Inside FlowX.AI: AI Engineered for the Enterprise.

Mike

Welcome, I’m your host Mike Parsons, and I’m here at the FlowX.AI headquarters and I’m with the FlowX.AI technical gurus. So this session is going to be great. Today we’re going to get into this meeting place between mission-critical enterprise applications and the technologies as it’s applied to banking, insurance, and logistics. And what we really want to do is have a conversation with the experts. We want to get into a dialogue to understand what’s really happening now and also what could happen next, because this is a fast-moving universe, where some days you open your feed and it’s doom and gloom at all at the same time. And we’re all left wondering, what’s the real story? Well, it’s so good that today the experts that I have are two of the technical wizards behind everything that’s happening at FlowX.AI. They’ve also been my friends for over 12 years. So I’m excited to welcome both Serban and Bogdan. Guys, welcome.

Mike

Yeah, 12 years, huh?

Mike

We had some adventures in San Francisco.

Mike

We had some adventures in Moscow.

Mike

And here we are. 12 years later, and it’s one of the most exciting times in technology. For you, Serban, when you look back at your whole career building applications, thinking about technologies, when you think about this moment right now. As a CTO and co-founder of FlowX.AI, how does this moment feel for you in technology?

Mike

Well, it’s crazy, isn’t it? Because if you think about that moment we saw cloud was the big thing and then maybe it’s AR and VR. And now we’ve got more acronyms. It’s obviously AI. I feel like it’s a bit of a golden age for us in terms of possibilities. What do you think?

Mike

Now, Bogdan, you’ve had a hand in making some of the large language models that FlowX.AI uses. That must feel pretty special to be putting together tools and technology like that. Do you feel like it’s a bit of a golden age, or is this just another technology for you?

Mike

It feels like the challenge in building has really changed, right? I remember in some of the projects we did in the early years together, we were kind of asking. I’m not sure, can this be done? It’s going to take a while. There was so much mystery. Now I think the question is how fast, right?

Mike

So it’s well we used to have our colleagues staying overnight and working all night just to get a bare bones application together, which is something now I think we could probably all put in a prompt. Yeah, I mean, heaven forbid you might even get me making some code, right?

Mike

Well, that kind of sets us up for the topics that we want to talk about today. I’ve got some really fun questions and formats for both of you. I think what we want to do for all of our viewers is really debate and wrestle with some of the really big challenges and hot topics that are in the minds of the enterprise. Because we are really at this moment where everybody, up until now, they’ve kind of had their chat AI off to the side and asking a few questions to get an answer. And now we’re really talking about putting that agentic in the workflow and actually getting the AI to take action to work and execute and produce on our behalf to be a very active part of that workflow. So it’s going straight for mission critical heartbeat of the enterprise. So are you guys ready for some some fun but also some challenging questions? Questions? Okay. Let’s go with you, Serban, firstly. I want to do a scenario for you. What’s more likely? Okay, here we go. What’s more likely in three years from now, most routine banking work is still being done manually? Or agents are doing most of it?

Mike

Okay, so here’s a follow-up question. What do you see today when you deploy your technology into a bank? And it’s interesting that banks are actually showing quite a lot of willingness to engage here, which is a very positive sign. What do you see today that gives you this confidence? What do you see in the features that your clients are running on your platform in their enterprise on mission critical. What is it that you see that gives you this confidence?

Mike

But then this idea of running essentially the same query and always getting the same answer, I think makes everyone feel like they can trust the system. Trust is a huge topic around AI and unleashing it, putting it, if you will, in the banking context, putting it closer to the money, that becomes the big risk. Will it be consistent? Will it be reliable? Is this. Is this kind of not only about governance but being thinking about AI and being able to assure that AI is going to be doing the job, that it can truly be mission critical? How do you think about this moment? How do we how do we know we’ve got it when we can ask the same query, get same consistent inputs and outputs? And so the trust can be there. Like, how does this actually happen? And what are you doing? And what are you working on today that gives you the confidence to say, yeah, look, agentic is for sure going to be taking over a lot of those manual, highly repeatable, consistent actions? What gives you that confidence?

Mike

And now you’re talking about these deterministic workflows. For those of us that don’t know what that is, just explain deterministic workflows.

Mike

And banks are doing that today with agentic workflows?

Mike

This is great. So, deterministic workflows. It sounds like it is really at the heart of trust building within banking. Let’s go to another industry that’s like super regulated, a lot of compliance needs, which is insurance. This one for you, Serban. What’s more likely? AI mainly helps insurance employees do their jobs, you know, query this, summarize this, or agents start owning large parts of the workflow themselves. [unclear], we’re in insurance now, so it’s a different industry.

Mike

Okay, but let’s let’s test that for a second. Do you see this happening with claims, underwriting, distribution? Is this actually happening today in those key parts of insurance?

Mike

Okay, okay. So we’re pretty confident here in both banking and insurance about what’s more likely to happen in the future. I think it’s interesting that you bring up human in the loop. Because we have to be a little bit careful, don’t you think, Bogdan? Because if we put humans in the loop at every check, we might actually end up creating a lot more work for the humans. So there’s like almost a design question about when and how you bring humans in the loop. How are we thinking about that in insurance?

Mike

That sounds like the quick win for any insurance company. What’s the highly repetitive standardized workflows that humans are just spending way too much time on.

Mike

And the real hope, though, is that then when you do bring the human in, it’s a high-impact area, but you also can better understand judgments, but I would hope that you can spend more time helping the customer because you’re doing less of the back office paperwork.

Mike

Okay, so we’ve got the evidence now, I think. Banking and insurance, they’re on the way. Agentic is coming in. It’s making those efficiencies, helping move faster. Well, if we talk about moving faster, logistics is an industry that’s built on moving fast lots and lots of component parts, lots of complexity. The nice thing about this is we don’t have the same degree of privacy and security issues, but we have a lot more frequency of transaction. So Serban, what’s more likely? It’s quite a different industry. Logistics companies they’re just going to get better dashboards. It’s too complex. There’s too many moving parts. Or agents actually start doing some real operational work in logistics.

Mike

Now one of the key things I discussed with Cristina, the FlowX.AI expert in logistics, is what’s key here is the integration across a di number of different platforms. It seems like if there’s the integration with the data, in logistics, they can then put agentic to work and make real mission critical outcomes. How do you see that? Do you see that happening already with clients and with the technology?

Mike

And do you have any COBOL as well? I mean, come on, that’s got to show up somewhere.

Mike

So what I’m getting the feeling from both of you is that you’re quite confident about agentic acceptance. I think what’s interesting is if you look in the press and media at the moment, the questions around AI and controls and how much and how embedded. It would sound that you’re much more on the boom side of the discussion than the doom side. Is when you think about the future and what’s more likely, is that a fair characterization of your point of view?

Mike

Perfect. So now we’ve kind of looked into the future. We can see a lot of evidence that we’re going way beyond ChatGPT on the side over here and it’s coming into the workflow, into the decision making, into the reasoning. I think it’s also we see the potential here, but what we have to do is really ask ourselves, like, what’s fact and what’s fiction? And are we overreacting to some of the hot topics in the industry at the moment? So I want to take a step back and I want to frame this next set of questions for you both around. Are we overreacting? So we need to be a little self-critical here. Let’s see how we go. Let’s talk about banking again. Is this the following statement? Tell me if you guys think this is an overreaction. Nobody will manually process a routine banking workflow in three years. So in three years from now, all that routine stuff is fully agentic. Am I overreacting in this statement?

Mike

Bogdan, some people listening and watching us might say, well, these guys are optimistic. In three years, all that routine stuff at the bank is fully agentic. Why do you think many companies many enterprises and their technology partners struggle with the sort of outcomes that we’ve been talking about. Where do people get wrong the deployment of agentic from the sidelines and putting it into the heart of the game? What do you see the mistakes that others make?

Mike

So do you think wha I’m as you’re talking, what I’m seeing is I think if I was to say it back to you, is that when people get agentic wrong in the enterprise, they’ve thought only of it as just a model, just a prompt. It seems to me you’re talking about systems thinking and seeing it more holistically. Is that the key to like unlocking all the value?

Mike

Beautiful. Well, let’s go from banking to insurance now, Serban. So you have to tell me, am I overreacting? And let’s figure out how this comes together. So straightforward insurance claims will become almost entirely autonomous. Is this an overreaction? So this is the full chain of custody in the claim. Can we really get to such a nirvana?

Mike

I can see how excited you are about putting agentic at the heart of the enterprise and all these mission critical apps. I want to bring you back though to insurance and the claims in insurance. This is like as well, obviously, is underwriting and distribution. These are big parts of the insurance scheme. You have that same you don’t think we’re overreacting. We’re not getting too excited about agentic when we talk about insurance claims being largely agentic done. That’s not an overreaction. You think it can happen there in logistics and the banking equivalent too?

Mike

So why? For some people, I think they’re going to think where the three of us are drinking the Kool-Aid, right? We’re all on the train for AI. What’s your opinion on you’re obviously seeing material value and evidence that agentic is already having today, in the enterprise, why do I feel like when I reflect on what people talk about publicly seems more negative? What are they missing? We talked earlier about going beyond the prompt and the systems thinking. What else are people missing? Because it feels like you’ve figured something out that I don’t think the industry has really figured out itself. What do you see as the missing ingredient?

Mike

So this is quite interesting because when we’ve challenged ourselves a little bit, are we overreacting? Are we drinking too much Kool-Aid. I think in both of your answers, and I’ll check this with you, Bogdan, I’m hearing the following. It’s way more than the model, way more than the prompt. It’s systems thinking. It’s looking at the patterns and putting the human in the loop at the right time and not having the limitation of pilot proof of concept thinking, production thinking, institutional AI. Are those the secrets that you guys have figured out?

Mike

Now, this one, recursive self-improvement is obviously a huge topic right now. If you listen to the frontier models, that’s what scares them a little bit about the future. Are you actually doing recursive self-improvement to make products info is better today?

Mike

This is great. So let’s just recap before we go to the third installment of our talk. We’ve really kind of captured this idea of what we see happening in the enterprise right now. We’re seeing a lot of characteristics of how agentic can win, how it can provide incredible value. Not just on the side with a few answers, but putting it into action. We’ve had some key concepts come up like deterministic workflows. We’re going into systems thinking. We’re thinking really what it takes to work at large-scale, cross-border, cross-department. This seems to me, as exciting and as elegant as just a prompt is, there’s so much more to the story of winning when you talk about mission-critical and agentic workflows. So now, what I would like to do together is wrap up this conversation and help people understand your experiences in the reality of these mission-critical operations inside the enterprise. So, Serban, I want you to really give me the proof. I want you to give me the evidence that you have for putting agentic in the heart of the business. The first one’s for banking. What I’d like you to prove to me and to all of the viewers. AI can run mission critical banking workflows in production. And here’s the killer word today. So I want you to think about like maybe an example of a function that you see working today from FlowX.AI on a bank in a mission-critical area, give me something real and tangible that is really happening. It’s not Sam Altman and Dario getting too excited and promising the world. Give me the reality.

Mike

And are we just talking about a summary?

Mike

Is this just on the financial docs? Or is this also like deeds and titles of properties? Can it or is it just the finance docs?

Mike

You see that happening today on clients executing on real mortgages in a bank?

Mike

Just one country, or have you seen it in more than one country?

Mike

Okay, okay. Let’s take insurance. Obviously, claims, underwriting, there’s a whole dimension of risk analysis that’s happening. What are you seeing today in insurance on those key actions inside the business, the mission critical steps? What are you seeing? What evidence do you have? What’s getting executed today?

Mike

Is it getting involved in maybe, Bogdan? Okay, so we’re scanning for the handwriting, we’re looking at the policy, but there’s a there’s some decision making and reasoning that needs to happen after that, right?

Mike

So, are you saying that right now you’re seeing FlowX.AI process handwriting, digital documents, digital images as part of that claims process?

Mike

And is it just humans making the decision at the end?

Mike

So it’s doing the full chain and then presents it all to the human.

Mike

If it’s doing all of that, like I just imagine, like I struggle to read my wife’s handwriting, right? You’re saying you can do handwriting, you can do images, you can do the digital docs. That sounds like a huge gain in efficiency.

Mike

And it’s happening right now.

Mike

Let me do my check. Multiple languages, multiple countries.

Mike

Okay. Serban, you were going to say something?

Mike

When I was talking with Cristina about logistics, she talked about this paradigm of: we’ll figure it out later. And it’s that moment where the truck with the goods gets to reception and they can’t match up the digital document, workflow, and acceptance. And then what they do is they put that package to the side in reception, like just put it over there. We’ll figure it out later. So, surely, when we think about the evidence, you can’t. Are you what are you seeing at this moment where they’re all saying we’ll figure it out later? Can your technology actually make a difference so that we can actually marry up that analog and digital world together? So we don’t have this massive inefficient, all these things that get reconciled days, sometimes weeks later. Is that really happening now?

Mike

So it’s interesting, isn’t it? Like the themes of what the three of us have been discussing, we’re seeing like a promised world where speed but also potentially accuracy because there’s like so many opportunities for humans to make natural human mistakes, particularly on those low level, highly repetitive workflows. So it seems like there’s a lot of confidence and a lot of optimism about that future. But on the other hand, I think there’s hard work to do, deterministic workflows, systems thinking, adaptive intelligence, self-improvement, making sure that it’s production ready, not just a nice pilot. Let’s start with you, Bogdan. So for the next year, what’s your focus to take things to the next chapter when we talk about mission critical enterprise technologies, when we think about a lot of agentic workflows? What’s the big thing you’re working on? What’s the next big jump you want the FlowX.AI platform to take?

Mike

Okay, so let’s just capture what you just said. So most people will experience AI as this question and answer. And we’ve firmly moved it into agentic workflows, reasoning, decision-making, orchestration. Okay. And one of the topics that’s come up today is maybe, you know, like building MCPs. Then there’s, of course, all the agents that FlowX.AI already has. But did I just understand what you said? You’re now thinking about building small, large language models for the clients of FlowX.AI. And this must open up local, open, and sovereign AI topics.

Mike

Now, that would have some huge advantages, namely the sovereignty of it, meaning that you’re not providing data to a frontier model. You’re not helping them train their own models. You’re training your model.

Mike

You get the control and the governance around it because it’s your model. It’s trained on your data. That sounds hard. And if we look at the balance sheets of OpenAI and Anthropic, it looks rather expensive. Is that too good to be true, Bogdan?

Mike

And you’re saying that you’re you’re referring to I think you have more than twenty large language models already on Hugging Face, right?

Mike

So anyone watching this could go and download those.

Mike

Okay, well, it sounds like you’re going to be busy for the next 12 years. Serban, with all of the progress with FlowX.AI, working on clients, and all the opportunities we’ve talked about today, what do you focus on for the next year? What’s got you captivated, curious, energetic? What are you building next?

Mike

That has a big theme for me of democratizing mission-critical, agentic workflows for the non technical user. That I mean, that sounds like a like a game changer.

Mike

Yeah. Well guys, I want to say thank you very much. I think we’ve firmly made a case for not only agentic, but how we’re going to accelerate the enterprise. It feels to me like you’ve given so much evidence for the real way to do this properly. It’s not just a prompt, it’s systems thinking. It’s deterministic. It is self-improving. I think what’s so fantastic is I feel like today we got a little bit of sense of how to make smart decisions to accelerate the enterprise, to put a bit of agentic AI magic dust on top of all the good work that those enterprises have done. Thank you very much.