The Swimm Podcast · Episode 02

Alex Glushenkov: taking R&D AI-native

Taking a whole R&D organization to AI-native: how onboarding changes, and the new roles that appear along the way.

GuestAlex Glushenkov
Season1
Listen on YouTube Coming to Spotify

In this episode

Alex Glushenkov leads engineering and implementation at ThetaRay, an AI company that helps major banks detect financial crime. In under a year his organization changed how it delivers software, and this episode is the account of how. It starts with a two-day internal hackathon that produced 37 initiatives and a projected saving of 500 person-days a year, and runs through Project Nova, in-house AI tooling that cut customer onboarding effort by 40 to 50 percent. Taking an R&D organization AI-native changed the people side too: a new role, AI Delivery Manager, folded five job functions into one, and a single developer shipped in a month what a team of four had scoped at two years. Alex is candid about where it gets hard, running AI against legacy code with a human still in the loop, and closes with three practical steps for any engineering leader starting now. Recorded in Hebrew.

Episode transcript

Translated into English from the original Hebrew recording, based on the YouTube Hebrew auto-captions, and lightly edited for readability. Timestamps follow the video.

00:01Oren Toledano, Swimm CEO

Hi Alex, how are you? Welcome to the Swimm podcast, and thanks for agreeing to be interviewed. I'd love it if you could start by telling us a bit about yourself: your background, the path you took, where you are today.

00:18Alex Glushenkov

Great. First of all, it's fun to be here, thanks for inviting me. My name is Alex. I've been in the industry for over 20 years. I'm married, a father of two, and I live in Givat Shmuel. I have to say that, otherwise I'll hear about it at home later. [laughs]

00:36Alex Glushenkov

So, about 20 years in the industry. I started my career as a software developer at Orange and was drawn from there into management roles, whether in product management, technology, or project management. After 14 years at Orange I moved to a company called Kaltura, which did an IPO a few years ago and focuses on video solutions. There I also led engineering and implementation teams for our largest customers, like Beeline in Russia and Kazakhstan, Entel in Chile, Cellcom in Israel, and many other large media organizations. After about six and a half, almost seven years at Kaltura, I moved to a small company called Facebook. [laughs] I was there for about a year, working on compliance in Meta's fintech area. After Facebook I moved to a company that provided proptech solutions to organizations, and I was there for about two and a half years. Today I'm at ThetaRay, where I lead the engineering and implementation teams for the largest customers in the financial world.

01:53Oren Toledano, Swimm CEO

Quite a career. Such small, insignificant companies. [laughs] ThetaRay is a company not everyone knows. Can you tell us a bit more about it, so we can learn about the company too?

02:09Alex Glushenkov

It may not be well known in Israel, but abroad it's fairly well known. ThetaRay helps large financial institutions, like banks and fintech companies, find financial crime: money laundering, financial fraud, and people who are under sanctions from American and European bodies. ThetaRay has been in the market for many years, and it has a unique engine that provides AI solutions to financial institutions and finds financial crime with that engine. It's really one of the unique things in the market today. We work on this day and night and really try to help financial institutions find this financial crime. Our biggest customers are banks across Europe and the world, like Santander, Crédit Agricole, and other financial institutions.

03:20Oren Toledano, Swimm CEO

That's really a company with a mission. Nice.

03:23Alex Glushenkov

Right.

03:24Oren Toledano, Swimm CEO

OK. We're here to talk about a transformation related to AI. In our prep call I understood that a year ago you were in one place and now you're somewhere completely different. So what changed along the way?

03:42Alex Glushenkov

I joined a year ago, although this project started a few months before me. Basically, when we saw that the AI tools were already mature, at a level where you could use them to deliver implementation and onboarding of our biggest customers onto the ThetaRay product, and we saw that this process takes quite a long time, not always because of us but also because of us, we said: let's see if we can improve the onboarding of customers to ThetaRay with AI tools. So we started a project called Nova, which is about how we improve and speed up customer onboarding to the ThetaRay product. We really started small. We broke down all the tasks involved in onboarding and implementation into very small pieces and looked at how we could build AI solutions for them, to make things easier both for our employees who do the implementation and for our customers.

04:52Alex Glushenkov

Today we're more than halfway there. I think by the end of the month we'll finish the AI onboarding solution for our fintech customers, and we have a fairly full roadmap through the end of the year to finish it for the rest of our customers. And it already works. We've used the tool with these customers, and I think we've already cut somewhere between 40 and 50 percent of our effort in the onboarding process, just with the tools we built for implementation and onboarding of our customers.

05:29Oren Toledano, Swimm CEO

You even told me you created a new role called AI Delivery Manager. Can you say a bit more about that? It's really interesting.

05:36Alex Glushenkov

When you onboard a customer to a product like ThetaRay, many roles in the organization help. There's a project manager. There's what we call a financial crime expert, who maps the customer's financial risks into something we can then develop. There are our data engineering and data science teams, who take what the financial crime expert did and turn it into an implementation in code. So at any given time you have roughly four or five roles working on a customer's onboarding. We said: let's change the approach and create a role that consolidates all of these into one. We called it AI Delivery Manager. The other teams sit behind it, so if there's something the AI Delivery Manager can't do alone, they open a proper ticket to the FC, data science, or data engineering teams so they can help with the implementation and onboarding.

06:42Alex Glushenkov

I have to say that along the way we realized we could do more consolidation of roles. For example, we merged data engineers and data scientists into a single role now called Full Stack AI Data Engineer. And it goes further: our R&D teams are consolidating roles today too. I think that's the trend in the market. With AI we'll be able to consolidate many roles into one, be much more efficient, and give our customers much better service. In the end it's much better to talk to one person than to four.

07:17Oren Toledano, Swimm CEO

It sounds like you managed to find an area inside the company that could create significant impact quickly with AI. On the other hand, we're also heading toward a world where every company is making some shift to AI, changing the operating system of how a company is run and how employees work. How did it spread at your company? You said it caught on across the rest of the company and people started all kinds of initiatives. Can you tell us how that worked, and how people who aren't developers started building with AI?

07:54Alex Glushenkov

It's really amazing to see. Our HR teams and our finance teams, for example, use AI on a daily basis and do very nice things with it, to improve reports and all sorts of other things. In the technology teams I manage, what we did was run a hackathon for all our data engineers, data scientists, and others. We told them: guys, take small problems and try to solve them with AI. We did one in Madrid, because we have a very large site in Madrid, and we did one in Israel for the developers here. From just those two days, about 37 initiatives came out that could potentially save us around 500 person-days a year, plus around 50 thousand dollars annually. And that's just from those two days. Of course we looked at every initiative at the level of the actual task and saw that it really holds up.

08:55Oren Toledano, Swimm CEO

I wanted to ask you: 37 initiatives is a very big number. How do you decide what to implement and what to set aside? You know, at hackathons there are always nice ideas, but how does it end up becoming something real?

09:07Alex Glushenkov

Broadly, we looked at it at the task level, as I mentioned a second ago. There were many more initiatives, I have to say, about 45 or 50, and not all of them were accepted. We asked everyone to really look at it at the task level. Every data engineer, data scientist, and so on said: today a task like this takes me, I don't know, five days. If I invest two days now in building a technical solution, I'll bring that task down to one day. And we saw that it worked. People are now in implementation, and by the end of September we're supposed to finish all these initiatives.

09:44Alex Glushenkov

By the way, people are doing this on the side. It's not something that was in our work plans. People love it so much that they do it alongside their work, and they understand it's important for their careers and their future. The more you develop with AI, the better your chances of succeeding in the new world. Whoever doesn't will simply not survive the revolution that's happening today.

10:11Oren Toledano, Swimm CEO

OK, let's go back to the world of development. One of the nicer things about AI right now is the ability to ship products fast, and you have a good story on that. I'd like you to tell us about the fun side: how you get from an idea to a product quickly, with a small team or even one person. On the other hand, I'd also love for you to talk about the tension with existing code: how AI deals with it, and how you need to bring people into the process so you don't break things. Because AI knows how to create fast, but it also knows how to break things fast. If you can talk about both sides of the coin, I'd love that.

11:02Alex Glushenkov

Let's start with the second question, because that's the biggest challenge today: how you take existing code, really bring AI into it, and make sure the code that comes out is high quality and doesn't break things. Right now there's still a human in the loop. AI generates most of the code, but all code goes through a human review. A person looks at it and checks that the code is correct. We also do very thorough testing with our QA people, to make sure that everything leaving our hands, whether written by a person or by AI, comes out at a high quality level.

11:38Alex Glushenkov

We did see that we took legacy code and, quite simply, managed to turn it into more modern code. I won't say it was 100 percent bulletproof, not 100 percent, but almost 90 percent of the code was at a very good, reasonable level. The other 10 percent we had to fix manually. But I think that's improving too. You see a new model coming out every day. Yesterday Anthropic launched two new models. With these models, and we see it ourselves, with 4.8 and also with OpenAI's 5.5, I have to say the code is generated at a very, very good level.

12:24Alex Glushenkov

As for your first question, we have an interesting story. A while ago we developed and launched a product called Ray Platform. It's an agent that sits on our product, though it doesn't have to sit on our product; it can sit on other compliance products. The agent helps the analyst get to an in-depth investigation much faster once we've identified suspicious activity. It collects all the data from the internet about that case and brings it to the analyst, cutting their investigation time by almost 90 percent, because it brings them all the information and the analyst only needs to look and check that everything is correct.

13:01Alex Glushenkov

The development process for this was really amazing. One developer sat with the customer, understood the customer's requirements, and basically developed and deployed it. The person who did it told me that he did in a month something that four people would have done in two years. That's pretty impressive. He used the newer models, of course, like 4.7 and 4.8. It's really crazy. Every morning we sit and look at the progress, and the progress is like a hockey stick. It's moving very, very fast, and I think it gets better every day.

13:55Alex Glushenkov

And if you remember, we talked about consolidation. I think that's the next stage. Probably all of our development as we know it today, where there's a product manager, a designer, and a developer, will also change in the medium to long term, and there will be one person who can do all of it. And I don't think we need to be afraid of that. I think we need to...

14:22Oren Toledano, Swimm CEO

I have a follow-up question about the cultural side of this, and we'll get to that in a moment, because it's also interesting how it's felt inside the organization. But take the example of this project, which was developed very efficiently, with a lot of code created very fast. You can imagine this becoming an exponential process for you: more people will do a lot more, so there will be a lot more code and a lot more products. One of the common conversations right now is how you preserve context over time, so that agents stay effective inside something that's growing much faster than before. If we talked about the challenge of existing code and how you work with it, when you generate new code you're actually growing the challenge. Have you put any processes in place in the development organization to preserve that connection between the models and agents and the existing code, so there's no drift in context?

15:25Alex Glushenkov

You're completely right, but I have to say it's something new. We really only started with this about a month ago. We still need to solve the problems. Not everything is solved, and probably not everything will be solved in another month or two either, but I believe that over time we'll find the right solutions. In the end, the most important thing is to make sure the product that leaves our hands is a good, high-quality product.

15:46Alex Glushenkov

Right now, because there's one person working only on this, it's fairly under control. I think the next challenge is to see what happens if we add another person alongside that person and double the development capacity: how they work together, how they don't overwrite each other's code, how they make sure that one person's development standards are also the other person's standards. So yes, we use skills and MCP and make sure there's grounding, but to be honest, it's not 100 percent solved yet. I think we're still at the beginning of the revolution, and as I said, you don't need to be afraid. You need to fix things on the fly. This probably isn't the only challenge right now, and there will be many more. You just need to be clear-headed, not afraid, and go into it with an open mind, and we'll find solutions.

16:41Oren Toledano, Swimm CEO

Speaking of not being afraid: do you get hallway conversations with developers about this improvement, about whether we'll become redundant? What's the feeling? As an engineering leader, how do you deal with that yourself, personally, with people? Things are changing very, very fast. How does it sound in the company's hallways?

17:04Alex Glushenkov

I think some people are afraid and some are less afraid. I think whoever isn't part of this revolution and doesn't adopt the new ways of working will probably be pushed out of the job market. But I don't think it will be a loss of jobs the way people talk about it today. I think new opportunities and new roles will be created that we probably don't even know about yet.

17:34Alex Glushenkov

I myself developed, in a few days, two new pages on our website for Ray Platform, the new product we launched, and it was simply amazing. I had already forgotten how to code, but I came back to it and it's simply addictive.

17:53Alex Glushenkov

In the end, I see AI as a revolution quite similar to what happened with iOS and Android, or even refrigerators. Someone builds the infrastructure, and someone uses that infrastructure to make much bigger profits. Take refrigerators, for example. When refrigerators launched, a lot of people developed refrigerators. But who profited most from refrigerators? Coca-Cola, right? Because they were the ones who used refrigerators and grew their profits much more, or all the other food companies. Same with iOS and Android. When iOS and Android launched, Apple and Google of course made very good money, similar to what Anthropic, OpenAI, and Google are doing today. But there will probably be a great many opportunities, and a lot of money, for solutions built on top of those AI capabilities. And you see it today. People talk about SaaS companies being hurt a little, or a lot, by AI, but I assume people are already working on other solutions that will adopt the same tools to build much better and much more efficient products.

19:04Oren Toledano, Swimm CEO

OK, I agree. To wrap up: somewhere out there is an engineering leader who hasn't started this journey yet. Everything is still very new, but they're watching us now and want some tips on how to start. What would you tell that leader, if there are already lessons learned from the short process you went through?

19:33Alex Glushenkov

Sure. First, as I said before, don't be afraid, and start small. Take a small problem, break it down, understand whether it can be solved with AI, and solve it with AI. When you see that it works, move on to the next problem, and go step by step. You'll see that within a month or two the pace will be crazy. Once you start, it's addictive, and you'll understand that you can do a lot of things much better and more efficiently than you did before.

20:03Alex Glushenkov

Two, you need to make sure that all the processes in the company today support this development. For example, as we discussed earlier, around consolidating the code, or how you deploy that code to production. You need to make sure your processes can support it.

20:24Alex Glushenkov

And I think the third piece of advice: be in podcasts all the time, be in WhatsApp groups all the time. In Israel there are lots of WhatsApp groups on AI. There are meetups all the time. Meet people and check things out. Not all the wisdom sits in one place; the wisdom is spread around. For example, today our biggest problem is tokens, right? We waste a lot. There are many people already working on that, and again, it's not 100 percent solved, but you can get help and you should ask others for advice.

20:57Oren Toledano, Swimm CEO

I'm afraid to open our token bill at the end of the month.

21:02Alex Glushenkov

So I'm telling you, there are a lot of things you can do much more efficiently. There are special skills and all kinds of solutions already working on reducing token consumption. But again, consult others. Many people want to help, and many people are new to this, so knowledge is flowing very smoothly right now. And the revolution is already here, so either you're in or you're out. Better to be in.

21:33Oren Toledano, Swimm CEO

Right. And just like you started small and then scaled up, that's true for anything new in life. So you're saying: let's use human wisdom for this challenge too. Alex, thank you so much for the time you spent with us and for the lessons you've learned. We're learning together with you a bit from the experience you shared with us. Thanks a lot for your time, and good luck with the rest of this transformation journey.

22:06Alex Glushenkov

Thank you very much, I was happy to be here. Thanks.

More episodes

All episodes