you, you know, extract the data and in the way you want it because you may have some limitations of that. That's the thing. You own it. Yeah. But yeah. How are you how do you you know, when you pull it out of one big software and you pull into another, you gotta put it somewhere to compare to do the the matching and cross in order to you know? So there's there's still that that component of complexity to pull it out. So but I I hear your point, and I I agree with you that the large if I was one of those I was at one of those large tech companies, I would be nervous because it is quite easy to rebuild these days even with screenshots. You're absolutely right. I'm not sure if you heard or not, but Salesforce, HubSpot, you know, all those even Figma right now, they needs to be nervous because to give you an example. Right? I was doing this, well, like, having accounts with HubSpot, and I get a little bit upset about certain functionality they did or didn't give me an access to certain functionality I wanted to. For my subscription, they won't charge me, like, more for very simple things. Yeah. I took the screenshots. One week later, I had complete CRM done in my own application. I can do whatever I want. I don't need them anymore in one week. Yeah. It's pretty wild. Yeah. Awesome. Alright. Slava, thank you so much for joining us today. Appreciate your time. Yeah. Thank you for having me. It was a pleasure. Thank you.