AI can build it. But should you? 🤖 Konstantin Klyagin

Let me tell you something about hypes. I have seen hypes before. When the first smartphones emerged, everyone said that they must have a mobile app for whatever they they sold. So we got tons of crap apps that no one used. Then came the cloud. Instead of self-hosted solutions, no matter how stable or good they were, businesses started moving them into the cloud, which resulted in a very expensive hosting bills. And so with AI, we are now witnessing the scene that. many businesses, they apply AI just for the sake of it. There is a very common mental trap that the more code I produce, the more successful I am, which is not true anymore. Code is indeed very easy to produce, but as you build features, you need to be disciplined. And the question nowadays is not whether you use AI or not. The question is how you use it and if you understand what you are building. I think that's the common misperception that a development team can build anything. They can't. A good team is the one that pushes back on stuff, that says this is not very realistic, or this can be done this way and suggests another way, or that keeps you in the loop and honest about the deadlines and delivery timeframes.

This week's Millennial Master is Konstantin Kliagin, who spent 2 decades running software agencies, building and fixing products for others. Now that AI has made software much quicker to build, this also means you can build the wrong product much faster. Konstantin has seen what happens when that speed gets ahead of the thinking behind it. In this episode, we talk about why building faster can create more problems, where vibe-coded products can start to break, and what gives you an edge when everyone has the same AI tools. More people are starting searches inside AI tools now, especially when they're researching a business. I do it myself, and quite often I find it more useful than a normal Google search. That made me curious about how discovery works from the other side. So I connected Data for SEO to my ChatGPT to look at that using current search and AI visibility data. I asked it to analyze how Millennial Masters and my name currently appear in AI and web search, and to find any obvious gaps. It found that Millennial Masters already has a clear association with founders, businesses, and AI, and several of my AI articles are surfacing in search. But it also found that those individual articles have a stronger search presence than Millennial Masters itself as a publication. And there was an even bigger gap with my name. Search understands who I am when you add Millennial Masters, but search for my name on its own, and I'm competing with several completely Setup takes under 5 minutes. There's no monthly subscription, and Data for SEO charges only for the requests you actually make. If you want to check how your own business or content is appearing, use my link in the description. New users get $5 in Data for SEO credits to start testing it. Konstantin, welcome to Millennial Masters. Where are you joining us from today?

Hi, thanks for having me. I'm joining from Lisbon, Portugal.

That's a popular spot right now. I've had at least 2 other founders from Portugal on the show. founders who are based there. So let's go straight into it because I know you spent your life inside software teams up until now. So I'm wondering what changed the most for you and for founders in the last 2 years that people outside tech haven't noticed yet.

All right. So I actually spent my entire life building software and 21 years of which I've been running 2 software agencies. One is called Redwork, which develops software, and the other one, QA work, that tests software. So in the past 2 years, what happened? Obviously, we are talking about GenAI. What else? And what changed is that producing code and building features and building software became incredibly easy and fast. It also gave birth to the common misperception that AI gives us speed and generates code for us so that we can scale faster. But it's not true.

If everyone says that AI is making companies faster, granted, you say that that's only half the story. So what's the other half?

Yes, absolutely. So obviously you can ship features and code faster with AI, but on the other hand, you need discipline because there is a very common mental trap that the more code I produce, the more successful I am. Which is not true anymore. Code is indeed very easy to produce, but as you build features, you need to be disciplined and you have to be checking in with your users more. So by discipline, I mean specification and quality assurance. And by users, I mean that you really need to focus and you need to know what you are building because you can be shipping features that no one uses. But yeah, the real catch is what you're building. Building became obviously less expensive. And faster, but the discipline still needs to be there.

Let's break that down then. Let's start with the users first, because you've been working in software companies, and that's one of the main things that you have a whole list of features that you want to deliver in your product. You don't know exactly which ones your users are going to need. So how do you pick which features to develop next? What would be the good process to do that versus just doing everything that you had on your mind, or you want to keep up with your competitors?

Well, you need to be doing customer development constantly. So you have to be working together with your users. You need to study what they use and how they use it, where they spend most of their time, and what can be automated or built for them to make their life easier. And according to that, you need to prioritize your roadmap and be building those features One by one, taking them from the roadmap and then properly implementing them, integrating them, educating your users, testing them, making sure they work well, they scale well, the infrastructure is ready for them. And actually you need to slow down because like everyone's got a Ferrari now. I like this comparison, but no one builds the brakes for it, or not many people.

How do you apply this rule then? If you need to slow down, but AI is making you faster. Then what's the right approach? Yes, obviously talk to your users, but I know from my experience and a lot of people, you only hear from your users when things go wrong. When things are good, you rarely hear from them.

That's a good question. So I can give you a few examples from my daily practice, and that is working with customers who utilize AI properly. So utilizing it properly or improperly. So how we got into white code cleanup. Business, for example, because there is a lot of white coat nowadays, right? And it's like, it's great for building prototypes and people come. And there is another mental trap is when people think that the prototype is the final product. But if you don't take into account the infrastructure, the architecture, security, and overall quality of software, then the prototype is good for demos, but 6 months in is when we need to come in and clean it up. So there is an example of a customer who came to us with a vibe-coded app, right? And on the surface, on the UI, it looks great, but payments fail, duplicate bookings, and many other workflow issues are there. So it needs to be tested, fixed, then it should be shipped to production. So to work live with the, with the users. There are though examples of proper AI usage. And those are, for example, PrideFit. It's a fitness app for LGBTQ+ community.

Yeah, because now you can niche down in any area because you— and create an app for every single small community because developing it is so much cheaper.

Yeah, yeah, for sure, for sure. And that's the case. So how they use AI is that they build features on their end, they showcase them, they check them with their users. And they, then they come to us for proper implementation for, uh, uh, for, for live readiness. And this is the protocol we are working with them on. Another example is, uh, an entirely AI-built product with the development team on their end, UK-based Granola AI, evaluated at $1.5 billion. And—

Heard of it, of course. Yeah.

And how they built? They work with GenAI on their end, but their discipline and their software experience, first of all, lets them define and spec out the features truthfully and AI-ready. And then they use my agency QAWork for testing. So they don't release anything until it's been tested on our side. And this discipline makes them roll out fast, but also Quality product.

Okay. So we all know that a lot of people are now building their demo product or their MVP with various AI solutions. Makes sense. It's so much cheaper, so much quicker, but once you go into the real world with the real customers, what's the first thing that usually breaks?

Well, regular workflows. So if you just ship and test the happy path, You just make sure that it works, whatever you had in mind when you were building that using GenAI. Then those things break when faced with real users, right? Then the infrastructure breaks. So consider one concurrent user against 100 concurrent users.

Yes, I've been there. And even at like 50, it can slow down your servers or go through a bottleneck. Absolutely.

Yeah. Also consider technical debt. So as if your only specification is the prompt, at some point there's gonna be a lot of technical debt that you also need some software architecture experience to put in order. And that list goes on.

You said something I really liked before we booked this interview. It was something along the lines of AI isn't really a scaling agent, but engine. But a multiplier. And I'm wondering what you meant in this context that we're talking right now.

Well, the multiplier is something that lets you accelerate your decisions and your roadmaps and your plans faster. So if you are accelerating something that you haven't thought through well, then you're getting more trouble, more garbage, more like unexpected results. But if you know what you're building, you're building it faster.

And where would you say that the bottleneck is moving once AI starts writing, say, half of the code? Because overall, when you go in with your agency and your type of work, you don't replace all the code. How much would you say you're replacing and where are the key issues that AI still gets wrong at the moment?

Well, it depends on the amount of technical debt, so it depends on how much effort it took you to build it. The original codebase. About multiplication. So it multiplies also the discipline that you have. If you have proper discipline, this discipline is amplified. If you lack discipline, if you don't test, if you deploy garbage, and if you don't spec out or think through every pipeline, every workflow that you want to build, it also multiplies that uncertainty.

Where does the bottleneck move once AI starts writing half of the code?

Well, the bottleneck moves to the specific application and also to the quality of the code. For a fairly sophisticated product, you're not talking to individual developers, you're talking to the CTO. Now, a well-experienced software engineer can become a CTO on his own, but a person with no technical knowledge whatsoever cannot think about proper architecture or infrastructure for a large-scale software product. They can build a prototype, which Which enables them big time to validate their idea, to show it to the customers, investors, whatever, but then to build it out properly, you do need software engineering knowledge. It didn't go anywhere.

So would you say that founders still need a CTO at this stage, or when they first demo a product or try to raise money for their products, at what stage do they actually need a CTO if they can vibe code an MVP?

Of course, that stage moved way down the timeline. So you can build a prototype, you can demo, you can raise investments, and you can even get your first users without a CTO role or without a software engineer role. But then, as you want to build a complex, full-scale product serving multiple customers, well protected against security hacks and all that, you obviously need software engineering experience. On your team to build it out properly, also with AI. So whenever we clean up AI-generated code, we obviously also use AI, but we know what to tell it and how to organize that code so that it works well. And we also have a QA team which tests the user journeys and workflows to make sure it works right.

So it's not that you don't use AI, it's, you know, the right frameworks and you have the technical technical knowledge to use AI to speed up your work? Because that comes close to a piece that I've written recently, that you can't outsource what you can't understand. If you don't have the technical knowledge, then outsourcing it to a development agency will not help you get you very far. Similarly, if you don't have the marketing knowledge, you can't outsource your marketing to a marketing agency if you don't know how to evaluate what they bring you back and the leads versus the money that you put in. So you would still need a CTO in a startup if you want to have a good technical product. Is that the correct reading?

Yes, yes, absolutely. Absolutely. And also with GenAI, every experienced software engineer can become a CTO on their own.

So Konstantin, what would you say then separates founders who use AI well Versus the ones who just use AI a lot? Because you can use it a lot and still be average at it without saying bad.

Well, as a founder, it depends on the stage of your software product you're in. So if you're in the beginning of your business journey, then you can be vibecoding and demoing and implementing with your early adopters your product easily. But then if you want to build it out and actually serve multiple customers and scale, and increasingly build features, then you need either an internal team with the software knowledge using AI properly or an agency helping you with that and that knows how to use AI properly. So you do need that. And the question nowadays is not whether you use AI or not. The question is how you use it and if you understand what you are building. Because all the tools, all the pieces of the puzzle, all the building blocks in software didn't go anywhere. You still need to know how they work and how they will behave when put together this way or another. You can consider that as an investment into something that will work out eventually. So you try this prototype, you try that prototype, and something clicks with your users, something solves their pain or their problem, and then you've got yourself a Nice product idea. Like prototyping got easier and faster. If you misunderstood the needs of your target audience, then you'll be solving the wrong problem, no matter with AI or without. It's just that you fail faster because AI is the multiplier and you fail faster. You understand it doesn't, it didn't work out and you move on to the next idea. As easy as that.

Speaking of misunderstandings then, why would you say founders consistently misunderstand about software teams, which is where you come from, especially when all they have known is a vibe coding tool?

What they misunderstand about development teams, founders. It's very difficult to generalize all the founders, you know.

From, from the ones you've worked with?

Well, I think that's the common misperception that a development team can build anything. They can't. And a good team is the one that pushes back on stuff. So that says that this is not very realistic, or this can be done this way and suggests another way, or that keeps you in the loop and honest about the deadlines and delivery timeframes. So that's a matter of communication mostly.

Now, Konstantin, just before we started recording this, you've told me that you've never taken outside funding or an investor as a founder yourself for your software company. So I'm wondering whether bootstrapping has made you a better operator or just a more cautious one.

Well, the problem is that I have never worked with external funding on any project. Maybe I missed on something by doing so, but I actually like the principle where you grow organically and you don't buy what you can't afford. So this has been my main philosophy. principles, and that's been keeping me afloat and growing for the past 21 years.

So why won't you then say hire ahead of demand when so many businesses do?

No, we don't do that exactly because of that principle where we don't buy what we cannot afford. So we scale as any service business, I presume, we scale together with the customers, with the demand, with new customers, and we hire Accordingly, whenever we don't need a full-time hire, we hire a contractor. And also growing organically, in my opinion, makes you care more about your reputation. The overall culture of caring about customer satisfaction is crucial because you're growing with that. So you're growing with recommendations, with satisfied customers, with word of mouth, and you go that extra mile for your customers. If you do so.

Now, about growing organically, one of the things that AI has made it in a way easier and simpler for a lot of companies is to ask, when do I need to hire a new person? How much can I augment and improve my current workforce with AI before I hire a new person? How much more can I squeeze out of a developer, out of a marketing person, or any other role? What has been your experience in your business? How much more productive have they made a— AI has made a software developer?

Well, we consider it a tool, a helpful tool that amplifies your productivity. So you develop using, I don't know, libraries, IDEs, frameworks, external APIs that make you faster. AI makes you faster because it can generate code instead of you and you are not stuck with the boring stuff and you can do more fun stuff. So this is how we've been introducing this in my agencies. We are talking to our people and encouraging them to use AI to be more productive, to produce more for our customers, but we are not squeezing anything out of them because they work the same 8 hours as they used to.

So has that made Your client work cheaper or faster at the same price?

Well, we maintain the same hourly rates. They didn't get cheaper or more expensive much because of AI. Like we are growing with the market, but, but not, not because of the AI. It's certainly made us faster and more efficient and more productive for our customers. So they're getting, I think they're getting more for their money for sure.

And that's in line with other software companies as well, and other agencies of any kind from what I'm hearing right now on Millennial Masters?

Well, you cannot ignore it. There are 2 ways that you can go. You cannot go without AI at all, but you can have discipline with it and build reliable quality software, or you can go berserk and build whatever you don't understand. So, so we prefer to keep that discipline that we carry from the pre-AI era. With us into the modern day.

Coming back to that, if every company can build faster, every agency can develop quicker, what actually becomes the competitive advantage when everybody has the access to the same tools? It's not necessarily about the people anymore. So AI is augmenting everybody or leveling up everybody to a certain point.

Well, the, I guess the priority shifted to understanding your market. Even more so because everyone can ship a product, but not everyone understands the real pains of their users. So you need to be either feeling the pain yourself and filling that gap in your industry or know how to address and talk to your users and do proper customer development cycles with them before you build something.

And what would you say is a proper customer development cycle? When everybody can move so quick?

Well, the proper customer development cycle is when you run studies, surveys, and also personal interviews with your users, asking them what takes most of their time, what they find tedious, basically what eats most of their time and where do they feel stuck. And if you can speed it up or completely relieve them of that work, here you go. You got a product-market fit.

When AI levels the playing field technically, everybody can use the same tool. Everybody can have access to the same resources. What's the next gap and what's becoming maybe harder to copy? Because you can tell any vibe coding tool to make me the Airbnb for whatever sub-niche, like pet lovers and people who travel with their pets.

Have you tried? Building an Airbnb for pet lovers?

Not personally, but it has been given as an example. So I'm using that one. And you just spoke about the example for a fitness app for LGBTQ people in New York. So a niche within a niche.

Yeah, absolutely. So you need to know, again, your audience, you need to know how to make them happy. And you also know, you have to know how to get in front of them. Because the competition for attention is not going anywhere. Maybe even—

Okay. Now we're getting into another serious topic. Let's do that. Yep.

Yeah. Maybe it becomes even more harsh. You need to know how to reach them and you need to know what they need.

Yeah. Building features ahead for a version of the product or a size of users that you don't already have, because you can build this great technical infrastructure that would support 100,000 concurrent users when at the moment you will only have 1,000 or 100. You can overbuild and spend a lot of money at the beginning before you actually need that capacity.

Of course, of course. I'm a big fan of solving problems as they appear. So you definitely shouldn't shoot for like much into the future, but Come on. We also need to recognize that the majority of entrepreneurs are like pathological optimists and they—

You have to be to be an entrepreneur.

Otherwise you will not start, right?

Exactly.

You're not gonna start any product or any business for sure if you're not optimistic. So I think this kind of thinking is making people build for the future and I cannot blame them for that.

So just zooming out now, Konstantin, to a bit of a big picture thinking. If you were starting a brand new software company tomorrow, Why would you build completely different because AI exists versus from when you started your company?

What I would build different? Man, it's been 21 years.

And that's exactly why I'm challenging you with this question.

What I would do different? I would still prioritize quality over speed.

Thanks again to Data4SEO for sponsoring this episode of Millennial Masters. If you want to see how your own business or content is showing up in AI search, use my Data4SEO link in the description and you'll get $5 in credits to try it yourself. As a final question for you then, Konstantin, because AI has been a thread throughout what we've been talking about today, what's something that everyone is excited about in AI today that they'll stop talking about it in a couple of years?

Well, obviously that AI must be anywhere. So let me tell you something about hypes. So I have seen hypes before and also in my business, for example, mobile technologies. Like when the first smartphones emerged, everyone said that, like many, many businesses thought that they must be there. They must have a mobile app for whatever they sold, for whatever industry they had. But the people and the ecosystems were not ready for that.

Yeah.

So we got tons of crap and apps that no one used, right? Then came the cloud, right? So instead of self-hosted solutions, no matter how stable or good they were, like businesses started moving them into the cloud, which resulted in very expensive hosting bills. And with AI, we are now witnessing the same. That many businesses, they apply AI just for the sake of it. And then, for example, they, I don't know, rehired their human employees. Like, who was that? There were a few examples like Klarna, IBM, at least in the support, because support seems logical for AI, but then somehow they realized they made a mistake and started rehiring those people. So with AI, You should also start with the problem, not with the solution. So it's not a solution looking for a problem. It's the problem that you find the solution for. If AI fits that role, that description, great. You should apply it there. But if it can be solved with the usual deterministic workflow, then there is no need for AI there. It's just— it will just make things more confusing and complex and more difficult to maintain. That's it.

I think that's a good lesson to wrap up our chat today, Konstantin. AI is not the solution for everything and shouldn't be the solution for everything, even if you just want to put a sticker, say, on your product that it's powered by AI. So I think that will be one of the main takeaways from our chat today. Thank Thank you so much for joining me on Millennial Masters and trying to look into a crystal ball and see where things are going to go next.

Thank you, Daniel. Thanks for having me.

Thanks for tuning in all the way to the end of this episode. If you enjoyed it, please share it with someone you think would get value from it. And don't forget to subscribe on YouTube, Spotify, or Apple Podcasts so you never miss an episode. For the best way to stay in the loop, head to millennialmasters.net. That's where you'll get all the latest features and new episodes straight to your inbox every week. Catch you next week.