AI agents aren’t plug-and-play and that’s where most companies get it wrong.

In this episode of the Nearshore Cafe Podcast, hosted by Brian Samson, we sit down with Oleg Baranov, Founder & CEO of FlameTree AI, to break down how AI agents actually work in real-world business environments.

Oleg shares why AI should be treated like a junior team member—not a finished product—and how continuous training, iteration, and feedback are essential to unlocking real value.

They also explore how companies can combine AI with nearshore talent to build scalable, high-performing teams across Latin America.

In this episode, we cover:

• Why AI agents require continuous improvement—not one-time setup
 • The biggest misconception companies have about AI adoption
 • How AI compares to training and managing human employees
 • The role of iteration in building effective AI systems
 • How AI + nearshore talent creates a competitive advantage

If you're exploring AI, automation, or nearshore hiring strategies, this episode provides practical insights you can apply today.

🎧 Host | Brian Samson – Founder of 💻 Plugg Technologies
🔗 https://www.linkedin.com/in/briansamson/

🎙️ Sponsored by Plugg Technologies – Connecting U.S. companies with top-tier software developers across Latin America.
🌐 https://www.plugg.tech

🌎 Follow The Nearshore Cafe Podcast
🎵 Spotify: https://open.spotify.com/show/6KYcgpmN77fJm6B25469B8
🎙️ Apple Podcasts: https://podcasts.apple.com/us/podcast/the-nearshore-cafe/id1775525954
💼 LinkedIn: https://www.linkedin.com/company/the-nearshore-cafe
🌐 Website: https://www.nearshorecafepodcast.com
🔗 All links: https://linktr.ee/nearshorecafepodcast

Transcript

Welcome And Sponsor Thanks

SPEAKER_00 0:12

Welcome to another episode of the Nearshore Cafe Podcast. I'm your host, Brian Sampson, and this is the show where we talk about the stories and people doing business in Latin America. Today's uh really interesting. It's a show about a uh long career seasoned tech entrepreneur. He's got an AI company, and uh we don't talk about this a lot, but how to sell to the Latam region. In this particular case, it's gonna be Mexico. So we'll get into that. Before we start, let me thank our sponsor, Plug Technologies, PLUGG.tech. Great way to connect talent from all over Latin America with growing U.S. companies. Let me welcome our guest, Oleg Baranoff, the founder and CEO of Flame Tree.ai. It's so great to have you. Thank you, thank you. It's pleasure to meet you here. So, Oleg, I'm just gonna um share your bio for a second with our audience because I think it's a really impressive bio. So, in addition to Flame Tree, serial tech entrepreneur, venture investor, you've worked in the banking and fintech sectors, uh 20 years on the software space, a lot of work with complex business applications with banks. And uh, you're a hands-on guy, I can tell. I believe you like to get your hands dirty into the tech. And uh, in addition to that, uh some really impressive um institutions you've attended, Stanford and Harvard. So uh really excited to have you. I would love to um just kind of set the scene. Tell us a little more about you know where you grew up and uh your career journey. Thank you.

Growing From Nine To 1,500

SPEAKER_02 1:55

I am in business already about 30 years, and uh my major part uh of this journey, I am acting as an entrepreneur. I managed to start several startups and uh businesses successfully. My first uh startup was uh internal startup when I worked in a software company. It was uh development of a new generation of core banking system. Our target audience was CIS countries. We started this software platform from to develop it from scratch with a very small team, like five five persons. I was young and a bit ambitious, and um it was successful, successful internal startup. We managed to create this software, it's a full-scale core banking system. We managed to sell 400 copies for 12 years. It's uh it's okay. And um, as I had a stock option in my first company, I I got paid in the end of this story and this journey. And uh using this money, really what I did uh in 2005, I started my first first my own company with my partners, of course. Uh, but I was a founder and uh uh uh CEO of this company. It was another idea. Uh we decided we believed in this moment after more than 10 years working for a software development company, we believed at professional services, consultancy, it's a very interesting area, very interesting topic. And we started a company focused on custom software development for banks, of course, because at the moment we had uh relations, we had expertise, and it was a very difficult moment in my life. It was uh again from scratch, second time. Uh but in this uh in this case, uh I didn't have like a big company which could help me. First year was very difficult, difficult uh period of time. We approached um banks, people we worked before. Unfortunately, many of them uh did not uh decide to take risk. And it's it's all it's it's always uh for such uh undertaking, it was very difficult to find first customers, but we managed to to do this. And uh finally this company became not very big but uh rather big. When I exited, it was 1,500 engineers. Wow. So incredible. And uh we we started it without any external funding from scratch with a team of just nine nine individuals.

SPEAKER_00 4:39

Please, yeah, please go go more more into details here because it's not every day we talk to people that can go. Finally we're sorry. Yeah, yeah, I was gonna say it's not every day we talk to people that have gone from you know nine to fifteen hundred. Uh, you know, myself, I've I've gotten stuck around the hundred mark a couple times in my career. I can't seem to get past it. So I would just like to learn from you, and I'm sure our audience will tell us more about just the building of that that company. That that sounds amazing.

SPEAKER_02 5:13

Thank you, thank you. So finally, I I got two different uh experience experiences. First is like out-of-the-box software development, and the second it's like building of uh IT professional services company focused on custom software development, but in both cases it's software development. And in both cases, uh we especially my second business, we was we had uh a lot of AI-related machine learning related projects, especially in the area of uh banking risks mana risk management. So we use machine learning models and uh trained machine learning models starting from 2005, and that's why when uh I started my next company, it was in 2022, less than four years ago, with headquarters in Cyprus. We initially decided to more or less repeat, but on the different market, uh the idea of our previous company. And uh we tried to build one more IT professional services firm. But frankly speaking, um uh it it grew, but not not very fast. Not fast enough. And uh one year later, in the summer of 2023, uh when uh large language models uh appeared and became like available, we came up to the idea to build Agentic AI platform, which now called Flame 3 AI.

SPEAKER_01 6:37

Yeah.

SPEAKER_02 6:39

We were not uh the very first guys who decided to do this, but uh anyway, it was more than two and a half years ago.

SPEAKER_01 6:45

Yeah.

SPEAKER_02 6:46

And uh we invested a lot in this platform during uh this period of time. And uh hope what we have now and what um we continue developing now, it's uh interesting product. At least uh we have customers, we have paying customers, we have uh customers among banks and financial industry, and I believe uh we have good perspectives.

Leadership Lessons On Team And Learning

SPEAKER_00 7:09

Yeah, yeah. Um I want to dive deeper into Flame Tree, but I don't want I don't want to overlook your career right before. What are some of the key lessons that you learned from before Flame Tree about building companies?

SPEAKER_02 7:24

It was um many lessons really. We did um a lot of mistakes.

SPEAKER_01 7:29

Yeah.

SPEAKER_02 7:31

But of course, what I think now, uh of course, the core of any such undertaking, it's uh uh the team.

SPEAKER_01 7:38

Yeah.

SPEAKER_02 7:38

It's a team, uh it's uh people, not only their knowledge, not only their expertise, but their attitude, your relations inside the team. Because you can't you can't do anything new using like a standard management approach, using KPIs or just giving directions. It should be common shared idea, and only in this case uh you can really achieve anything uh serious, especially in difficult times, because for sure, at least in my experience, in any projects and any company I started, of course we had difficult moments when everything looked like crazy, and uh only support of my team and support of each other could help in such situations. From my perspective, it's very important. And the second thing I would like to mention, you mentioned that I had a couple of educations. Really, I I try to maybe I'm so-called lifetime learner, or I graduated like four universities, and uh I uh try to every year or at least uh once in a couple of years to participate in any uh new educational program. For example, during the last four years, I took courses in in best universities. For example, in Barclays, I took a program regarding venture capital in IMD. I took a joint program with IMD and uh MIT regarding disruptive innovations, of course, uh bigger programs about finance, general management, corporate governance. So the second idea for me it's uh education and constant participation in uh this educational process, not only for myself, but of course for my team members.

SPEAKER_00 9:22

Yeah, yeah. How did you learn uh agentec AI?

SPEAKER_02 9:26

As I mentioned, in my previous company we was in were involved in uh machine ML and AI projects uh for risk management. Two areas, uh it's uh credit scoring and uh all related with lending. And second, we developed models for regulatory reporting according to standards like IFRS, Basel, and so on. So my team had data scientists already, and we had experience with the models of like previous generations. So that's why when uh LLM appeared, it was not very difficult for us to notice this in time and uh start using them uh widely in our projects, not on not only in Flame Tree, but in other projects, because we still have like professional services like branch of our company which continue uh provide like key professional services for for banks.

SPEAKER_00 10:21

Yeah. And and before we get into Flame Tree, just your your past, were there certain countries or places that you had most of your engineers working?

SPEAKER_02 10:31

Sorry, say it louder, please.

SPEAKER_00 10:33

Yeah, where where were your engineers? Which countries were your engineers in?

SPEAKER_02 10:40

Uh our main office in Cyprus, but we have a couple more. Uh we have uh Dubai, we have uh Johannesburg, South Africa, have uh Serbia and Armenia, two countries where we have like engineers uh with uh CIS background and groups.

SPEAKER_00 10:59

Okay, okay, good, good. And then with Flame Tree, that are your engineers in all those countries too, or where have you decided to put your technical team for Flame Tree?

SPEAKER_02 11:10

Core team for Flame Tree is on Cyprus. Myself and uh chief architect, my chief my product owner of Flame Tree. So core team in Cyprus, but we have some engineers in other in our other offices. We don't have um office in LATAM yet, but we're seriously thinking about this, and I believe this is like a neck next step for us.

Flame Tree Agentic AI Platform Explained

SPEAKER_00 11:33

Good, good, got it. All right, so let's talk more about Flame Tree. Tell us more about your vision for Flame Tree and the ideal customer.

SPEAKER_02 11:43

As I mentioned, our background in banking automation. So that's why when we started this project, this platform initial idea was to create a platform to automate customer service for banks, for retail banks mainly. And uh we thought about conversational AI agents. So to substitute call centers employees, to substitute uh legacy chatbots, which are unconvenient, with like you need to choose a lot of options before you understand that you really can't get the information you need and you you can't be sourced. So it was an initial idea, and because live language models can uh understand natural language, I'm up to the very simple, obvious idea uh right now that uh such agents, AI agents can help banks in two ways. They can make their customers uh more happy, and it's important. I believe all of us uh have like bad experience in some retail businesses, not only banks, it could be like telecom operator airline when you need to solve any urgent issue and you're waiting in the queue, or you're chatting with the like not very clever chatbot, or the the the employee on the other side can't answer your question because simply didn't know or can't understand the situation because you understand that in call centers of bigger companies, not only banks, maybe not best uh best people with not enough sometimes knowledge, sometimes not enough motivation to to help you. Um AI agents from that perspective have uh a lot of advantages, they can work 24 by 7. Became tired or frustrated or angry, and uh, another important thing, they can't uh leave the company with some knowledge and experience, and they learn on each iteration. Because the idea of such platforms like flame, for now I believe not all people understand this. It's impossible, from my perspective, to buy ready AI agent for a particular business, or it's impossible to develop or train it at once. Really, it's a process of constant improvement. It's like with real human. If you hire a junior specialist, you need to explain him what to do. And after that, he is learning, he he does mistakes, he does mistakes, and uh with each iteration, each and every iteration, he became better. The same story with AI agents. That's why one core idea of Flame Tree, it's not just we can give you AI agents which can solve your problems. We provide a technology, a platform which allows to organize continuous improvement of AI agents. So we have special tools, special uh modules to analyze all conversations, all outcomes, all results every day, and to get data for AI agent improvement. It's by the way, one of the maybe not so obvious uh uh things, even even now. If you go to the internet or any like trade show, you easily find like dozens of uh AI agents offering of companies who offer AI agents, platforms, and so on. But I just came yesterday from uh Web Summit from Qatar. You know it's like a big big conference.

SPEAKER_00 15:10

Of course, yeah.

SPEAKER_02 15:11

It was, if I remember correctly, 1,500 startups on this event. And I'm not sure, but at least several hundreds worked in the area of Agentic AI, Agentic AI platforms. What I can say, not all of them at least, or maybe even it's a rather rare situation where they speak about this vicious cycle of constant agent improvement. People try to offer like a ready solution. For me, it's maybe it's possible for very very simple business, for very standard business. If you have two restaurants, maybe you can develop an AI agent for one restaurant and offer it to a second one. It could be maybe okay. But if this is a bank, regarding our functionality, of course, currently we have, I can give you an example, one bank when our customer uses five different AI agents developed based on Flame Tree platform for customer service, for debt collection, for customer onboarding, for cross-sell upsell, and and so on. So, and of course, such AI agents can be specially trained, setup, integrated with different backend systems to be able to perform particular tasks in particular situations. Of course, uh I believe it's relevant to mention a couple of technical technical things. Our platform is uh LLM agnostic, so we don't trade our own light language model because it's like a very serious and expensive uh business. And I believe that company like companies like OpenAI and Meta and other giants can do this uh very good and we do not compete with them. We can connect any LLM from privately or publicly so publicly served to any of our AI agents. Moreover, if tomorrow, I don't know, one of these big techs uh release a new version of LLM, better than previous one, we can easily substitute it, change it just in like 10 minutes. So our customers can use best of breed LLMs, but our platform just helps to organize relevant workflows, relevant integrations with channels. Of course, we support different channels, starting from all messengers, telephone calls, emails, our agents typically integrated into web or mobile application of the bank. By the way, it's an interesting story. Now we have not only uh banks as a customers, we manage to get some interesting customers from different industries because technically our platform industry agnostic, not only LLM agnostic. As I mentioned, we cover different business cases. We call it inbound and outbound. Like typical example of customer inbound AI agents, it's customer service AI agent when customers ask questions and we just need to answer questions. And outbound cases, the best example is debt collection, when we need to outreach client to remind him, for example, that it's time to pay his debt, to ask to get uh agreement to pay, and so on.

Selling AI In Mexico And LATAM

SPEAKER_00 18:12

Yeah. If I could um if I could shift gears for a second, if I could shift gears for a second, Oleg, and um focus more on Latin America for a few minutes. So tell us more about doing business in Latin America. I think you said Mexico is where where you've um started first in selling, selling your product. Uh what's that been like? Um, you know, how did you decide on Mexico? How is selling to Mexico different from selling to other countries and other buyers?

SPEAKER_02 18:46

It's a funny story, uh, really. But first, I would like to start that um, as I mentioned, uh initially our focus uh was in uh Europe, Middle East, and uh Africa. Really, we got it was opportunity, it was not strategy. The person who worked uh with us in his previous company joined Mexican Mexican microfinance company, like uh it's like a bank, monoliner bank, and invited us. And uh so as I mentioned, as I said, it wasn't strategy, it was opportunity. We of course decided to to take this project. Hopefully, we we had uh a couple of people who speak Spanish, who hablo Espanol. By the way, myself, I I'm not very good in Espanol, but uh I used to live in Spain and uh more or less understand uh and can write in Spanish and Espanol. And we had, as I mentioned, some customers in Angola, in Portuguese speak countries. So we have some people who have Espanol as a native language. So finally we started this project and uh it was rather successful. And uh in a couple of months, really, maybe three or four months just after starting, we were recommended to another Mexican microfinance company by the people who worked with us in in the first one. So we got second Mexican bank, and again it was opportunity, it was not strategy. The third step was uh more or less uh similar. Chief marketing officer of one of these banks left his organization and uh joined another company. In this case, it was not a bank, it was a company which sells cars and motorcycles. And uh, in two or three weeks after joining this new company, he called me and said, Alec, we need uh your AI agents to to improve sales of our products. Can you do this? We are not a bank. And I said, I don't know because I don't know your business, but let's try you you may believe me or not, but in uh just three or four weeks, the first Air Agent for this company have been launched successfully, and it started to uh to provide 700 conversations daily with potential clients, and uh I believe it was one of the fastest implementations of Flame Tree. And so next step was again very very not similar but uh opportunistic. So we have uh we use uh partner models in our business. So how to do to do sell flame tree? Of course, we have some our own sales forces, but we have partners besides that. And these partners are typically IT professional services companies or software companies. And one of our one of our partners uh invites us to some tender for two banks, one in Guyana and the second one in Belize. So it's uh countries in the Caribbean region. Okay. I never I never been in any of them. And so after that fact, we decided that we need to to create strategy for for Latin for Latan, for Latin America. And what we're going to do, most likely, we will open an office in uh Mexico this this year, and um we'll uh we'll create some special strategy for this market.

Closing Thoughts And How To Connect

SPEAKER_00 22:14

Yeah, amazing. I love that. I love that. That's a very interesting story, and you know, I've been spending a lot more time in Mexico myself, and it's a unique place, and I think it's important as we talk about on this podcast a lot that Latin America is a unique place, but it's not all one way, you know, even though Spanish is the common language amongst most of the countries, it's just a very uh Mexico is its own special, special country. Well, I want to start wrapping up here, Oleg, uh, you know, as we as we get close to time. This has been really, really interesting to have such a globally successful entrepreneur as you. You're the first person we've had on the show that's in Cyprus, so that's exciting. And we like the uh agentic AI platform, flametree.ai. I think that's gonna be really exciting to watch. We'll be cheering for you and all of your success. So thank you again for coming on the show on the Nearshore Cafe Podcast. Again, we're sponsored by Plug Technologies, pl-g-g.tech. And Oleg, if anyone wants to reach you, what's the best way to do that?

SPEAKER_02 23:22

Uh my LinkedIn, I believe it's the easiest way, but of course I can share my like email. Uh you have it uh then we'll we'll I will drop it everything in the show notes for people to find you.

SPEAKER_00 23:36

So that's all we have today. Thank you again for listening to the Near Short Cafe podcast. Thanks again to Ola again, our sponsor, and we will see you again next time.

SPEAKER_02 23:45

Muchas gracias, Brian.