AI Engineering in Action: Diffco's Approach to Scaling Companies with Vadim Peskov

Summary

Vadim Peskov, CEO of Diffco, shares insights on AI development, focusing on practical applications, innovation, and navigating rapid technological changes. Discover how Diffco translates client needs into cutting-edge AI solutions and the future of AI interfaces.

Takeaways

AI development philosophy
Client-centric AI solutions
Innovation and rapid iteration in AI
Future of AI interfaces and tools
Challenges in AI growth and adoption

Soundbites

"We experiment with new AI tools to stay at the forefront."
"Avoiding hype and focusing on ROI is essential in AI services."
"Agentic AI is the most exciting trend right now."

Chapters

00:00 Introduction to AI Speed and Vadim Peskov
01:10 Diffco's Core Philosophy in AI Development
02:03 Evolving AI Products and Innovation
02:43 Translating Client Needs into Practical AI Applications
03:58 Focus Areas for Diffco as CEO
04:53 Ideal Client Profile and Scaling AI Solutions
06:02 Challenges in Growing Diffco
07:38 Adapting to Rapid AI Advancements
08:54 Avoiding AI Hype and Focusing on ROI
11:01 Raising Awareness and Business Development
12:21 Addressing Challenges and Future Plans
14:00 Expansion and Growth Targets
14:49 Exciting Trends in AI and Software Engineering
16:15 The Future of AI Coding and Development
16:21 Vision for the Next 12 Months
17:18 Closing Remarks and Key Takeaways

Video

https://youtu.be/JYS8YlpfwhI

Transcript

SPEAKER_00 0:20

Welcome to AI Speed, the show where AI-powered companies talk about what actually works in the market right now. Business doesn't move at internet speed anymore, it moves at AI speed, and the people who figure out how to turn models into money will own the next decade. I'm Evan J. Chaulfin, founder of Luxhammer and growth partner to high-performing brands. Today I'm thrilled to be joined by Vadim Peskov, CEO of DIFCO. Vadim leads a cutting-edge software engineering company that specializes in bringing complex AI solutions to life for clients across diverse sectors from healthcare to fintech. He's an expert in navigating the AI landscape, ensuring practical application, and delivering measurable business impact. Vadim, thank you for being here.

SPEAKER_01 1:11

Pleasure to be here.

SPEAKER_00 1:14

Absolutely. So DIFCO is at the forefront of AI development. What's the core philosophy that guides the company's approach to building AI solutions for clients, especially and ensuring they deliver real business value?

SPEAKER_01 1:28

Man, this is so different project to project, realistically, and changing basically almost every week. But if I want to give a like a summary view here, realistically, we have kind of a mantra of constantly exploring and innovating and changing things and not sticking with one approach that we had before. Again, when AI development and enhanced development started, originally it was, hey, it will be autocomplete. And now it's like full agent workflows. Slightly different, different logic, different systems. And the product that we used three years ago doesn't really make sense to use today. And even like the product that we used a month ago, maybe doesn't really make sense to use anymore. So it is constantly innovating and trying new things and exploring what we can do better with this. So it's not a specific workflow that exists anymore. Things are changing, and even like agile doesn't really make sense in a way of INA enhanced development. So I think the core idea is to be in front of those things and delivering accelerated growth to our customers and allowing them to see the future and build for the future, not built for even current things, because things will be different, and we will explore this uh in a few minutes.

SPEAKER_00 2:49

That's great. And what makes TIFCO's approach unique in the AI development software engineering space, particularly and how you translate client needs into cutting-edge practical AI applications?

SPEAKER_01 3:02

So, first of all, we had a lot of experience building regular B2P SAS products, which really helped understand about what uh could be done. However, the core aspect of creating all is actually not the specific tool that you use would be cursor, cloud code, whatever you use doesn't really matter. The core aspect here is the context and the symbol you do it. Um AI still cannot be uh everything, and you still need to guide this, you need to explain it, what needs to be done, and what does the best outcome actually looks like. So, in many cases, the management approach and really digitalizations of the specifications and understanding what the architectural system needs to be built, I think this is the core aspect of DIVCA. Because again, if you ask AI, it wants to build the Kubernetes clusters with a bunch of microservices. Great, but in many cases, you don't need this. And ability to understand and navigate those aspects, even on those like architectural simple levels. I think this is the core.

SPEAKER_00 4:07

Yeah, definitely. So where are you focusing most of your energy right now as CEO of DefCo?

SPEAKER_01 4:13

Right now, to be honest, it's uh testing your tools and testing what can be done in the future. Plus, um, I do involve with a lot of product strategy for some of our clients, uh, as well as on my my investment side. And there are my focus is to see how the interfaces of the future will look like. Because it definitely will not be interface where you're doing things. It will be interface where EI will do things for you and you guide it, approve it, or you enhance this. It's not the interfaces where you need to do multiple clicks to get results. It's interfaces where you see the done work and just basically move on or provide a new context there.

SPEAKER_00 4:57

Absolutely. And so who do you serve best? What's the ideal client profile that benefits most from your AI development expertise?

SPEAKER_01 5:06

We do work with specific type of clients for right now. Um again, always depends on what uh they really want and how they want to do this. So most of the our project is mid-sized enterprises and growth stage startups. So this is the core aspect of the folks that really uh benefit best. We do work with uh some real stage folks. Most of those are second-time founders that sold their company multiple times, or at least one time, and uh want to enhance and build really great products there. Most of those projects are large complex systems. However, we also have some some of those projects that are like a simple build that we will do in a couple of months, etc. Typically, this is the folks that already have something working or have some prototypes and they want to scale. So we're here to scale, can we can build MVPs, but most of the time it will be best to scale with us and build MVPs now is like wherever our bytecoding tools exist and you like.

SPEAKER_00 6:06

Yeah, absolutely. So digging in a little bit, uh, what's been your biggest challenge in growing DevCo this past year?

SPEAKER_01 6:14

It is a challenging question, man. Uh realistically, the biggest challenge that I see primarily is all about um uncertainty about the result that you can get with the tools and how you can rely on tools that's now, let's say, 90% efficient, but you want to see them uh at 100. And how to develop a system where you know that AI will get better, you just don't know how and when. And those logics I think is the most challenging approaches. And basically seeing how the software industry evolves, it doesn't really make sense to develop things without AI. So the folks that doing this in the old way, unless it's specifically required for whatever reason, this is the way to nowhere. And the best approach from our perspective is to build systems that can enhance and can grow with AI. And again, sometimes this means optimizing and changing things almost monthly, or sometimes more from depending again on a product. I think we're not longer in a way that we can build a product that can be sustainably growing without any updates for longer. Again, HDLC in a wake of AI is quite different.

SPEAKER_00 7:32

Yeah, yeah, definitely. And how are you handling the need to continuously adapt DEFCO's capabilities and stay at the forefront of rapid advancements and AI tech and evolving client demands?

SPEAKER_01 7:45

I I think like realistically, again, it's just trying new things and experimenting with new things um and suggesting this to the clients. So that's why we have a longer relationship with uh most of our clients. Because we want to um bring in new tools to them and allow them to explore and make this much more efficient. And now basically, in many cases, 95% of the things actually written with AI, sometimes hundreds. And realistically, we can enhance many projects with this. And give them not only the technological edge, but also the product edge. I think this is where it's not a question how to build. I think this is primarily solve the problem. The question is still and will be what to build and how to build the product that people love. And especially, again, our expertise is in a business side of things, how to build something that large businesses will adopt and be happy to use and recommend. So this is where we basically pushing this forward.

SPEAKER_00 8:50

Yeah, definitely. What makes it difficult to avoid sort of the hype traps of AI and focus on ROI when presenting AI development services to potential clients?

SPEAKER_01 9:01

I think like it's a lot of traps that happen every day with AI. Exactly. And a lot of hype. And realistically, you don't know about the hype until it's a hype. So you like playing the with a new tool and you're trying to accelerate with this, and realistically it's end up being like the most kind of named thing ever. And uh, for example, you like everyone would be like agentic, agentic, agentic, and you like come on, like stop this. And now like everyone is again obsessed with uh different levels of agents, demon level of autonomy that you can do with uh open claw, a cloud code, and others. Uh and I think this is the an amazing tools. However, the core question is how you not just like, hey, this is the best tool ever, how you can actually implement this realistically and meaningfully, because in many cases it's so many security issues with many of them, and people again doing the stuff that you like, yeah. So they try to accelerate their work, great, but realistically they put in so many things at risk and creating so many problems in the same way. So I think especially when you're doing a complex systems where you have a lot of people relying on you when you have a legal responsibility, this is the different way to look on hypes. It is to plan those things and account for every possible thing that you can imagine, but realistically, it is not an end game ever, because tomorrow it will be another thing, or this thing will break, or they will find another vulnerability, and you will need to enhance and you will need to change this. So realistically, you just need to move this forward and always try to add new things and see from the prism of making sure that whatever you do is actually helping the business, not just doing because of like it's a hype.

SPEAKER_00 11:01

Yeah, absolutely. What challenges do you face raising awareness about your company and its services?

SPEAKER_01 11:06

Man, like again, we work with uh a lot of clients directly, and we have a wonderful position where we have a lot of uh a relationship that we can grow. However, realistically, right now we will be launching a few new products under our umbrella under DivKI. And launching there, I think this would be more challenging to enhance and build a much more comprehensive um awareness, uh, especially uh across a larger sectors than we're currently working with. I think the most challenging thing is differentiation from the any kind of noise that you have and trying to explain in a professional way to the customer what is actually the most important thing for them and what they actually need to focus on. I think this is the uh the best kind of logic here. However, again, we'll see because now we will be half product, half service in a way, and this is will be an interesting interesting playbook. Again, I did this multiple times for many clients of ours, and now we're kind of like pushing this into the public realm for Divca.

SPEAKER_00 12:17

Yeah, oh that's great. That's exciting. What do you do to address that challenge?

SPEAKER_01 12:21

I think uh right now, just talking with directly with many clients as we can and trying to explain them the logic about the processes there. And after I think we'll be more public about this. Ever right now, I think we have a good uh internal place to play with those things and enhance those. And we probably will be going semi-publically on this slightly sooner, probably again, beginning of next quarter or so. We'll see. So it's uh again, I don't have like specific challenges because we don't disclose those things yet, because it's a work in progress. And I think realistically the AI service business is will be changing. And the most important thing that I just want to make here is that AI service business is actually will be changing every public quarter. I think we're in a wake of really constant pivots, not in a wake where one product will, that's all. I think this is the where you will need to enhance and add more things to ensure that whatever you do is actually relevant. Because no one wants to use uh OpenAI GPT 4 anymore or even five. Everyone wants to use the best products and the latest products. And I think this is the logic that we'll have with the services as well.

SPEAKER_00 13:42

Yeah. Yeah, that's totally understandable. So uh I understand, you know, you have a lot of great clients. Are you looking to continue to expand your client base? And if so, like do you have certain growth targets?

SPEAKER_01 13:56

So I th I think like for us, our core expansion goals is in terms of the enterprise sector on the Fortune 5000, on mostly combination of non-technical clients and technical clients. For technical folks, we have a slightly different approach how we work with them. However, in terms of the non-technical, it is mostly upgrading a lot of legacy systems and making sure that the development that they do is actually relevant and allowing to optimize them a lot, all of the development that they actually do.

SPEAKER_00 14:28

Yeah, definitely. So I'm curious um if there are any trends that you're seeing in AI development, software engineering, or digital transformation that excite you the most?

SPEAKER_01 14:39

I think the the most exciting, I think the autonomous agents is probably still the most important and uh still the most interesting from my perspective right now. We have um uh few projects that we fully utilize in this approach, where we completely switch from writing code to actually directing AI. So we're not longer writing the code itself for those projects. It's still again previewed in many cases, etc., especially in complex type of scenarios. However, realistically, I think the fact that we will probably not write the code itself uh in the next half a year or a year for many projects. Again, folks still need to read and understand and be in the bill have an ability to do this. However, our approach that we have most only senior folks and uh staff levels, this is actually helping us to build this kind of uh logic here. I think that this angle is really what will allow to change how the current software built. Because if you have the folks that understand how to build large systems, how to build them in an enhanced way, I think this is where you will have the best configuration. And this is the uh probably the best playbook that you can have uh for software development uh enhancements with AI right now.

SPEAKER_00 16:04

Absolutely. So if we were to have this conversation again in 12 months, what would make you feel it's been a successful year for DIFCO?

SPEAKER_01 16:11

I think realistically, again, everyone talking about like cutting jobs for developers. We actually hiring. However, we pushing into a different level of the folks. We only hire in the senior folks, we only hire the folks that can work with CI and can make better decisions with AI. I think the probable configuration in a year, we stopped writing code whatsoever, and we built autonomous systems really fast. And I think that the realistic fact that collecting requirements is more challenging and more expensive than a development itself. I think this is now absolutely true. And again, it's almost true right now, especially for large systems. However, it probably will be true for even the smaller uh MVP playbooks as well. Absolutely.

SPEAKER_00 17:06

Well, that's it for today's episode of AI Speed. A huge thank you to Vadim Peskoff for sharing his invaluable insights into how DIFCO's helping clients turn AI models into money and for navigating the cutting edge of AI development. If you're building or leading an AI native company or a service business that uses AI under the hood, and you care about revenue adoption and market share, make sure to subscribe to AI Speed. Learn how the best AI operators ship faster, sell smarter, and stay ahead. Thanks for listening. Until next time, keep building, keep selling, and keep moving at AI speed.