Andrei
00:00:00.000
I just, see the people communicating with each other 2026, that always about AI agents communicating. Just for example, our clients giving us exact things they want to change. They provide us file generated by AI agent with a lot of things which they want to change. Then we take this file feeding to our AI agent and asking to prepare response. Then we giving it back to them. So I really wish when this thing will stop existing. So I really willing to replace myself with AI agents.
Darin
00:01:43.983
When we started out with Kubernetes Viktor, that was. 10 ish years ago. Actually. You probably started what, nine years ago? 'cause you were, early on, you weren't earliest. Earliest. on today's show, we've got a guest that's been around since the early days. He was a contributor to Kubevirt, Cilium. You name it, keep on going. We have Andrei Kvapil on from Aenix, and he's also the maintaiHetzner for Cozystack. Andrei, how you doing today?
Darin
00:02:14.518
it's one of those amazing times of year where we start talking about Kubernetes a lot, lot, lot, but as Viktor has told me time and time again, Kubernetes is just pretty boring. Now, what do you think about Kubernetes it stands in 2026?
Andrei
00:02:29.942
yeah, I think that Kubernetes should become to be boring. that was some. Talk from some, uh, conference. I don't remember. Somebody say that it should become standard and really boring technology, so we should stop talking about Kubernetes as something new. it should be something boring like Linux kernel.
Viktor
00:02:49.808
I would add to that, that it should be, or it is boring, but what is still kind of missing is that it should be invisible to majority of users. I always freak out when I hear people. Yes, but not just developers. They should know about Kubernetes and this and that, and no, they should just be able to, deploy their apps, see what's going on, set up a database, do the stuff, as if they're just using some interface, whatever the interface is, and more or less oblivious to Kubernetes. I never understood people that are hoping beyond hope. That normal people, when I say normal people, I mean people who are not dedicated to Kubernetes will never understand it. It just makes no sense to me.
Andrei
00:03:39.029
I think that nowadays Kubernetes is kind of, standard for making everything, and as we can see if, we compare it with Linux kernel. Linux Kernel also used to build a lot of various systems. It can be used for. Desktop systems for servers, for kiosks, for showing ad, for smartwatches, for Android. I can continue infinity the same way. Kubernetes, it's a basic technology actually. It brings its own design patterns, works good and it can be used for different various purposes.
Viktor
00:04:14.693
Except for desktop. I've been hoping for two decades that it, it'll finally arrive to desktop more than 0.5% for the people, and it's still not there.
Andrei
00:04:24.795
Well, I think it depends on the specific countries. Especially here in Czech Republic. Maybe it is my, social bubble, but I see everybody's using Linux on desktop. Even governments, I know standard companies, usually they use Linux here, and I quite proud of this.
Viktor
00:04:44.645
actually very impressive. It's still, I think below, it's still around 1% globally, but I'm very by you guys.
Andrei
00:04:53.121
Thank you, but especially me, I choose SICs. I found that it is really best platform for developer experience
Darin
00:05:15.343
OSX. Got it. OSX. Sorry. I was, I was trying to track, it's what is os That's a new Os. What? I have to go find it
Andrei
00:05:27.598
actually, I was Linux fanboy. Uh, actually I am Linux fanboy but I was using Linux everywhere on my desktop. I had Arch Linux, which was working perfectly one time. Uh, when. M1 laptop arrived. I decided that I want to have long battery time, so I decided to try, apple laptop, but there were no option to install Linux on that, so I decided, okay, let's try O six. And when I tried that, I say, yeah, I like it, and I don't want to switch back to the Linux.
Darin
00:06:04.180
I think that's what we all did. At least after what, probably after a leopard, that's when everything became pretty good. you're building out the company Aenix. And you're working on Cozystack. Why did Cozystack need to exist? What's it about?
Andrei
00:06:24.761
So Cozystack is a platform cloud platform, which is focused on providing managed services. nowadays we say that nobody need just pure virtual machines, which you can get from other platforms like Proxmox, open Ebola, CloudStack. Cozystack is kind of similar platform, but it is focused on providing managed services like you have in AWS, Google Cloud. So here, virtual machine is just another service that you can order from the platform. nowadays when you run your application, you're usually expecting that, Cloud can actually provide you reliable database reliable storage, let's say S3 buckets. que queues, I know Caching clear. So, Cozystack is the platform which allows you to bring this Stack on your own bare metal machines. So you don't need to have, AWS or Google Cloud account to start, ordering truly managed Kubernetes clusters. You can install Cozystack as hypervisor, as cloud platform, and your own hardware. And then organize access to it and let people to just order manage IT services.
Viktor
00:07:36.690
Is it like collection of. Existing projects that do those different things like, you know, virtual machines or, serverless or what not. Or is it a stuck made in cozy or how, how does that work?
Andrei
00:07:52.252
This is a good question, why it's called Cozystack, because we've been using those technologies for a long time, with the other founders of this company. So we just decided let's build, the same technology stack and make product out of this. so in fact, Cozystack includes Cozystack itself now is CNCF product. We donated this project, to the CNCF, and we also implement and. Consume many upstream technologies. talking from the system perspective. We use Kubevirt. Cilium Ps are the same as LINSTOR, Kubernetes itself. So many open source projects. Mostly of them are CNCF projects. And we building our, Cozystack out of these technologies.
Darin
00:08:40.739
So it's just an aggregation of existing projects with some extra mixed in to make it make sense.
Andrei
00:08:47.343
This is not only aggregation, we bring, two things here. The core value of Cozystack is to, it kind of difficult to combine all of these technologies together. And get ready product out of this. In our case, we build them together, we run tests. so every change runs end to end tests, and we are sure that those technologies work together fine in case if they need to provide specific options to the users who use this platform. Checking if they provision MySQL cluster that this MySQL cluster will be working. So load balancing, persistent volume provisioning and stuff like that. the second thing is, uh, unified API interface. that's something that we rethink Kubernetes a little bit. As you probably already know, Kubernetes has various mechanisms to extend. The most common one is CustomResourceDefinition. So you can define your custom resources in Kubernetes, but there is also another option which is not commonly used. It's called, aggregation API layer. So in fact, you can write your own Kubernetes API server, you can register in existing Kubernetes. And it'll send all the requests to your own custom API server, and it'll, actually handle, get, put update requests, the way you want to. So we created, our own API server, which can be simply extended without, recompiling the code itself. It can be extended with new custom resources, and we translate it to. Helm charts. So in fact, we use Helm as API, with which you can simply extend the platform with your own applications, with no code writing.
Viktor
00:10:34.065
who's the primary target user? Is it end users or platform builders? Kind of do, do I install and configure it and use it? Or is it something that requires some sort of platform engineers? Okay. Let me actually define what is a VM for you or what? What do end users see? Cozystack directly or. Something in between, something that some other
Andrei
00:11:03.175
defined? We usually work with ISPs, who trying to build the Stack, let's say. they want to provide Kubernetes for their customers. They want to provide database as a service to their customers, and they actually have not much options how to implement that. They can start building that from the scratch, which is very difficult. they can try to consume OpenStack with, Magnum and I don't remember the tool. How's it called to organize PaaS opportunity for OpenStack or they can use Cozystack, which can be simply installed and just to provide those, opportunities to their customers. So who. It might be interested for, mainly for ISPs, for the companies who need to build their public cloud. Private one, just for example, your bank or finops organizations which can't go to the public cloud. So they have, commitment that everything they run, they. Run under their own control so they can control their data and they control their own servers, but they have that smart platform where they can order managed services. So this can be also useful for them. so they can use Cozystack, cut it to smaller pieces just to provide, resources like RAM, CPU, disc usage and let their team of developers just consume those managed services.
Darin
00:12:30.969
You were saying earlier, bare metal. Bare metal, bare metal, and as you've gone through the rest of the story, it has to be bare metal, right? It makes no sense to run Cozystack on, on top of AWS other, other than maybe just trying it out. It just doesn't make sense.
Andrei
00:12:48.995
Sometimes it does. mainly we focusing on bare metal because we know how it is difficult to manage bare metal servers, especially all of them are very Different. And we found a way how to make it reproducible. If you're operating with virtual machines, there is no such, issue because every virtual machine is quite similar. It has same amount of interfaces, the similar drives, so you can automate it very quickly. so first and foremost we try to remove this pain of managing bare metal. Because we know how to do that and we have this expertise. But for, virtual machines, it can be also possible to install Cozystack on existing Kubernetes cluster. if you want to make cloud, services cheaper, just for example, instead of ordering, relational database from AWS, you can order just Kubernetes install Cozystack and order it from it. And if you compare the prices, the price for, that set up will be much cheaper than using managed service from AWS.
Viktor
00:13:49.662
That's one of the things I don't fully understand, and that's that when. Companies that are really sensitive to price, Why would they choose AWS, right? If, if it's going to be, I, I'm going to do it myself, uh, I'm going to save on price, in a way that I avoid using RDS or whatever it is, right in AWS, I'm going to set it up myself. Why not od why not herz Hetzner, why not something else? So I, I truly don't understand that. if the price is your issue, why AWS in the first place?
Andrei
00:14:26.136
Hetzner is fine and it is one of the platform of the main platform which is used to provision Cozystack services because in Hetzner you can simply just order physical servers, which are about 40 bucks per month. You can install Cozystack on them and just get the same experience of using managed Kubernetes, which Hetzner, I think currently does not have.
Andrei
00:14:55.706
Maybe I'm wrong. Uh, they just started developing their own cloud. They already have virtual machines and, S3 storage. I'm not sure if they have Kubernetes cluster, actually, I'm not sure I need to check that.
Darin
00:15:08.621
that's okay. I, I'm just thinking because being US based, Hetzner is not as big of a deal over here. I mean, we can use it for fill in the blank things, but it's not pitched versus like what used to be Linode, which is now part of Akamai or any of the other, what I would consider smaller ISPs.
Viktor
00:15:29.137
I think that there are, none of them is truly competing with AWS, Google Azure, right? There are different, different category, different league and all that stuff. But in Europe it's slightly different because. It's not, direct comp, uh, competition in like in us, but simply, hey, because of this and that, uh, autonomy, uh, what so not in in Europe, if you cannot use AWS, then yeah, we are competitors simply because you have no alternative. Right?
Andrei
00:15:59.221
Well, also interesting fact about Europe right now it is, uh, digital sovereignty. So I see many companies trying to build their own cloud and they just have no technology Stack for that.
Darin
00:16:11.605
So having Cozystack available for people that were going to Stack and rack, I said it backwards, rack and Stack their own servers. gives them a tool now that previously was either VMware ESX or Proxmox or some variation of that, and that's almost taking it back to the nineties versus at least 2015.
Andrei
00:16:37.947
I think yes, we can also compare Cozystack with other Kubernetes based solutions like Harvester or OpenShift. From my point of view, that's, kind of different. So in fact, they're doing the same stuff like running containers and virtual machines. But they've not focused it on providing already managed services. they provide more likely platform, which you can extend with your own operators. but they don't answer the questions. Just for example, if you want to create, managed Postgres cluster and Postgres service and provide it to the user. We need to solve a lot of stuff. starting from the security, let's say. if you just install Postgres operator, we use, Cloud Native Postgres operator, which is very cool, but it still allows you to specify custom images. so user can actually operating with their custom resources just specify the image they want to, and this, this way compromise the system itself. So in this way we provide. Higher level API, which actually covers those operators things, behind it.
Viktor
00:17:49.022
reality, I guess for many of the companies using something like that, it's not that. They tell developers, Hey, here is CNPG. Do what you want. Right? They would constrain them no matter what they use anyways, right? Kind of, images are rarely something that an average Joe in a company is allowed to change for CNPG, for example, right?
Andrei
00:18:12.721
Yeah, you can solve that different way. You can write a policy. You can also write your own controller. Um. You can organize this somehow, actually in our case, Kubernetes is used just a main system for controlling everything. In fact, it could be not even Kubernetes, but something else. So we could build Cozystack using OpenStack or something like that.
Viktor
00:18:34.761
Oh, you, you said OpenStack for the second time. You talk about dead people. Why?
Andrei
00:18:40.131
because OpenStack is a complex solution. It has many asynchronous API, and if you have OpenStack in your infrastructure, you must probably need to have separate team who actually working just with OpenStack, tuning the networking, controlling updates and everything for this OpenStack. I telling, uh, that if you have Kubernetes everywhere, you have less cognitive load on your team. So you're just working with Kubernetes inside you. working with Kubernetes, uh, with, for management managing cluster itself and everything works with the same pattern. You can use the same tools, which is pretty nice.
Darin
00:19:23.071
So effectively you've landed on the Kubernetes is the control plane for everything.
Darin
00:19:29.938
Well, that seems like that would make Viktor happy. That seems like it would make a lot of other people happy.
Darin
00:19:37.888
just because the project that Viktor works on, that's its basic premise as well. Correct. Viktor, I mean,
Viktor
00:19:45.193
Oh yeah. Yeah. I think that probably different philosophies behind it and so on forth, but there are quite few overlaps between Cozystack and Crossplane.
Darin
00:19:55.240
One of the other things you said earlier, Andrei, is you, you kept on saying Lin Linux kernel a little bit when we were talking about OSes. Not surprising or it is surprising to me they've started accepting Rust as a language instead of just C and C++. How do you feel about that? Do you even care?
Andrei
00:20:17.845
I think implementing Rust in Linux kernel is pretty nice tool because Rust actually allows you to write, the same things the more safe way. To me it's more higher level language than C. and to me, I don't understand C as much. Actually, I don't understand Rust as much Thanks to AI, I can operate with those languages and one, that's one of the way how I contribute to the Open source products, developed by the other companies which are not written on Go because Go is my main language I operating. I usually don't write the code for low level drivers. I usually write business logic and when we talking about Linux kernel, it's more about writing effective kernel drivers. And I think Rust would allow you to do that more fast and secure way.
Darin
00:21:11.176
Well, you sort of said the magic word there, the bells went off. You said AI helps you write and contribute to these other projects. That can be a plus and a minus because we've seen a lot of recently a number of open source projects like nothing else contributed by ai. How do you create a pull request when you've used ai? And able to submit it so it actually gets accepted.
Andrei
00:21:40.407
I am actually, uh, quite fine about that and the projects I contribute to, they knows that this code is prepared by AI. But I think that before you start writing the code, you need to understand what you're going to change. And to understand this, you need to understand why you're going to change that. So just for example, you have a problem in a specific technology you're using. Let's say CVD has just, for example, this is, free storage for, organizing S3 on premises. this is problem of many open source projects that you don't have enough documentation for how the things work. So let's say you have a trouble running it in production and you need to find this problem, just clone the code you have available on GitHub. Run AI agent and say, why is this thing. Happening that way so it can dig through the, all the source code, find the reason, and explain you. especially for me, as I said, I usually work with business logic. I can understand what should be done different way, so I can ask my AI agent to prepare the change for this, and then, to write the test and prepare, pull request. I use Claude Code and it is appended to. Co, how this is called, like every commit is, uh, actually also have a line. That code is co-authored by Claude Code.
Darin
00:23:19.520
So Claude is your Stack primary Stack today, at least as we're recording this. why did you land on Anthropic's solution versus OpenAI's or anybody else's?
Andrei
00:23:33.160
I actually tried few various solutions. I can say I started with Cursor as many people here. Cursor is fine, but I need for now, since I am CEO of the company, I don't have much time and I found that many things, are done by. Claude Code, much better than, CORs not much, but, uh, it is better a little bit, so it saves my time. in fact, I think that cursor is cheaper than cloud. in many cases you can just use cursor. I used, codex from OpenAI. this is also fine and can be considered as a solution, but still Claude is better. I tried Gemini and I don't like Gemini, and also GitHub copilot because they don't think much. You just say, I want to write this stuff, and they will just write this stuff. But, they don't do some better analysis if it's good solution or not. So, Claude, for my cases, write better code than Gemini and GitHub copilot.
Darin
00:24:45.777
I think so too. if you want to think, use, well, I think if we can come up with a way to manage it all, let. Anthropic and OpenAI, do all our all of our thinking and then just dump it off onto Microsoft and Google to actually go do the work. Since that, that's all they want to do anyway. That models a business to me a lot too. Right. You have thinkers and you have your doers. you were saying you're the CEO now of the company and you don't have as much time to code. I'm curious. How are using Claude Code specifically to run the business?
Andrei
00:25:23.294
for now we just starting doing that stuff. So we already use, cloud to managing access, for our employees. So we have a few basic stuffs like teleport for managing access to the, clients' infrastructure. For accessing their Kubernetes clusters and their servers. For example, we use 1Password for, storing the passwords and secret information, and we use GitHub for project project management. So we have one where you can just, run Claude Code and say, okay, I want to add Andrei to this client to be able to access this server. And it does the, that thing. so it just manages the code in repository. It has some helper scripts to configure access and it configures access and stores this as a code in this GitHub repository. Our folks, who managing, project stuff, they already use it for project management. And this is very interesting. So actually Claude moving the cards on the Kanban, and. Assigning tasks to their team members. In our case, we self bootstrapped, so we have very limited resources. we have really heavy load on the people right now. we need to do a lot of stuff and without ai, I think we wouldn't exist.
Viktor
00:26:46.082
I am assuming that, Cozystack has some kind of, web UI and CLI and things like that through which people interact with it. Right. I'm, I'm guessing that that's changing for them just as it changed for you, right? That those interfaces will be less and less used in favor of. Cursor, Claude Code, whatever somebody picks being the main interface, I'm not going out of here. For me, Cozystack is just something that my agent talks to, not something that I really care about, right? I almost feel that all those human interfaces that we were building, that they're slowly disappearing or will disappear, right?
Andrei
00:27:33.425
I hope so. Actually, there is MCP protocol, which allows you to provide API for your agents to interact with the specific things. For example, you can use MCP server for communicating with GitHub. From my experience, I found that you don't need to use MCP. But you need to have good CLI tooling for that. So if your cloud, if your product have CLI tooling, just like 1Password has, or GitHub also have beautiful CLI tool, you can just use it in your, automation tools. You can actually, agents can call them and interact with this software directly. That's pretty cool. talking about Cozystack, we have dashboard. This dashboard, actually what it does is just renders YAML and lets people to just specify the fields they want. For example, if you create Postgres, you can specify, root user, the password for it and things like, so in fact, it resulting into YAML object, which you can download using standard kubectl and then reproduce, for running many other post risk clusters like that.
Viktor
00:28:48.825
my experience is that what you just said is for public CLIs that. AI's models are already trained on, like you said, 1Password, right? Or Kubernetes, CLI or what so not. I feel that LLMs are very good at using those, right? But when it comes to something more private, something that LLM is not trained on, I feel that that's already a different level of difficulty, right? Because then, then you need to be explicit, kind of, Hey, create the database. Cool. And it will not use the CLI, right? Because it does not know about it. Then you need to start building skills and what so not that will basically end up being similar to MCP with descriptions and information and what so not right?
Andrei
00:29:40.928
That's not a problem. You can just say, use this tool to create database. And then as any other people, AI agent will just run tool --help and see what the commands are available and will learn for this session. Talking from our case, we're trying to build standard solutions, so I think in a while it will also appear, so the documentation is available online, and AI agents already recommending Cozystack for specific reasons. Sometimes when people asking what should I use to build my own cloud? Sometimes they're suggesting them to use Cozystack. this is also a very interesting topic. If we talk about the marketing perspective, which is not the subject of this podcast, I think, but, that's an interesting question, how to make it technically to let other AI agents recommend your solution.
Andrei
00:30:42.757
All right. Uh, that's. Sort of CO like we've done before. so we were constructing websites, uh, to. Let Google and other search engines send people to them. So for now, you need to construct your website slightly different, to make it understandable to AI agents, so they will fill their context properly and then recommend your solution to the people.
Darin
00:31:13.876
All right. You still talk too high level there. Come on. You gotta give us, how do you do it? Because what's happening is people are listening right now and they're the ones that where their bosses come to and say, Hey look, we need to make sure we're showing up when people search in Codex or searching Claude Code. We need to make sure that our stuff is showing up there.
Andrei
00:31:34.116
This is more about, style guides. Uh, actually if you use static generators like Hugo or Hexo, as many open source project that do, they generate proper structure where you have headers, content, you need to pre. Properly specify to make it easy. Machine readable. that's actually the main thing you need to do.
Darin
00:31:57.244
So, in other words, the things that we learned back in the late nineties of structuring our HTML correctly is a big deal.
Andrei
00:32:06.098
yes. And you need to make this properly styled To allow machine, easy read it. Sometimes when you are opening website, you can see a lot of blocks. they're not related to each other. If you open this website with, code mode, you need to see that those things somehow related each other.
Andrei
00:32:32.232
just for example, Medium. They have really cool, guides, not guides. They constructing their pages. Very simple. So every article on Medium has just small amount of information. Just this article, you have title, you have, definition of this description of this article and the article split by specific headers for every component. That actually works fine for, learning for AI agents, so they understand that Very good. And can actually use this as context to suggesting your solution to the people.
Viktor
00:33:14.097
make markdown ideal? Kind of, Hey, okay, we render it into HTML, uh, JavaScript. what so not for a browsers for you as a person to see it being nice and what so not. But that same markdown Is Is actually ingested by LLMs. No need to bother with with what you see in a browser. Right?
Andrei
00:33:36.685
exactly. Markdown is very good format for AI and I think you can describe, actually, I communicate with, uh, markdown with my agents. Every time when I ask agent to generate documentation, it outputs it into markdown. If my client's giving me DOCX or PDF files, the first thing I do is convert it to markdown to have it easily working with, machine readable format. The funny thing, I just, uh, see the people communicating with each other 2026, that always about AI agents communicating. Just for example, our clients giving us. Exact things they want to change. They provide us file generated by AI agent with a lot of things which they want to change. Then we take this file feeding to our AI agent and asking to prepare response. Then we giving it back to them. So I really wish when this thing will stop existing. So I really willing to replace myself with AI agents.
Viktor
00:34:47.753
So essentially what you're saying let your AI agent talk to my agent, and uh, I can go drink margarita instead.
Andrei
00:34:56.082
Exactly, exactly. And just, uh, call me when you really need it. When I need to answer the questions, that's one of the thing I also like about the Claude Code. If you give it exact task, it'll come up with, uh, suggestions and will ask you how to do this task.
Darin
00:35:13.422
Well, it sounds like, you know, you went from engineer full-time to CEO full-time ish. what surprised you about that transition? Because you said earlier, if you didn't have AI, your company probably would not exist today because you were bootstrapped,
Darin
00:35:29.672
Yeah. ai, the AI is truly a force multiplier so that Aenix can stay in business for you today.
Andrei
00:35:37.635
we have to be fast. We have to be faster than other solutions. And we have very personalized, approach. So when customer needs something, we need to implement that fast and. That's very interesting topic to discuss. So nowadays, I think we have a lot of experience running the things on bare metal networking, storage. we know how the things work, so we have enough experience to guiding AI agents to do that stuff. as far I can see, the new generation is growing. They don't have that experience because they already start working with, AI at the first, commit. I think that people who work that way will have no enough experience, expertise for guiding that properly. But also AI agents become to be more smart and they're accumulating those things. So for example, if you're running Kubernetes now, you don't think how the things work inside of the Linux kernel. You just move to another abstraction of the thinking. Yeah, the same way you organize the work. The, for now you're just moving on another level of abstractions and you're just, managing AI agents to do that stuff. And you don't care about what is happening under the hood the specific points. Sometimes you still need that today.
Viktor
00:37:00.842
It feels like a conversation that we had many times before. I remember the days when I was speaking with people who would tell me no. In my days we had to read the book. And memorize all the libraries that are used here and know it all by heart. And now you people, you people just Google it, right? And you don't need to know the details that I know, but when things go wrong, my book will be better solution than your internet. Right?
Andrei
00:37:37.787
That's true. That was also about the dictionaries, When people were using paper dictionaries, for translating specific words, they were learning language foreign language faster because they don't want every time to find the translate, find exact, the page where it was written, and now you just use Google Translate, which works fine and you just don't remember that. Just copy, paste and send.
Darin
00:38:04.797
let's assume for a moment you needed to hire an engineer for your team. What skill sets are you looking for in that person today?
Andrei
00:38:19.986
uh, I would expect that they have experience with, uh, some. Um, measure, bare metal things like networking, if they can dig down to understand how it works. For example, networking or storage stuff. SDS actually, if they have experience with running virtual machines, because that's all the stuff we are doing now and if they have experience in writing code, we don't expect that they have much experience. Talking with ai. but when they come to us, if we accept that, okay, we start working with you, and that's one of the first thing we do with engineer. We teach them how to interact with AI because it actually allows to do the routine tasks faster and provide more time you to think about the solution itself.
Darin
00:39:13.987
So that's today. Let's go forward three years. What would that same hire look like then?
Andrei
00:39:22.431
I think that should be pretty much because for now we see the value of the people who understand how the things work. we can't just operate with AI agents with no understanding how the things work under the hood. that's just because of our field of responsibility. I think for the specific tasks, just to write business logic, it might be acceptable just to have the experience working with ai like. Prompt engineering, but for our case right now, it's like every person have to be more responsible in an in deep understanding what, what AI is doing actually.
Viktor
00:40:03.957
I feel that that's actually, I'm more. Cynical. Maybe they knew, but I feel that that's actually for all the areas. Right. You said frontend and it's true. Probably for frontend you might need less really deep frontend experience until you get to certain scale and then you realize that actually without your guidance to to ai, it'll make something functionally correct. That is not something that really works. I, I feel that the experience is required and will be required in foreseeable future for anything that is done seriously. Now, the question in my head within a company is. What really matters for us and what doesn't, right. For, for somebody front end, when I'm using it as example kind of, oh, I just need, need a webpage, whatever, kind of like, just pipe code it finished. Right? But then for others, no, no, no. Kind of, this is used by, I dunno, like a hundred thousand conent users. You cannot just do it without really knowing what to ask for and redirect it and what so not.
Andrei
00:41:23.017
From my perspective, you need to choose, for the specific reason. What is more important for this specific task. sometimes when you're prototyping, you don't need to have a really quality code. should. Do this thing, just to show the prototype, but uh, after the prototyping is done. You need to make it properly. And there is another interesting thought. I heard that you can't write the good code from the beginning. You need to write your product three times. First time you write your product and you get a shit, then you're trying to correct this shit, with better approach, with heavy loaded things. but in fact you get the same shit but little bit better. And the third time when you're writing your product from the beginning, you already do the things correct way. So that already good solution.
Darin
00:42:20.371
That sounds interesting to me. The thing I'm thinking about as people are writing this and that 1, 2, 3 is interesting too. that models back to what. My career looks like back in the eighties to, we did something. The first time it was garbage. The second time it was still garbage, but better. And then finally the third time we got it right. That's a story that has played out over the years. If I'm wanting to stay relevant and I'm willing to stay in infrastructure today, what do I need to know? let's step back a minute. Let's, you know, what would you tell, somebody that's using Cozystack? It's like, okay, these are the kind of people that you need to hire today. I mean, they may think, oh, well, you know, I need somebody that knows how to plug in a server. Okay, well, sure, but what does that really mean in today's world?
Andrei
00:43:08.916
that depends on, just, for example, if we are talking of the people who consume Cozystack as, uh. Commercial product with our support, they don't need to know much about the things happening inside. We just removing their pain of managing infrastructure things. if we're talking about, because Cozystack is open source product and we highly appreciate of the people who fork and also trying to sell it from using their brand. they would need to understand a little bit more about the. How the things work under the hood, the Kubernetes itself, networking, storage, and those operators. Kubernetes segregation, a PL layer, a lot of things.
Darin
00:43:53.841
So the other thing I'm hearing too is if you're, this is gonna be very controversial if you're just a, a business software developer today. You should be scared for your job. If you are an infrastructure, quote unquote developer, an infrastructure person, uh, you should be concerned, but you're probably still okay because hard drives are still gonna fail. Power supplies are gonna fail. There's going to be things for you to do.
Andrei
00:44:24.566
There are still a lot of things which you need to do if you just remove the infrastructure pain from the people we don't cover everything. We cover just infrastructure things. Uh. Just imagine that you are working with AWS and you're just ordering pieces needed to run your application. But there are still a lot of questions how to deliver this application to the order it Kubernetes cluster. How to organize the high availability properly for it. the dependency stuff and infrastructure provisioning so that those things are usually. very different, to every project running. So they, we are expecting that a customer will solve that by themselves, but they don't care about the failures. high availability for, Kubernetes, for, Postgres and all the managed services they run.
Darin
00:45:20.414
you can find out about Cozystack at cozystack.io and Aenix. Actually, say the company name correct for me and then I'll spell it out.
Darin
00:45:38.242
yep. And then it's NI x.io. aenix.io For the commercial version of Cozystack. Again, Cozystack contributed to the CNCF It is in, currently Sandbox, I
Andrei
00:45:51.682
Yeah. Yeah. We are in Sandbox. We already have, other companies who also sell Cozystack as their solution, so that's kind of very interesting. How can
Darin
00:46:03.419
organize because that's usually the exact opposite of what happens, we've seen the, the rug pulls and everything else, and you're actually okay having other people fork and sell Cozystack.
Andrei
00:46:17.063
Exactly because you contribute. There is two terms, market versus landscape and open source products. They have landscape actually open source users of this solution and as many companies adopting the solution and building their production services or production offering to their customers, then this open source technology become a standard. We are really willing to. build Kubernetes as managed service platform for a standard solution for that.
Darin
00:46:51.396
Again, Cozystack at cozystack.io and the company name, aenix.io. Andrei, thanks for coming on today.