Best tech stack for web dev (with AI) in 2026?

Englishالعربية Español 한국어 中文

MMaximilian Schwarzmüller

Computing/SoftwareSmall Business/StartupsInternet Technology

Transcript

00:00:00Thank you.

00:00:30Thank you.

00:01:00There would probably be an interesting YouTube video since I get the question quite a bit.

00:01:07So why not make it a stream and potentially pick up some input and some questions along the way.

00:01:13Now I'll just give it a couple of seconds.

00:01:19Hi, everybody.

00:01:20Thanks for joining.

00:01:23Super excited to have another stream.

00:01:25I think, yeah, last week I had to skip it.

00:01:28In case you don't know, I try to stream every Thursday.

00:01:32But yeah, it doesn't work out every Thursday.

00:01:35But today it does.

00:01:36And yeah, as I just said, trying to do a stream on a topic, which I get asked quite a bit here and there in some way or the other.

00:01:49Which tech stack to choose in 2026 and if it still matters, right?

00:01:55You could also make the argument we have AI.

00:01:57It doesn't matter anymore.

00:01:58So I want to share my perspective and, of course, also hear yours.

00:02:02So, yeah.

00:02:02Hi, everybody.

00:02:03Thanks for all the nice messages, for saying hello.

00:02:06Awesome to have you all here.

00:02:08Thank you so, so much for joining.

00:02:11And with that, I'd say let's dive in, shall we?

00:02:17Now, I think I did a stream somewhat like this last year where I compared Vue and React and Angular.

00:02:26And that's not what I want to do today.

00:02:28It's not about which framework is better and do some nitty-gritty technical comparisons.

00:02:35But instead, it's more about what should you consider in general when you want to build something, no matter if you're doing it with AI or without.

00:02:43And should you use AI is also another interesting question, I guess.

00:02:47Am I using AI?

00:02:48And I just want to share my perspective here.

00:02:51And I also want to be totally transparent.

00:02:53Obviously, it's not like I have used every single technology out there.

00:02:59And I'll also pretty much exclusively focus on the JavaScript TypeScript ecosystem, simply because that's what I'm working with.

00:03:06I can't say much about the Python or the .NET or anything like that world.

00:03:12So, I'll stick to what I know.

00:03:16And I think one big question you could have is, does it even matter?

00:03:21Does it even matter which framework you're using, which database you're using?

00:03:27Because you may be using AI to write the code anyways, right?

00:03:32So, which difference does it make?

00:03:35And I think that's a very critical, important question.

00:03:40Because, yeah, it does not matter for the same reasons it did in the past or not for all the reasons that mattered in the past.

00:03:50But many reasons, I would say, still matter.

00:03:53And the choice, which tech stack you use is important.

00:03:58So, let's maybe start with that.

00:04:00Does it even matter which tech stack you're using?

00:04:05So, yeah, my answer, of course, is a yes.

00:04:09But why?

00:04:10What are the important reasons?

00:04:13And I would say one important thing to consider is that tech stack, of course, is a lot of stuff, right?

00:04:21It's not just which front-end framework are you using.

00:04:24It is also which database are you using.

00:04:27How are you handling authentication?

00:04:30How are you managing your content?

00:04:31Where are you going to host or deploy this?

00:04:36Which AI agent are you using for building this or for assisting you?

00:04:42These are all important questions.

00:04:44But even if we would just stick at the front-end, back-end framework level, which is really not all.

00:04:51I mean, and by the way, I'm talking about web development here, of course.

00:04:55But even there, it's not just about the front-end, back-end framework.

00:05:00It is also about which database are you going to use.

00:05:02And as you'll see, that all will also be influenced by where are you planning to host it.

00:05:08But I'll get back to that.

00:05:09So, yeah, it matters also because not all choices will work everywhere because cost matters and you can make choices that are pretty costly.

00:05:22Yes, scaling also matters or scalability.

00:05:25But there are trade-offs here and that is all what I want to dive into.

00:05:29But even if we ignore that, in the past, so before AI or maybe like in the early days of AI, but mostly before it, what mattered a lot, of course, is stuff like developer experience, right?

00:05:43And certainly for me, it also mattered how much do I simply enjoy working with a certain framework if we stick at the framework level for now.

00:05:53And I'll get back to the database stuff too.

00:05:55But if we stick at the framework level, yeah, of course, it mattered a lot how much do you enjoy the syntax, the ergonomics, that was important.

00:06:07So, we had syntax, code style, ergonomics, that mattered.

00:06:17We also had the ecosystem, right?

00:06:20So, for example, you may have chosen React instead of Vue simply because there are more libraries that help you build stuff, charting libraries, routing, all that stuff and more courses, tutorials, resources to learn it.

00:06:39And I would say these points are a bit less important or changed.

00:06:47Certainly for me, the entire syntax, code style point changed.

00:06:51It's not entirely gone, but it's definitely less important because, like I guess most of you, I would guess, but feel free to drop it into the chat.

00:07:00I want to know how you're working, but many developers, I would guess, are using AI for assistance, at least.

00:07:09Some may be generating all of the code with AI.

00:07:11Some are definitely vibe coding.

00:07:13That's not what I'm referring to here.

00:07:15If you are vibe coding and if you are maybe using a platform like Lovable, right, then the entire tech stack thing does not matter because the tech stack is chosen for you there.

00:07:25I'm really referring to projects you're building on your own, but of course you may be doing that with AI.

00:07:31Still, for me, the syntax and how much I like the code, that has become less important with AI.

00:07:39It's not entirely gone.

00:07:40I still don't feel like working with a language or a library where the code I write looks like garbage to me.

00:07:50For example, I never enjoyed Python that much, not because the code is garbage or Python is garbage, but to me, it never really was pleasant to write Python code.

00:08:01I did it.

00:08:02I know Python.

00:08:03I'm definitely no expert there, but it was never my thing.

00:08:07Now, with AI, depending on the use case, I might not care as much about it.

00:08:11I still don't use Python a lot because I'm no expert there and because, I would argue, knowing your tech stack, knowing the technologies, being able to understand the code you may be generating with help of AI, that still matters and is as important as ever, at least if you care about the product you're building and if you don't want to get totally disconnected from it.

00:08:37Again, you can go down the wipe coding route, that's a different story, and wipe coding can be fine if you're building your own throwaway software, some internal tools, but if you're building a product which you are monetizing, which you are selling, distributing, I believe it's important to understand the code and being able to understand the code.

00:08:55So, for that reason, I still use mostly TypeScript, JavaScript, and that, but the library choices are less influenced for me by how much I like the syntax.

00:09:08Now, the ecosystem, for example, still is important.

00:09:13It's maybe a bit less important how many tutorials there are, but it's not entirely not important because even if you are using AI for coding, having the ability to read up on something or take a course or being able to point the agent at certain blog posts or have the agent find it if it's doing web search for you,

00:09:42So, that matters because, of course, I want to try to use AI and steer AI such that good code is being generated, right?

00:09:51Code that follows common best practices, uses common patterns, and of course, it helps if those patterns and best practices are documented.

00:10:00So, that can be the case in the official documentation, but that's not necessarily the case for all technologies out there.

00:10:07It often is the case in blog posts and so on.

00:10:11So, it still matters, but it's maybe also a bit less important.

00:10:16What definitely has changed for me is the importance of the ecosystem regarding libraries that are available.

00:10:24So, for example, in the past, in like 2020, you may have chosen React because there is a library for everything in React, right?

00:10:34You want to build charts, there are probably 10 or 100 different libraries to choose from.

00:10:40You want to handle form submissions, here you go.

00:10:43There are tons of libraries for that.

00:10:45And no matter what you're thinking about, there are libraries for that.

00:10:48I mean, just think about all the state management libraries we had and have for React.

00:10:54It's a bit less important now, I would say.

00:10:57Also, because of all those supply chain attacks we had, I definitely tried to build more myself, which got easier now with AI.

00:11:10And it got easier now with AI for two main reasons.

00:11:14For one, of course, the speed advantage.

00:11:16In the past, building my own, let's say, charting library would have just not been feasible.

00:11:22Now with AI, it is definitely feasible.

00:11:25And especially if I'm building it for myself, if it's not a library, I open source and distribute.

00:11:30But if it's just a building block in my application, it doesn't need to be able to do everything.

00:11:35It just has to do what I need it for.

00:11:37So that got easier.

00:11:39And with AI, for me, I think it also got easier to explore technologies I don't have a lot of experience in.

00:11:48Let's say, SVG, if I want to build a chart in my application.

00:11:52I'm no SVG expert.

00:11:54I don't need to be.

00:11:55With AI, I can let AI generate the code.

00:11:57And I can definitely look at that and understand it, make sense of it, because that is different from writing it.

00:12:03So here, AI definitely has shifted that for me.

00:12:07And therefore, long story short, I think my point of view on those parts here has changed due to AI.

00:12:16It's not like it does not matter at all, but it definitely changed.

00:12:19Now, these are not the only considerations, of course.

00:12:21Another important consideration regarding all technologies, but especially for frameworks and libraries, is maintenance.

00:12:30Maintenance.

00:12:36You know that when you just can't figure out how it's written.

00:12:41Yeah, sorry.

00:12:42Okay, you, of course, want to definitely use something that is well-maintained.

00:12:48And well-maintained, funnily enough, can mean different things now in the age of AI.

00:12:53Of course, for one, it means what it always meant, that it's frequently updated, patched, kept up to date, right?

00:13:02That matters because that is security relevant.

00:13:05That also means that new features can be used, that performance can maybe improve over time.

00:13:10You want to use something that is well-maintained.

00:13:12Now, with all those supply chain attacks, there also is a downside to frequent updates, for sure.

00:13:18But in general, I don't really want to use something which is not being updated at all anymore.

00:13:25And by the way, I will also get to the questions and comments in the chat.

00:13:28I just want to finish that thought.

00:13:31Now, regarding the maintenance, it's also important for me to, or as I said, for one, I, of course, try to use less libraries.

00:13:46So, that dependency has degrees.

00:13:49But if I am using one, I want it to be well-maintained.

00:13:54I want it to be well-maintained, of course, and that's like a bonus.

00:13:57But ideally, by people I trust, for example, the Vue team, but also the React team, Angular team, has been around for a very long time.

00:14:07I know or I trust that I know or I trust that they know what they're doing and that they're not just aggressively pushing Vype Slop into their AI Slop, wipe-coded AI Slop into their library.

00:14:20Because, of course, that is a danger.

00:14:22You could be using a library these days that is aggressively updated by some maintainer and it's full of wipe-coded AI Slop.

00:14:28I mean, let's be serious, are we all really inspecting the code of the library we're using?

00:14:33Probably not.

00:14:35And that is not necessarily a problem, but, of course, could lead to bloated bundles, performance issues, potentially security issues.

00:14:42So, that's something to consider or I'm considering, but these are important things I want to look for.

00:14:50Now, I'll get back to the rest here.

00:14:52I got many, many more thoughts, but let me quickly also dive into the chat here and pick up some comments here and questions.

00:14:59So, React with Java or Spring Boot, has that still takers or should we learn niche like Golang?

00:15:04So, Java Spring Boot, definitely still super popular, especially also in the enterprise sector.

00:15:09Not something I'm an expert in, so that's just what I read, not what I personally have worked with.

00:15:15But I don't think that is going anywhere right now.

00:15:20And if you – I will get back to which programming language to use, so, yeah, I'll hold the thought.

00:15:27And, yeah, ooh, got a new follower milestone.

00:15:32Awesome.

00:15:34Z-sharp, Blazor, MAF, .NET, Maui, yeah.

00:15:40I mean, there are many amazing stacks.

00:15:42And, as I mentioned, I'll primarily focus on the TypeScript, JavaScript world because that's just the world I live in, so to say.

00:15:49Max doesn't deserve Udemy.

00:15:51Udemy doesn't deserve Max.

00:15:55Yeah, I mean, I've had very good times with Udemy and still have.

00:16:00The future is a bit unclear, but we'll see where this all goes.

00:16:05Still confused between React and Next.js.

00:16:07Yeah, I'll get back to that, too.

00:16:12Astro.

00:16:13Astro is amazing.

00:16:18AI always tries to wipe code its own functions from scratch rather than using existing helper functions.

00:16:22Yeah, and AI also, depending on the model you're using, at least for me, has a tendency to add lots of pointless helper functions into the code and be super defensive and try to have a fallback for a fallback for a fallback.

00:16:36And that's what I mean with, I want to be able to understand the code.

00:16:39I want to be able to steer AI and get rid of such stuff if it ends up in the code base and not blindly accept everything.

00:16:46So, yeah.

00:16:47So, which frameworks should you use then?

00:16:51And does it matter?

00:16:51Well, it matters because these are the things I'm looking for.

00:16:55But then, of course, it does not matter.

00:16:57And it never mattered in many situations if you're using React, Angular, or Vue.

00:17:03Of course, it will matter if you're working at a company.

00:17:05If they are using Angular, you'll be using Angular.

00:17:08But let's say you're building a little project for yourself.

00:17:12You're self-employed.

00:17:13You're starting your own company.

00:17:15It does not matter which framework you use there.

00:17:18AI definitely has a favorite, and that is React and then also Next.js.

00:17:23And, again, I'll get back to how these are related in a second.

00:17:26But that does not mean that you have to use it because, and that's important, we're not in those early days of AI anymore.

00:17:33Where you had to mostly work with the knowledge built into the AI.

00:17:38Instead, all these new coding assistants like Codex, Cloud Code, Pi, and the models they're using,

00:17:45they are pretty good at, for example, doing web search and visiting web resources.

00:17:50And you can point them at specific resources or just copy in resources, of course.

00:17:54And, therefore, you can, and I have, make AI agents work with brand new libraries that are definitely not in the weights yet.

00:18:05For example, Vercel released a framework for building AI agents, and I played around with that the last few days.

00:18:12I might be creating a course on it, I find it quite interesting.

00:18:15This is definitely not in the AI weights and the model weights, right?

00:18:21The models themselves know nothing about that.

00:18:23But if you point it at the documentation, give them, like, the starting page, give them some hints and steer them.

00:18:29That's your job there, right?

00:18:30They really can learn or understand how to use that framework, and you can make it work.

00:18:36So, just because React is the favorite of AI models does not mean that you have to use it, I would say.

00:18:42I would use whatever you feel comfortable using, because, again, it's important that you understand the code, at least in my opinion.

00:18:50Now, regarding the language you're using, as mentioned, I'm a TypeScript guy.

00:18:56But there, of course, you also have options, right?

00:18:59Not regarding the language, but regarding the runtime, if we're talking about the server-side code, because client-side, it's TypeScript, right?

00:19:07And then compiled away.

00:19:09But on the server-side, you can use Node.js, you can use Bun, of course, you could use Dino.

00:19:15These are the three important ones, I would say.

00:19:18And, again, comes down to what you like, but my personal favorite here would be Bun.

00:19:22And I got a course on that coming up soon, too.

00:19:24By the way, a little plug, if you want to stay informed, you can sign up for our newsletter and or join our Discord community.

00:19:31I share new courses there, of course.

00:19:33And I like Bun a lot because speed, it's really fast, that's great.

00:19:38And it has a bunch of useful stuff built in.

00:19:41And I know that there are definitely quite a few people that don't like Bun because it has so much stuff built in.

00:19:48It's not that pure, you could say.

00:19:52But I enjoy the fact that it, for example, has a built-in SQLite client.

00:19:56I'll get back to SQLite.

00:19:57It's an amazing database.

00:19:58But I like the fact that it's built-in because, again, also because of all those supply chain attacks, reducing the amount of libraries doesn't sound like a bad idea to me.

00:20:10And using a runtime that has stuff built in for which I otherwise would need a library also sounds pretty good.

00:20:18And also because if I'm using more and more libraries, even if I ignore the supply chain stuff, these are all libraries where I have to hope that they are well-maintained, that they work well together.

00:20:30And, yeah, that's why I like using Bun.

00:20:33It doesn't have everything built in I need.

00:20:35And I'm not using everything I could be using.

00:20:38But certain features are quite nice.

00:20:40That, however, does not mean that Node.js, for example, is a bad choice.

00:20:45And, for example, and I created a video about that on my Academy YouTube channel, Node.js has quite a bunch of stuff built in not a lot of people know.

00:20:53For example, it also has a SQLite client.

00:20:56And, again, not a lot of people know that.

00:20:58But in Node.js, you can import Node.js SQLite.

00:21:03And you don't need an extra library for that there, too.

00:21:06So Node.js has come a long way.

00:21:09It even has built-in TypeScript support.

00:21:11A lot of people also don't know that.

00:21:13And ultimately, again, therefore, it's up to you.

00:21:15Bun has that little performance bonus.

00:21:19So that's my choice.

00:21:21But, again, just trade-offs, just something to consider.

00:21:24And you can't make everything work with AI.

00:21:27You just, as always, have to steer it in the right direction.

00:21:33So let's see.

00:21:34Chat.

00:21:44Yeah, Laravel.

00:21:45Just so Laravel in the chat.

00:21:46By the way, also amazing.

00:21:48I have been a Laravel developer many, many years ago, 10 years ago.

00:21:52Haven't really used it since.

00:21:54But it's an amazing framework, which has basically not everything, but almost everything built in.

00:22:01Or for many situations, it has everything built in you may need for building an application.

00:22:09It has authentication built in, queues built in, jobs, scheduling, all that fun stuff.

00:22:13And, by the way, there is an equivalent, kind of, on the JavaScript side, Adonis.js, which I explored in a live stream three weeks ago or so, which is also a framework, a full-stack framework with everything built in you would typically need for building a full-stack or an application in JavaScript.

00:22:33So, yeah, that's interesting, too.

00:22:35And it kind of brings me to the second part of this question here.

00:22:42I talked about React view and what matters there.

00:22:45Of course, what can be confusing and what was confusing in the chat already is you have those meta frameworks there, too, right?

00:22:52You have Next.js, you have TanStackStart, you have Remix, which now is changing and is no longer a React framework in the future, but you had it in the past.

00:23:05React Router is the new Remix to make it even more confusing because you can use React Router in two different modes.

00:23:11In the framework mode, where it then is an alternative to Next.js, or in the data mode, in which case it's not.

00:23:18Super confusing.

00:23:21Ultimately, the thing just is React, Angular, Vue are front-end frameworks.

00:23:27They run in the browser.

00:23:29You build or you use them to build applications that run in the browser.

00:23:33That allows you to build highly interactive web applications, but, of course, pretty much all websites also need a backend, a server.

00:23:40And there you have two main options.

00:23:42You can, or you get a couple of options, but two big options, let's say.

00:23:46You can either build a decoupled backend, a REST API, let's say.

00:23:51And you could do this with BUN or with Node.js, often combined with frameworks like, for example, Express.js or HONO is another great framework I like a lot.

00:24:02And these are amazing frameworks for building backends.

00:24:05And again, it gets more complex here.

00:24:09They're not limited to that.

00:24:10You can also render HTML pages with Express and also with HONO, but these pages then are not React applications.

00:24:18Instead, they are standard HTML pages, which can use client-side JavaScript and where you could also make React work, but it's a bit harder.

00:24:26So, building REST APIs with Express and HONO and so on is super easy.

00:24:32Building full-stack apps, depending on how interactive the client-side should be, can be a bit trickier.

00:24:38That is where Next.js comes into play.

00:24:41For example, Next.js simply makes building full-stack applications, where you have the server-side and an interactive client-side, a React application on the client-side, easier.

00:24:53And it's the same for tan-stack-start or for analog-js for Angular or for Next.js or just Next, as it's called now, for Vue.

00:25:02The idea is always the same.

00:25:04You build full-stack applications, where you have both the server-side and the client-side, but building a highly interactive client-side, like a normal Vue app or a normal React app you may be used to, is simply easier.

00:25:16That is the problem these frameworks solve and why they are so popular, because typically you want both.

00:25:22If you want server-side code, because you need to store user data in a database and that database can't be in the browser, it must be on a central server.

00:25:30But at the same time, you want to be able to build interactive client-side applications.

00:25:35That's where all this complexity is coming from.

00:25:37And it's also coming from the fact that this entire area evolved heavily between 2010 and 2020, something like that.

00:25:47So the internet in 2009 was vastly different regarding how you build websites.

00:25:54And believe me, I've been there compared to how it looked like in 2015, compared to how it looked like in 2020 and how it looks today.

00:26:02Yeah, so that's that.

00:26:04Again, which one should you choose?

00:26:06Up to you.

00:26:06AI definitely has a favorite, and that is Next.js.

00:26:09But as I mentioned before, you can make all of them work.

00:26:11And for example, I built all of my applications right now, like build my graphic, which I built a couple of months ago as a side project.

00:26:18I built them with 10 stack start, because I like that a lot.

00:26:22And you can absolutely make AI work with that too.

00:26:26Again, doesn't replace the fact that you should also know it.

00:26:31Okay, so lots of rambling there.

00:26:33Let me get back to the chat.

00:26:38What do we have here?

00:26:42My opinion on FastAPI.

00:26:45Well, yeah, FastAPI, I know it's very popular and I guess fast, but it has one big disadvantage from my perspective, and that is that it's a Python framework.

00:26:57And I'm just not a Python developer.

00:26:59I know Python, as I mentioned, I'm not, yeah, it's not my main language, and therefore I haven't ever really used FastAPI.

00:27:09I played around with it a bit, but I therefore don't have a good opinion on that.

00:27:13But I'm sure it's great, because I know it's very popular, and it certainly isn't very popular without a good reason.

00:27:20So, yeah, definitely a good choice, I would imagine.

00:27:25Just purchased the Codex course on Udemy.

00:27:27Awesome.

00:27:28I hope you'll get a lot out of that course.

00:27:30Thanks so much for the purchase.

00:27:33What's happening with Udemy?

00:27:34Are you going to stop publishing courses on Udemy?

00:27:36No, I'm not going to stop publishing courses on Udemy.

00:27:40Udemy just has been acquired by Coursera last year already, and the deal is final since like two months or so, or one month, I don't know.

00:27:48And it's unclear how exactly things will evolve there in the future, but right now it continues as always, and I continue as always.

00:27:54But not all my courses are published on Udemy.

00:27:57I do have exclusive courses on my own platform, Academy Pro, and the courses that are on both platforms, the new courses tend to be a bit more extensive on my own platform, a little bonus there.

00:28:08But, yeah, I'm not stopping publishing on Udemy in general.

00:28:13So, I'm not leaving Udemy.

00:28:14No plans about that.

00:28:19What do you think about ASP.net?

00:28:22Also not something I'm an expert in, so not a lot I can share there, sorry.

00:28:26Your Udemy courses helped me more than my computer science degree.

00:28:29Any advice or courses for going from mid-level to senior level?

00:28:33Thank you so much.

00:28:34First of all, super amazing to hear that.

00:28:36It means a lot to me.

00:28:37And regarding going to a senior level, the boring answer is, unfortunately, really only build more and more complex projects.

00:28:46At least that is the answer for me.

00:28:48Resources or tutorials or courses are more scarce there.

00:28:53It's all about gathering that experience by building stuff.

00:28:56As boring and annoying as that answer may be, sorry.

00:29:05So, yeah.

00:29:08By the way, if you feel like I'm missing something in my analysis here or something that absolutely matters, definitely also drop that in the chat, right?

00:29:15So, I want to make this an interactive session and, of course, try to cover different angles and not just my angle.

00:29:26Why am I not over 100K?

00:29:28Well, I have a second YouTube channel, which used to be more about tutorials and still is more about tutorials.

00:29:35And that channel has 900,000 subscribers, the Academy channel.

00:29:40This is my rambling and I share my opinions channel.

00:29:43But it's also growing nicely.

00:29:44So, I hopefully will make the 100K subscribers at some point.

00:29:51These are my thoughts regarding frameworks and how I choose them and if the choice still matters.

00:30:00But, of course, if you're building a web application of any kind, even if it's just an API.

00:30:06Oh, that's also important, by the way.

00:30:10Especially nowadays with AI agents, I feel like full stack web applications are still definitely important and will stay important.

00:30:23I don't believe in this idea that in some, in like five or ten years, we'll only be interacting with services through a chat.

00:30:33But I definitely see that there is a vastly rising importance of building services that are easy to use by AI agents.

00:30:43And if anything only because it's a hype, but I also see some true value there because I myself see that I conduct research through chat GPT or through a locally running agent.

00:30:55So, of course, whatever you're building can and likely will be a full stack application that has a web fronted for humans, but it may also just be an API.

00:31:10It's not unthinkable that you're building a service that is like a REST API, which could be consumed by an agent, maybe with a MCP server, maybe with a CLI in front of it, right?

00:31:21The CLI is also a client, just not a very visual one, not a nice website, but it's a client that can interact with your service.

00:31:29And that is 100% something you could build, depending on what your service is about.

00:31:34If your main audience is people that are using AI agents for whatever problem you're solving, you could be building just an API, maybe with a CLI in addition to it.

00:31:45So that is also worth noting.

00:31:48And again, if we're talking about backends and stuff, I already mentioned it, but HONO, Express.js, Elysia.js is another great framework and many, many more.

00:32:00Don't get me wrong.

00:32:01Like Fastify.

00:32:03There are so many.

00:32:05Is this still maintained?

00:32:06Yeah, it's absolutely maintained.

00:32:08So there are so, so many frameworks out there and you maybe don't even need a framework.

00:32:13I think that's also worth noting here before I move on, especially if you're using something like BUN, but also, as I mentioned, modern Node.js, there's a bunch of stuff built in, right?

00:32:22So it is possible that you're building something which doesn't even need a framework because I like HONO a lot.

00:32:30It's a great framework.

00:32:32One of the most important things it adds is convenient routing and middleware and stuff like that.

00:32:38And that is worth a lot.

00:32:39That can make your life easier, right?

00:32:42But it is also important to mention that BUN, of course, also has built-in HTTP request handling and Node.js has that too.

00:32:52I mean, that's kind of the core thing or one of the core things you can do with these runtimes.

00:32:57And BUN has built-in support for setting up routes and stuff like that.

00:33:02So you may be able to build an API with just BUN or with just Node.js or with just whatever you're using, with just Python.

00:33:11I don't know.

00:33:12You don't always need to reach for a framework or a library just because you're used to that.

00:33:18And I think, again, with AI, that's kind of changing because in the past you may have used a library because using the raw runtime would have required a lot of boilerplate code, a lot of typing.

00:33:29And guess what?

00:33:30Boilerplate code is kind of free these days.

00:33:32Not entirely.

00:33:33It costs you tokens, of course.

00:33:35But it's easier to generate such code on your own or to have AI build a little helper library for wiring up a middleware.

00:33:42So that is definitely also worth considering, I would say, that you can't just use these raw tools.

00:33:50And also important, since I mentioned the CLI part, you are not limited to using BUN, Node.js, Dino, and so on to build servers.

00:33:59That's one of their main use cases, but it's, of course, not the only one.

00:34:02You can also build CLI tools with them.

00:34:06I mean, again, BUN.

00:34:08And this is really not a BUN advertisement.

00:34:10It's just a runtime I like a lot.

00:34:13But BUN has this single file executable feature here, which means if you build a TypeScript application that uses BUN features, maybe, you can build it into a tool that you can just distribute to other people and they can run it without having BUN installed on their system.

00:34:31So that can be great for building CLIs, for example.

00:34:34Just worth a thought.

00:34:36Okay, but now with that, I really want to conclude this.

00:34:39That is how I think about this right now.

00:34:43But when planning a web application, it's about a bit more than just which framework do you want to use, right?

00:34:51Or which framework do you know and feel comfortable with?

00:34:54It's also about all that other stuff.

00:34:56And for example, the hosting and deployment question is also important because, for example, let's say you're choosing Vercel as a host.

00:35:10Yeah.

00:35:12Yeah.

00:35:12And of course, again, there are others like Netlify.

00:35:28And again, I'm mostly talking about the JavaScript ecosystem here because that's kind of my thing.

00:35:34These are great services that make hosting really simple.

00:35:38And there are many others, as mentioned.

00:35:41But they have one disadvantage.

00:35:43This simplicity comes at the price of less configurability and you're kind of limited to the features they offer you.

00:35:52So, for example, when it comes to a database, if we search for Vercel database, we get this Vercel storage page.

00:36:02And we learned that right now, at least, at this point in time, there isn't really a great database solution here.

00:36:11You'll need to bring an extra database service, which, of course, is 100% an option.

00:36:16But if you decide that maybe you want to build an application and you want to use SQLite, let's say, again, an amazing database, I'll get back to it.

00:36:24But let's say that is your choice.

00:36:25You want to use SQLite.

00:36:27Well, Vercel and SQLite, not working.

00:36:32So, yeah, I mean, here, they offer other storage solutions.

00:36:44It can't be used with Vercel.

00:36:46That's what I was looking for.

00:36:47Again, not necessarily a problem.

00:36:49You can go for alternatives.

00:36:51You can use something like PlanetScale.

00:36:55They give you hosted Postgres, essentially.

00:37:02So, that's basically database as a service, you could say.

00:37:05You could be using AWS Aurora or AWS RDS or the equivalents by Azure or Google Cloud.

00:37:15You could be using Convex, which is quite popular.

00:37:18These are all database services you can use and you can build your application.

00:37:22You can build it with any framework you want.

00:37:25You can build it with Next.js, for example, also, or you build just a REST API.

00:37:30And you can use Convex or PlanetScale and the database they give you and use that service and connect it to yours.

00:37:37You can do that.

00:37:38But that, of course, is some extra cost.

00:37:41You'll be paying for these services.

00:37:43It's also some extra complexity.

00:37:46It's an extra dependency.

00:37:47If one of those services shuts down, in the worst case, you'll have to migrate your data.

00:37:53If they increase prices, maybe a problem.

00:37:58If it's down, maybe a problem.

00:38:00But, of course, you also get some convenience.

00:38:02You get automated backups.

00:38:03And it's always a trade-off.

00:38:05You can use that.

00:38:07But for many applications, in my opinion, just using SQLite, for example, is an amazing option.

00:38:14Or you could use Terso, which is basically a modern version of SQLite written in Rust.

00:38:21But SQLite is also pretty amazing.

00:38:23Just because it's written in Rust does not mean that it's better.

00:38:26But it is another great re-implementation of SQLite, essentially.

00:38:33So, yeah, there is a lot of complexity here.

00:38:37But I personally like SQLite a lot.

00:38:40The website doesn't look like much.

00:38:43But it's an amazing database.

00:38:45And the huge advantage of SQLite, for example, is that it's really just a file on your system.

00:38:50It's great for development because it's just a file on your machine.

00:38:54It's no service you have to spin up.

00:38:56Nothing you have to connect or pay for.

00:38:57It's just a file on your machine.

00:38:59And if you deploy it, it's just a file on the VPS, for example.

00:39:03Because that's another deployment option to which I'll get back.

00:39:05It's just a file on a VPS.

00:39:08Backing it up is as simple as copying it.

00:39:11You should copy it in a certain way.

00:39:13But it's easy to back up.

00:39:16But it's your responsibility to back it up.

00:39:18And, yeah, if you have, like, millions of active users per second or per minute even is probably enough.

00:39:27It might reach its limits.

00:39:29But it can go very, very far.

00:39:31And if you're just starting, if you're building something, go for a simple solution.

00:39:36And if you are lucky enough and successful enough that you have thousands, hundreds of thousands of users and you suddenly feel that maybe load is becoming a problem, that's a good problem to have.

00:39:48You can then scale up.

00:39:50You can migrate to a better database solution.

00:39:53It'll be some effort.

00:39:54But it's probably worth it to not have, like, the final solution right from the start, which may be super complex to set up where you're paying a lot.

00:40:03And, therefore, that would be my go-to database in many cases or something like Postgres or MySQL running in a Docker container on your VPS.

00:40:12Owning the database comes with responsibility, like the backup part especially.

00:40:19But it also gives you a lot of flexibility and freedom.

00:40:23And it also can make development easier.

00:40:26So these are, again, trade-offs here.

00:40:28Okay, lots of rambling again.

00:40:30Sorry.

00:40:31Let me dive back into the chat.

00:40:33I'm actually using 10-stack start a lot.

00:40:38How to prepare a 10-stack project for AI development?

00:40:41Yeah, what I do with 10-stack start or in general, so frameworks which are maybe not as well known by AI, I like to, in a project, point it at the documentation.

00:40:54So I have my agent, like Pi or Codex, and say, hey, I'm using 10-stack start.

00:40:58Here's a link to the docs, and I'll take the entry page or some page.

00:41:03And I say, I want you to analyze the docs.

00:41:06And I want you to gather information, especially on ABC.

00:41:09And these are just some concepts I know that are important.

00:41:13For example, for 10-stack start, I know that it's important to get server functions right.

00:41:18I have run into problems there when you have some incorrect code here.

00:41:24So I point AI at that, and I ask it to gather all that information and then write a skill for me.

00:41:30And I know there are official skills, too.

00:41:32And 10-stack start publishes skills and documentation, I think, with their packages, which is nice.

00:41:38AI can find that, the installed packages.

00:41:41But I have it build its own skill, essentially.

00:41:43And then I can copy that skill across projects and update it from time to time.

00:41:47And if it, for example, makes a mistake, I can tell it to update the skill and make sure that this knowledge is reflected there.

00:41:55But again, that is why I do review the code and why I think it's important to understand what you're working with so that you can steer AI there.

00:42:03But that's basically my approach.

00:42:07Top three books everyone must read.

00:42:09So if we're talking about coding books, obviously my book, if you want to get, right, if, whoops, I got to fix that.

00:42:24If you want to get better at React, React key concepts, that is a strong recommendation.

00:42:31But, yeah, jokes aside, I think Data Intensive Applications, that book.

00:42:36Recently got a second edition, which I already got.

00:42:39That's the first edition, but there is a second edition available.

00:42:44That would be a recommendation.

00:42:48Then, let's see, maybe it's already here.

00:42:51Yeah, I like the philosophy of software design quite a bit.

00:42:55And I don't know which one it was.

00:42:58One of these software architecture books.

00:43:00I like the two, The Pragmatic Engineer, also a book I liked a lot.

00:43:04But, yeah, I would probably, I would definitely go with the signing data and intensive applications and philosophy of software design.

00:43:11I think these are pretty good books.

00:43:17Do you think AI is affecting my skills or our skills badly?

00:43:23AI is definitely having an impact on the skills, I would say, on my skills, too.

00:43:27And I try to stay sharp by really challenging myself to not fall into that trap of blindly accepting stuff, but to really understand it, to write some code myself here and there, and to really also read the docs myself.

00:43:42And not just point AI at it and accept whatever it gives me, but yeah, I mean, I think the skill set is also changing.

00:43:49It's simply changing, and it will be more about picking the right building blocks and being able to understand the code and about code architecture.

00:43:57And I think that's becoming more and more important, and I'll also create some courses on that.

00:44:01And, yes, they will also be available on Udemy then.

00:44:04But, yeah, there definitely is a change there.

00:44:07Absolutely.

00:44:12Oh, my.

00:44:13So, so much interactivity.

00:44:14That's amazing.

00:44:17And thank you all so much for all those amazing, very nice comments.

00:44:21It means a lot to me.

00:44:22Thank you so, so much.

00:44:23Really amazing.

00:44:25On headless CMS.

00:44:26Yeah, I'll also get back to that.

00:44:28Yeah.

00:44:31Yeah, boilerplate code is free.

00:44:34Debugging thousands of lines of wipe code at boilerplate when it breaks is not free.

00:44:37Absolutely.

00:44:37So, again, that's why it's so important to me.

00:44:41You got to understand that code that's generated and not blindly accept it all, including boilerplate code.

00:44:50I don't mean that you can let it generate boilerplate code and just don't care because it's boilerplate.

00:44:56I care about all the code.

00:44:58It's just easier to tackle certain problems which you maybe didn't feel like before or were too lazy to do.

00:45:06Because reading and analyzing that code, of course, still is something you have to do, and it's not free.

00:45:12It takes time, but especially for boilerplate code, you may be able to go through it fairly quickly, and that's the part where I think AI can help you.

00:45:26Library framework-wise, I think it really depends on the situation, since sometimes a library or framework has more than what you actually need.

00:45:32In those situations, I think it might be an idea to think going for building your own.

00:45:36Yeah, definitely.

00:45:37And, I mean, also there are certain things.

00:45:43Let's take authentication.

00:45:45I'll get back to it in more detail.

00:45:46But authentication is something which is not that difficult to do on your own against popular belief, but it depends on the scope.

00:45:57Adding basic email password authentication, doable.

00:46:02You definitely have to read up, do some research to get it right, but it's doable.

00:46:06But if you start adding OpenID Connect, if you are diving into OAuth, if you want to add a bunch of different connectors there,

00:46:15if you want to support authentication via Discord, via Google, via Apple, you may want to use a library there,

00:46:23because even with AI, getting all of that right will take time.

00:46:28You definitely don't want to mess up and have security issues there.

00:46:32And, of course, you don't have a guarantee that there won't be security issues with a library,

00:46:36but you're still offloading that task to someone who may be dedicating more time to that question than you are.

00:46:43So there is that trade-off for sure.

00:46:51Any advice on how I could apply the knowledge that I've learned from your course to build the actual project?

00:46:57As for now, I'm completely stuck and don't know where to start.

00:47:00What I like to do is really just build clones, simplified clones, a simplified Amazon clone, simplified Twitter clone, anything like that, and go from there.

00:47:11And with AI, you've got a partner.

00:47:12You can also ask to build some parts of it, but also to understand the code it may be generating, or you may be writing, or to discuss alternative solutions.

00:47:24You're facing a problem, you can discuss it with AI.

00:47:28It doesn't mean that its proposed solution is the best one, but at least it can get you started.

00:47:33I think that is something you can do.

00:47:38The video is getting choppy.

00:47:40Video is laggy.

00:47:45Okay.

00:47:50Let's see if that problem persists.

00:47:56Sorry about that.

00:47:58I hope the video is better.

00:47:59I had those problems before.

00:48:01Did change the encoding settings?

00:48:03I'll close some processes behind the scenes on my second screen to maybe free up some CPU power.

00:48:13Let's see.

00:48:14Voice has a delay.

00:48:17Yeah.

00:48:24Let's see.

00:48:32Give me a sec.

00:48:33Works good on Twitch.

00:48:45YouTube was choppy.

00:48:46Huh.

00:48:49Maybe it's a YouTube problem.

00:48:55The thing is, I'm also just not a live stream expert.

00:49:06Let's see.

00:49:10Let me tweak that a little bit here.

00:49:13I don't know.

00:49:17How is it now.

00:49:18How is it now?

00:49:18Maybe that will be better.

00:49:21We'll see.

00:49:22I don't want to waste too much time on this now.

00:49:24Okay.

00:49:26So, yeah.

00:49:27Let me continue here.

00:49:30And I hope the delay between video and audio isn't too bad.

00:49:35I think there still is some delay.

00:49:38I'm not sure why I suddenly have that.

00:49:48Yeah.

00:49:49But let's see if that problem persists or if that will get better.

00:49:58Okay.

00:49:58Good.

00:50:00Okay.

00:50:01So, a question for everyone.

00:50:03What is a better AI coding setup, Pi coding agent, codex subscription, or cursor pro plan?

00:50:08I'm currently going with Pi plus codex, but I'm also frequently switching to try out different options.

00:50:13But Pi plus codex works pretty well for me.

00:50:15Yeah.

00:50:16But, yeah.

00:50:17Just my current stack or what I'm currently using, so to say.

00:50:29Okay.

00:50:29So, what I said, audio, audio, hosting and deployment is another important consideration.

00:50:42There are hosts, services, like Vercel, but I already mentioned there are trade-offs.

00:50:52And those trade-offs influence the other choices.

00:50:56So, for example, database.

00:51:00I mentioned SQLite, pretty good choice in my opinion.

00:51:03But also, that has disadvantages, more responsibility for you, but also more flexibility, easier to set up, cheaper.

00:51:10It's always a trade-off.

00:51:11And especially, it's different, of course, for established, large projects.

00:51:16But if you're building something new, starting simple has a lot of merit, I believe.

00:51:22So, that's one thing.

00:51:27The alternative, or there are many alternatives, of course, to services like Vercel.

00:51:33There is, for example, Cloudflare.

00:51:35I personally like Cloudflare a lot.

00:51:38And they are really releasing new products every day, it feels like.

00:51:44Their main product for hosting is their workers' product.

00:51:50And it's a kind of a special product because it's not a server where you install stuff and then it just works.

00:51:56Instead, it's basically there to put your code there, the code that handles an incoming request.

00:52:04And when a request comes in, the worker starts, handles the request, does its thing, shuts down.

00:52:10So, obviously, no place to put a database there, right?

00:52:13There is nothing there.

00:52:15So, unlike Vercel, they have a database product, D1.

00:52:18But that, again, is like a managed, hosted database just coming from Cloudflare.

00:52:23Of course, you could also connect another database service to them.

00:52:27So, I like them because that workers' model can be super cheap for getting started.

00:52:33You're only paying for when it's running code.

00:52:36And even then, it's really cheap.

00:52:37There is a very generous free tier.

00:52:39And therefore, again, if you're getting started, you don't have any users yet.

00:52:44It can be cheap to get started with it, but the downside of choosing something like Cloudflare is that you're building for Cloudflare.

00:52:52You're building for their workers' architecture.

00:52:55On that worker, you also can't use Bun.

00:52:58It's their own JavaScript runtime, which has a lot of Node.js compatibility, but it's another trade-off.

00:53:04You're limiting the amount of features you're using.

00:53:07You may not be able to use all libraries if you make that choice.

00:53:11So, choosing where you want to host your applications, how you want to deploy it, matters.

00:53:17And you should choose it pretty early on because it influences the rest of your tech stack, or it can influence it at least.

00:53:25Now, of course, you could also host on a VPS.

00:53:28You could use something like AWS EC2, which is, in the end, just a server in the cloud you're renting.

00:53:34Or you could be using Hetzner, which is the same, and it's cheaper than AWS.

00:53:39You can do that.

00:53:42And they recently increased their prices, but it still allows you to get started at a fairly reasonable price.

00:53:50The advantage of that, and I have a course on that, too, in case you're interested.

00:53:54I have a course about configuring a VPS.

00:53:59Here it is, VPS Essentials.

00:54:01The advantage of a VPS, of course, is that you have full control.

00:54:05It's like your own computer, right?

00:54:07You can install whichever software you want.

00:54:09You can put your database there, no matter if it's a SQLite database or maybe Postgres in a Docker container.

00:54:16But it means you also have to deal with all of that.

00:54:19You have to deal with configuring that VPS, with securing it.

00:54:22You may be using Docker to Dockerize your application to deploy it, something you don't have to do with Cloudflare or Vercel.

00:54:29And suddenly there is new stuff you have to learn and you have to get comfortable with.

00:54:34And that, again, is a learning curve.

00:54:36It's always about these tradeoffs, but I like the control you have.

00:54:40And therefore, for me, for most projects, it's either Cloudflare because super cheap, but the restriction is or the problem is that I'm limited.

00:54:49I have to use their runtime or I'm going VPS.

00:54:52There are very good reasons to use Vercel or Netlify instead, of course.

00:54:57But it's super important to make that decision early on and to spend some time on it and not rush it.

00:55:02Maybe discuss it with AI, but also dive into the docs, read some articles on it, build an opinion to understand what the tradeoffs are.

00:55:11And maybe these are also things I could create more detailed videos if that's helpful.

00:55:16But that is, in the end, my thoughts here.

00:55:24Any open source models that can be used locally?

00:55:27Yeah, I like open source models.

00:55:28I like the JAMA models, for example, but not for coding.

00:55:31For coding, the models that I can realistically run on my machine are just not good enough yet.

00:55:37But for stuff like summarizing documents or something like this, I like the JAMA models a lot.

00:55:42Pi, since the question was in the chat, it's a coding agent like Codex or CloudCode.

00:55:48But unlike those, it's not primarily built for one service.

00:55:52CloudCode, of course, is bianthropic for Cloud.

00:55:55And Codex does work with other models too, but it's built by OpenAI for their models, for the GPT models.

00:56:03Pi is a super minimal agent that only has a few tools built in, but it's highly extensible and it can extend itself.

00:56:12And I got a course on Pi coming up as well.

00:56:15The documentation is also really great, though, if you want to dive in.

00:56:18And I like because it's fast, because it's extensible.

00:56:22There are lots of packages you can add to add new features in a very convenient way.

00:56:27So, yeah, all these agents are useful and there isn't like a best one or anything like that.

00:56:33But I personally like working with Pi for these reasons.

00:56:38And for me, it's better because minimal core and then I can add whatever I want to add.

00:56:42But I also just like tinkering with stuff.

00:56:45And that's probably also because I like Pi or why I like Pi.

00:56:53My thoughts on dock ploy, to be honest, I have to...

00:57:02Look at it.

00:57:03Oh, that's...

00:57:04It's like...

00:57:05Is it like Coolify?

00:57:07I think these are great services, which I have not used.

00:57:12But I absolutely see the merit in turning your VPS into more of a managed platform without paying a managed platform.

00:57:21So, that's kind of the idea behind Coolify, at least.

00:57:23And I think dock ploy is similar, if I'm not wrong.

00:57:25It's basically a tool you install in your VPS and suddenly there is less you have to take care about.

00:57:32Because they or this tool, the service, which now runs on your VPS handles a lot of configuration, makes deployment easier.

00:57:40And it's a bit like turning your VPS into Vercel whilst keeping the VPS advantages or many of the advantages.

00:57:50Again, I like tinkering, which is why I tend to configure my VPS on my own and stuff like that.

00:57:56But I definitely see the advantage of tools like this.

00:57:59And it's very good that it was mentioned in the chat because it is something you absolutely should consider in that deployment decision.

00:58:06You can go the VPS route and use a solution like dock ploy or Coolify to make that deployment easier.

00:58:17Yeah, the advantage of the PyAgent is customization.

00:58:19What customization have I made?

00:58:22I built a bunch of extensions or skills on top of it that helped me with planning, with implementing planning, with code reviews.

00:58:29I also use Py for non-coding stuff, for analyzing finance data, so for my bookkeeping in my company and added extensions that help it read PDF documents and stuff like that.

00:58:44So, yeah, that's what I built or what I had built Py for me, so to say.

00:58:52I only use Cloudflare for SSL and routing and traffic to analyze.

00:58:56Yeah, and I mean, I think Cloudflare, when it started many years ago, it was mostly about their CDN, their cloud delivery network, content delivery network, and about their protection against denial of service attacks.

00:59:15But nowadays, it's so much more.

00:59:17But even those basic features are pretty great.

00:59:20Cloudflare, I think, is a great company or service.

00:59:25And Cloudflare R2, for example, is my go-to solution for storing data in the cloud instead of S3, simply because it's cheaper.

00:59:34I completed the VPS course, great course, gained a lot of knowledge from it.

00:59:37Thank you so much.

00:59:38Super happy to hear that the VPS course was helpful.

00:59:41Thank you so, so much.

00:59:43Thank you for courses for Go and Vue.

00:59:46Yeah, again, thank you.

00:59:47And thank you all.

00:59:48So many nice comments.

00:59:49Thank you all so, so much.

00:59:52I'm super happy that the content is helpful.

00:59:54And I got more interesting courses planned, as I mentioned, on Py, on Bun, systems design, many courses planned.

01:00:03Do you think focusing on Docker and Kubernetes knowledge is one of the best things to master in this AI era?

01:00:13I'm not sure if it's the best, but I think Docker especially is something every developer should know.

01:00:20You don't have to be an expert, but you should know it.

01:00:22And, yeah, with AI, but also before AI, if you're building something, if you're deploying something, Docker can be a very useful tool.

01:00:31With AI, it can also be a useful tool for having, like, a sandbox.

01:00:36I could talk an hour about Docker and using it as a sandbox because there are some pitfalls there, too.

01:00:43It's not a perfectly secure sandbox out of the box, necessarily, though on macOS it's quite good.

01:00:49But, yeah, that takes us a bit too far, but Docker is a technology I think every developer should know and should have a basic understanding.

01:00:56Kubernetes, I think, is optional.

01:00:59Depends on if you need it for your job.

01:01:02And at some point, at some scale, it definitely becomes important.

01:01:07But Docker, I would say, is more important.

01:01:12DocPlay is exactly like Coolify.

01:01:13Okay, yeah, I got this right.

01:01:14And, yeah, these are great solutions for self-deploying on a VPS, I would say.

01:01:28What AI model providers do you use?

01:01:31Have you used something like Open Router to get all the models in one API?

01:01:34I'm mostly using my ChatGPT subscription.

01:01:37I got the $200 subscription, so I'm using that, trying to get some value out of it.

01:01:41Switching to the Anthropic subscription from time to time to see how those models work.

01:01:47And, yeah, Open Router or something like Vercel AI Gateway or the Cloudflare AI Gateway.

01:01:54I also used those or do use those from time to time, for example, to use models like GLM, Cloudflare, GLM 5.2.

01:02:12Models like this, which may be hosted by Cloudflare or in Open Router.

01:02:16You can access that from various hosting providers.

01:02:20And it's a great model and quite a bit cheaper compared to the Cloud models.

01:02:27So, yeah, I am using such services, too.

01:02:29And you can use them easily with Pi, too, which is nice.

01:02:37And, no, the VPS course does not cover DNS.

01:02:40It's really only about understanding VPS, doing that base hardening.

01:02:45It's that Essentials, that Basics course.

01:02:47It's not that Advanced Setup Everything course.

01:02:49So, just to set expectations right there.

01:02:52I do show how to deploy an application there with a reverse proxy,

01:02:57but it's not a deep dive networking or anything like that course.

01:03:01Now, back to that list.

01:03:02So, hosting deployment, super important consideration, which you should cover early on,

01:03:08has an influence on the database, for example.

01:03:11And I shared my thoughts there.

01:03:13There are good reasons to use a managed service.

01:03:15There are reasons to self-host.

01:03:18The trade-off always is like cost and responsibility and stuff like that.

01:03:21I touched on authentication already.

01:03:23This is one area where I think, as I mentioned before,

01:03:27building it on your own is not as hard as some people make you believe it is.

01:03:32But at the same time, once it gets a bit more complex, more advanced,

01:03:44I'm happy to hand off responsibility there.

01:03:48But then again, you have choices.

01:03:51I like the better of library a lot.

01:03:54It's a relatively new library which gives you all kinds of auth-related features.

01:04:10It's well-maintained.

01:04:11That's just what I wanted to check.

01:04:13It has a basic email password authentication.

01:04:17It has support for all these OAuth providers, right?

01:04:21So basically, every company that was ever founded can be used for authentication with BetterAuth.

01:04:29They got more advanced.

01:04:31They got integration guides for popular frameworks and so on.

01:04:34They got plugins for more advanced stuff like two-factor authentication or pass keys or for supporting organizations and teams or setting up API keys,

01:04:47having admin features, stuff like that, and also enterprise features like OpenID Connect or Skim and stuff.

01:04:55That is all really, really nice.

01:04:58And that's my go-to solution for authentication.

01:05:03But of course, you could also there use a service like Clerk or WorkOS, I think is another one,

01:05:10where they give you a library, you add to your project, and then you hand authentication off to them.

01:05:18So the user actually signs up using their servers.

01:05:22User data gets stored in their databases, and they charge you money depending on your plan.

01:05:28But the convenience is that you're totally not responsible for the user data, of course.

01:05:34So if your database gets breached, if your server gets hacked, they are responsible for the user data.

01:05:42They are responsible for securely authenticating users.

01:05:45But again, it's a trade-off.

01:05:47You may not want to give up that user data.

01:05:49You may not want to pay for something that's not that difficult to do on your own.

01:05:53But these, again, are really just trade-offs you should think about.

01:05:57I'm personally a fan of doing it myself or doing it with Better Off, to be precise.

01:06:03But there are good reasons for choosing a managed provider there.

01:06:07By the way, I just saw Superbase being mentioned in the chat.

01:06:13Another great service.

01:06:15And I know all the popular with AI models and AI agents.

01:06:22Personally, I haven't used Superbase a lot.

01:06:25Of course, I played around with it to understand how it works, which features it has.

01:06:29And it gives you database authentication, all of that.

01:06:34And you can absolutely self-host it.

01:06:37But you can also...

01:06:38Am I wrong?

01:06:42Superbase self-hosting?

01:06:44That works, right?

01:06:45Yeah, you can self-host it.

01:06:47But you can also use their cloud service.

01:06:48And they also have a free tier.

01:06:52And you then get one solution that gives you, like, database storage authentication, file storage.

01:06:58Also super convenient.

01:07:00Very nice.

01:07:01And you can self-host it as mentioned.

01:07:06But you then, again, have one service, even if you self-host it, where you may not be able to tweak everything exactly as you wanted.

01:07:15Or to switch stuff.

01:07:17And then replace certain parts as you wanted.

01:07:20Again, always a trade-off.

01:07:22And, again, for me, it's not...

01:07:25I find it not that difficult to set up a basic application with SQLite, batter-off, and some library of my choice.

01:07:33But there simply is no wrong or right here.

01:07:37That is really, really important.

01:07:42Have I tried Forge Yo yet?

01:07:47No.

01:07:47And I don't even know it, to be honest.

01:07:50So I don't know what it is.

01:07:52So I have not tried it.

01:07:53Sorry.

01:07:55Do you have any plans for a course on AI-powered web apps, like Vercel SDK, Transformers.js?

01:08:02I got some plans there, but mostly about building AI agents with these relatively new agent frameworks, like the EVE framework I mentioned before, or the FLU framework, which looks interesting.

01:08:14I find those frameworks quite interesting and intriguing.

01:08:17And I'm playing around with them to see if they, A, are useful for me, and, B, if I want to create some content or courses about them.

01:08:26Yeah.

01:08:31Did I miss the question I asked about Laravel?

01:08:33I did answer one question about Laravel earlier.

01:08:37Not sure if that was your question, to be honest.

01:08:40But regarding Laravel, Laravel, amazing framework.

01:08:42I loved it when I worked with it 10 years ago.

01:08:45I just haven't worked with it since then.

01:08:47But I know it's an amazing framework, and I mentioned basically everything built in you need for most applications.

01:08:53So, an amazing choice, in my opinion, if you are into Laravel PHP, essentially.

01:09:04Yeah, I already talked about Pi.

01:09:06I like the minimal core and the extensibility.

01:09:09I see people are creating AI sub-agent developer team to work for their project automatically.

01:09:14Do you think it's necessary, or do you think you still need human monitoring?

01:09:17Yeah, so, whenever I read about people that are spinning up teams of agents, where the agents are maybe collaborating and doing stuff, I think it's just marketing fluff.

01:09:29So, no.

01:09:31AI agents, or AI in general, is, of course, obviously an important technology, and I use AI for coding every day.

01:09:40And AI writes the vast majority of my code.

01:09:43And I use those coding agents, of course, and those agents also spin up sub-agents for research.

01:09:49But I'm nowhere near, nowhere near to having a team of agents that would do something on its own, or where maybe multiple agent teams would collaborate.

01:10:01It's always me who steers the AI and kicks off tasks.

01:10:06And I'm not sure if that works differently for some people, but I think it can feel super productive to have a bunch of agents doing stuff and maybe kind of communicating with each other.

01:10:17However, I also think currently, and I also think for the foreseeable future, it probably ends in, let's say, sub-optimal code.

01:10:30Have you done a course on tail scale?

01:10:34Is that the question?

01:10:35Chat scrolled up.

01:10:40Where is it?

01:10:47Course suggestion about Docker, by the way.

01:10:50I have a Docker course, which I would suggest.

01:10:57So, that is great for learning Docker, I would say.

01:11:00And tail scale, I have no course on that.

01:11:03I'm not sure, because I only use the basic tail scale features, but those are already amazing.

01:11:09Being able to put all your VPS and your other devices into your own private network over the internet is just amazing.

01:11:16And you can even use all of that for free.

01:11:18It's quite generous.

01:11:20So, yeah, tail scale, amazing.

01:11:22Whoops.

01:11:23But, yeah, no course planned, because for those basic features, I don't feel like a course adds a lot of value.

01:11:29And I'm not using any advanced features, to be honest.

01:11:33Yeah, Superbase is an alternative to Firebase.

01:11:44Maybe better, but most importantly, also usable.

01:11:48Well, you can also use Firebase for free.

01:11:51You can also self-host Superbase, though, which you can't do with Firebase, at least last time I checked.

01:11:56And it's definitely more popular, at least in the bubble I am on X and so on.

01:12:03But I think Firebase is not as bad as its reputation sometimes seems.

01:12:10I have used a lot of Firebase, especially around 2017, 18.

01:12:16And I thought it was a great service.

01:12:18And also a great managed backend, essentially.

01:12:23But, yeah, again, I'm currently not really using too many managed backends, except for all the Cloudflare stuff or for projects that I build more on Cloudflare.

01:12:34There, things look different.

01:12:37Have I played with Tori a bit?

01:12:39Yes.

01:12:40But mostly, if I'm dabbling around with desktop applications, I always get back to Electron.

01:12:45But I'm not much into desktop application development.

01:12:50Apple just increased prices.

01:12:51Yeah, I saw.

01:12:52They had to increase the price because memory and so on got cheaper.

01:12:55I mean, it's funny that with Apple for the last two years, we had pretty good deals because they did not adjust their prices, right?

01:13:04Am I writing any code by hands these days?

01:13:06Yeah, I am.

01:13:07I still am.

01:13:07But the majority of code is definitely generated by AI.

01:13:14But, again, I do validate and check all that code.

01:13:19And I do change it by hand or with AI.

01:13:22It's not like, hey, write that code.

01:13:24Okay, next code.

01:13:24I don't care.

01:13:26That's Vibe coding for me.

01:13:27And I'm not doing that.

01:13:29Except for throwaway software where I don't care about some utility tool, which I just need right now.

01:13:35I don't care about the code if it does the job.

01:13:38But for anything serious, no.

01:13:42Have you explored the idea of a second brain or building a knowledge base using AI?

01:13:46Yeah, I played around with it.

01:13:47Wasn't super satisfied with the results, to be honest.

01:13:51But also because I'm probably too messy and I got my stuff.

01:13:55I'm not good at keeping my stuff organized.

01:13:57So, yeah, AI has a hard job with me.

01:14:03I don't see many people talking about running any of these AI tools in a sandbox environment.

01:14:07Currently, our only approved AI use at this client is 365 co-pilot.

01:14:12Yeah, I mean, that entire sandbox stuff is also in.

01:14:18There's so much happening there right now.

01:14:20There's like a new sandbox startup starting every day.

01:14:24AWS announced their Lambda Micro VM service, which is essentially about sandboxing.

01:14:30So, there's a lot of stuff happening there.

01:14:33And it'll be interesting to see in a year or so, once the dust has settled a bit, which approaches actually survive.

01:14:40I mean, there, for example, is something JustBash is a tool, you could say, by Vercel,

01:14:47which gives you a super simple sandbox you can run on your system.

01:14:51And I find stuff like that quite interesting.

01:14:54But, yeah, that is all in so much active development right now.

01:15:00Have I tried spec-driven development?

01:15:03Depending on how strict you are, yes, I have.

01:15:06But I don't believe in building one large spec and then letting AI do its stuff.

01:15:13I've always been more about iterating.

01:15:17And I have to feel out the product, right?

01:15:20Obviously, I have a plan.

01:15:21I have a spec, but then I only notice where I realize something's missing, where something doesn't work the way I want it to work once I build the product.

01:15:31That has always been the case for me.

01:15:32And obviously, that does not work for every team or especially not every team size.

01:15:37But for me, alone or in a small team, that does work quite well.

01:15:43What do you think about Micron having an 80-point preserved margin in memory?

01:15:47Yeah, I think these are all things that will ultimately change at some point.

01:15:50But right now, I guess, good for them, right?

01:15:52But nothing lasts forever.

01:15:55So, let's continue.

01:15:58And there's the perfect question for continuing.

01:16:00How do you approach UI design?

01:16:02Do you just use Tailwind?

01:16:03There's a shortage of good UI frameworks these days.

01:16:06So, let's talk styling.

01:16:07Because it matters if you're building a frontend, at least.

01:16:11A web frontend.

01:16:13For styling, again, for me, it's mostly vanilla CSS.

01:16:22So, modern CSS has come a long way.

01:16:28And I have some videos on my Academy channel.

01:16:30They didn't perform too well, unfortunately.

01:16:33But modern CSS and the modern browser in general has pretty great features

01:16:39that make working with CSS and that make building modern UIs in the browser

01:16:46so much easier than it was three or four years ago.

01:16:49And I think many people missed that because, A, a lot of people don't like styling and CSS.

01:16:55That's just always been the case.

01:16:58And, B, now with AI, everybody's just focusing on AI agents and stuff like that.

01:17:03So, yeah, kind of a lot of people are missing how powerful modern CSS and the browser got.

01:17:10And especially for me, I mean, I grew up and I became a web developer back in the old days

01:17:16when we still had to support Internet Explorer 8, I think it was.

01:17:21And that was no fun.

01:17:23And it's so much better now.

01:17:24So, for me, styling is mostly not about a library but about using modern CSS.

01:17:29And, again, you can get the AI to do that quite well.

01:17:32And you can always kickstart it by writing some initial code or some pseudocode on your own

01:17:38by defining the style system and so on.

01:17:41But if I'm not using that, Tailwind is what I'm using

01:17:44because one of the big advantages Tailwind has, always had, and still has, of course,

01:17:51is that you can just go into the code and quickly iterate on your design

01:17:56by just switching those class names, right?

01:17:58If you're using vanilla CSS, one disadvantage is that you have to find the right CSS file,

01:18:04find the right rule, the right property, and change it.

01:18:08And with Tailwind, you just go to the markup, you know which box you want to change,

01:18:12and you go there and you tweak those classes.

01:18:15And that's super, super fast with or without AI.

01:18:18And that's one of the big advantages, and I think that's also why Tailwind got so popular.

01:18:23So, yeah, I still also use that.

01:18:25And this iteration speed when building a design is really great.

01:18:29Again, it depends on your approach.

01:18:31If you're using something like Figma for the design, that may not matter to you,

01:18:35but I'm designing on the go, so to say.

01:18:38And there, of course, Tailwind is pretty good.

01:18:41But with AI, it's also pretty easy with vanilla CSS

01:18:44because you can just tell the AI, we need more margin here.

01:18:48Or you, of course, dive into the file on your own.

01:18:51You can easily find it in the DevTools, for example,

01:18:53if you don't remember where something is configured.

01:18:56And you can, of course, also tweak the designs in the DevTools,

01:18:59and then just, once you're happy, instruct the AI or on your own make those changes.

01:19:07Do you think going DevOps as a junior is viable,

01:19:11or is it more of a role that someone pivots to after some work experience?

01:19:16No, I think going DevOps as a junior is absolutely viable.

01:19:20Of course, you'll still need those developer basics.

01:19:24I mean, it's DevOps, after all, not just ops.

01:19:27But why not?

01:19:29I mean, again, the question is, what exactly is the DevOps for you?

01:19:35But that entire part or that entire aspect of managing deployments,

01:19:39CI, CD, agent sandboxes,

01:19:43interacting with all these different services,

01:19:46that is important, that has been important.

01:19:51And I think that is absolutely a viable career and something you can do.

01:19:56And it's just a combination.

01:19:59Of course, you have to understand what gets deployed.

01:20:03So you have to have basic development understanding, obviously.

01:20:06But I definitely see that as a valid option.

01:20:13Oh, yeah, Astro.

01:20:15I just read about Astro, another pretty good framework, by the way,

01:20:20which recently got a brand new version,

01:20:22which gives us more speed, makes it usable with Hano,

01:20:27so that you can use Hano for routing and middleware

01:20:29and then have your pages defined in Astro.

01:20:32And that takes me to the next part, content management.

01:20:35You could combine that with database quests, of course,

01:20:38because a database is about storing data and content is just data.

01:20:41But what I mean here with content management is that

01:20:45most web applications or websites have content that's one of two types.

01:20:52You may have user-generated content,

01:20:55and that typically goes into a database, right?

01:20:58So if you're building a social network and the users can create posts,

01:21:01those posts are stored in the database.

01:21:05But you also may have content that's coming from you,

01:21:07the developer or the owner of the site.

01:21:10marketing pages, typically, blog posts.

01:21:13And the big difference there is that for the user-generated content,

01:21:17it's generated dynamically whilst the application is running.

01:21:21That's kind of the main purpose of the application.

01:21:24A social network's main purpose is that people generate posts and stuff like that.

01:21:29The other content, the marketing content, is kind of static.

01:21:33Of course, you also will be adding new content over time.

01:21:36But you as the owner, you obviously know when you're about to add new content, right?

01:21:41And therefore, you can store that marketing content in a different way than the user-generated content.

01:21:47The user-generated content is part of your core product.

01:21:50The marketing content could just be markdown files that are translated to HTML.

01:21:55That is a common approach.

01:21:57That is, for example, what you can do with Astro, right?

01:22:00Astro is a framework that is great at many things,

01:22:04but it's particularly great at content-heavy sites

01:22:07talking about content that's coming from you, the owner.

01:22:11So with Astro, for example, you can easily set up your content, your posts, your articles at markdown files.

01:22:19And Astro has a built-in markdown processor, which is super fast, which translates that into HTML.

01:22:27So Astro would be one suggestion here by me for building the content part or the marketing part of your site.

01:22:34By the way, a bit more advanced, but not uncommon.

01:22:37You can also combine multiple frameworks and servers.

01:22:43You could have one tech stack, one server for your main application.

01:22:47Let's say for your social network, you have maybe Next.js there,

01:22:51and you're using a Postgres database in a Docker container.

01:22:55But then your marketing site, and that site, the main site, is maybe app.yourdomain.com.

01:23:02And then you have your marketing site, which is yourdomain.com without a subdomain.

01:23:06And that may just be an Astro page with the marketing pages.

01:23:10And the login link redirects to your main site, to app.yourdomain.com.

01:23:16So something like that is not uncommon.

01:23:18And Astro is a great choice there.

01:23:20Obviously, you could also build it all in one app.

01:23:22You can also have your markdown pages rendered by your Next.js application.

01:23:28And it doesn't have to be marked down in the first place.

01:23:31You could have raw HTML pages.

01:23:33And of course, you could have your marketing content stored in the database, too.

01:23:37Nothing wrong with that.

01:23:38It's just worth knowing about these options.

01:23:41And using markdown, for example, has the advantage that you can often build those pages in advance

01:23:48at build time when you're making a new deployment, for example.

01:23:52And that simply means they can be served a bit quicker.

01:23:54They don't have to be fetched from the database.

01:23:56They don't hit your database.

01:23:57That saves you some load on the database.

01:23:59So, yeah, that can be reasons for doing it in that more static approach and why that approach is popular.

01:24:06And that's also something to consider.

01:24:08And also consideration, like all these considerations, that still matters in the world of AI and where you should not blindly ask and trust the AI.

01:24:18It's important to understand why certain decisions are made here.

01:24:21And then, of course, you have content management systems like Storyblock.

01:24:28It's maybe not the one most people know.

01:24:32It's just one I used in the past.

01:24:34And the idea here is that your content, your marketing content mostly, lives in a managed service.

01:24:43And they give you a nice editor and an API to which you reach out to get that data.

01:24:49And therefore, you don't have markdown files.

01:24:51You don't have your content in your database.

01:24:53You instead use an extra service.

01:24:54And there are many such content management systems.

01:24:58And we have to differentiate between headless ones like Storyblock.

01:25:02At least it was headless in the past.

01:25:04Where you have an API to load that data into your application to show it there.

01:25:09And, of course, then you have content management systems like WordPress, which do not just give you the data management and an API you can use.

01:25:22But they also give you a frontend.

01:25:23And that's, of course, great if you want to build a website and you don't want to, like, have your own code anywhere or stuff like that.

01:25:33So, yeah.

01:25:34But in general, managing your content in a content management system can be interesting.

01:25:39Personally, I don't do it because I don't see a reason.

01:25:42I just use Astro, markdown files.

01:25:44I'm happy.

01:25:45But, again, depending on your scale, that may be different.

01:25:49I'm working alone or in super small teams.

01:25:51If you're in a large company, using markdown files for content management brings new complexities because suddenly you have to synchronize those changes.

01:25:59You may have approval processes, which you then have to build on your own.

01:26:04Services like StoryBlock give you that.

01:26:06So, again, depends on your scale, on your task, who you are, and all that stuff.

01:26:15Yeah.

01:26:18Max is great.

01:26:19I'm listening to what he says, has to say always.

01:26:21Thank you so, so much.

01:26:22And all the nice comments below.

01:26:24Thank you all so, so much.

01:26:26Really, really nice comments.

01:26:28Mean a lot to me.

01:26:29Thank you very, very much.

01:26:30And I'm happy that this is helpful.

01:26:33Trying my best here to simply, like, brain dump what I think about all of that and what my decision making is here.

01:26:41The software development lifecycle is changing because the cost of writing code is collapsing.

01:26:46The bottleneck is moving from code creation to change, validation, review, and governance.

01:26:51Yeah.

01:26:55The entire software development lifecycle is absolutely changing.

01:26:58And the skill set is changing, too.

01:27:00It is super important, for example, to know about these things to make decisions, which then you can implement together with AI or maybe to large extents by the AI.

01:27:11But you need to be able to make these decisions.

01:27:14And these are just the high-level decisions, right?

01:27:16This is the tech stack.

01:27:18There are many other decisions about the architecture of your software.

01:27:21There are decisions, depending on what you're building, about do I need a queue?

01:27:26Do I need a fan-out pattern?

01:27:28Which provider do I use for managing that queue?

01:27:31Do I manage it on my own?

01:27:32How do I manage it?

01:27:33These are all more advanced questions that also become important, depending on what you're building.

01:27:38If you are sending out emails, the question is, how do you send them out?

01:27:42Do you use something like Resend, which is a great provider?

01:27:48Do you use something like AWS SES?

01:27:52These are all important questions that are on the next layer, I would say.

01:27:58These are the big, not architectural, these are the big tech stack questions.

01:28:04But then we can dive deeper and deeper.

01:28:06And that is precisely also what I want to tackle in my systems design course, amongst other things.

01:28:12How to think about building applications, about trade-offs, about building blocks, which options you have there.

01:28:20Because I believe this is one key aspect of not just web development, but in general of software development in the future.

01:28:28It will still be about the code and being able to understand and evaluate and judge the code.

01:28:37Because at least right now, we're definitely not at a point where AI is flawless, far from it.

01:28:43And I'm not sure if we'll be there anytime soon.

01:28:47So that stays important.

01:28:48You need to be able to make sense of the code, I believe.

01:28:51But you absolutely also need to be able to make decisions about tech stack, software architecture, patterns being used, because they all come with trade-offs.

01:29:02If there simply would be a best solution, it would be easy, of course, yeah.

01:29:07But it would have been easy before AI2 then.

01:29:09But these trade-offs exist for good reasons.

01:29:12You always have trade-offs between scalability, cost, simplicity.

01:29:16And there simply is no best solution.

01:29:18That is why it's so important to be aware of how to make these decisions and which trade-offs exist and which technology and which solution brings which advantages or disadvantages to the table.

01:29:34Do you think all will be cloud, that even our computers won't ever belong to us?

01:29:39No.

01:29:41So for deployment and stuff, obviously, that can't be on your computer.

01:29:45And you'll probably not be building your own data center in your garage.

01:29:48But for development or in general, for computer work, yeah, no, I don't see that going to the cloud.

01:29:55We are definitely offloading more already with AI, right?

01:30:00So we're constantly sending data to the cloud.

01:30:03But still, the development happens on our machine.

01:30:08Our data ultimately is on our machine or in our own networks.

01:30:12And I don't see that changing, at least for me, anytime soon.

01:30:16I love working on my computer and tinkering with it way too much.

01:30:23But you can definitely feel already that if you don't have an internet connection for some reason, there is a lot of stuff you can't do.

01:30:30That is definitely where the computer you have, your laptop, is definitely more of a client than it was maybe 10 years ago.

01:30:42Will I have a Cloudflare course?

01:30:43Is also planned.

01:30:44Yes, is also planned.

01:30:46You just see there are so many course plans.

01:30:48And I am actively working on many of these courses.

01:30:50The Bun and the Pi courses will be out soon.

01:30:54Yeah, so this is all planned.

01:30:56And I mentioned it before.

01:30:58If you want to stay updated about courses, the best way to do that is to sign up for our Academind newsletter on Academind.com slash community.

01:31:05And or join our Discord.

01:31:06We publish all the new courses there.

01:31:08And of course, we have a lot of courses already.

01:31:11So there may be something interesting there.

01:31:14And if you have a membership, our subscription, you get access to all our courses, all new courses included, whenever they come out, as long as you have an active subscription.

01:31:22So that's just a side note.

01:31:28Oh, oh, oh.

01:31:35Frontend design skill is super.

01:31:40By the way, I'm reading all the comments, just not talking about all of them, because some I already picked up on.

01:31:46Some are just opinions, which are valid, but where I don't have a lot to add right now.

01:31:53So, yeah, frontend design skill is super useful now for my project.

01:31:56Is it a good idea to depend on skills?

01:31:58Does CSS only take a lot of time and development?

01:32:02You mean depend on skills on your skills?

01:32:09That was always important and that does not change.

01:32:14And especially with AI, I repeat it over and over again because I strongly believe in it.

01:32:20It's all about steering it, about making decisions, about giving your human input.

01:32:29You can absolutely let AI make all the decisions.

01:32:32That is fine for some utility tool, some throwaway software you're building, but for nothing else.

01:32:39If I'm building something which I'm distributing, which I may be monetizing, that should be my vision, that I want to control how it looks like.

01:32:52I don't tell AI, build me that service, and then I look at it and think, oh, good enough.

01:32:57No.

01:32:57You can do that, and it may work to some degree, but I think you'll always stand out if you have your own influence there, if you bring in your own skills, if you just build a good product.

01:33:11And design is one important part of a good product.

01:33:14And I suck at design and still.

01:33:17That just means I have to spend more time trying to make it at least somewhat good.

01:33:30How should someone go about relearning development in the age of AI and learning AI as well?

01:33:35I don't know where to start and where there's a lot to learn.

01:33:39So about relearning development means you already have some development experience, and therefore I think the best way is just build stuff, build it with AI, so use AI, build it step by step.

01:33:52Don't tell AI, build me a social network.

01:33:54Don't tell AI that.

01:33:56Instead, think about what do you want to build?

01:33:58A clone of some site, as I said before, Amazon clone, whatever.

01:34:02Break it up into the core building blocks.

01:34:04Make your own decisions which authentication solution I want to use, right?

01:34:08So all these decisions here, make them on your own.

01:34:11Spend the time, dive into different options, and don't feel like you have to pick the best one.

01:34:15There is no best one.

01:34:17Just evaluate a couple of options, look at some code, see what feels good to you, and then go with that.

01:34:24And then also, yes, use AI, but evaluate it on your own.

01:34:27And yes, this will be slow.

01:34:29This will take time, but that is how learning works.

01:34:32AI can definitely give you the feeling, and the current time we're living in can give you the feeling that you have to be super fast, you're going to be left behind, but that's all total BS.

01:34:42You have time, because the future will very likely look totally different than we think it looks right now.

01:34:50And nobody can look into the future.

01:34:52And like, if you spent two years ago on mastering prompt engineering, which was more important back then than it is now, or it simply changed.

01:35:03You were not left behind if you did not do that two years ago, because again, it's different now.

01:35:09So take that time, don't rush it.

01:35:11The skills you build will always be valuable.

01:35:14And the worst thing you can do is just not doing anything, unless you just want to quit.

01:35:22But keeping your skills sharp, practicing, playing around with those tools, simply staying up to date, I think is a good idea.

01:35:32And for getting back into development, building those clones and spending the time to make your own decisions is a good approach, I would say.

01:35:40What about Redis?

01:35:41Yeah, Redis, also a good point.

01:35:46Not every application will need it.

01:35:48I would say most applications, when you are getting started, don't need Redis.

01:35:53But of course, Redis, I see mostly as a tool I add when scale becomes important.

01:36:01Because you can solve problems that you can solve better with Redis, also with, even with just stupid SQLite.

01:36:10It will be worse than Redis, but it will be good enough in many situations.

01:36:14But I would absolutely also learn a bit about Redis, understand what it is, its use cases, and how you could add it to an application.

01:36:25Adding it and using it is not difficult.

01:36:26I think it's just overkill for many applications when you're getting started.

01:36:31But having that cache database can 100% help your application scale, become more efficient, more performant once you reach a certain scale.

01:36:44And that's, in general, true, since you mentioned queues here.

01:36:51100% all these building blocks matter a lot.

01:36:55But, again, if you need a queue in your database, for example, to send out email in your application, I mean, you can also build a queue, a persistent queue with SQLite.

01:37:07SQLite, again, will not necessarily be as good as with Redis, definitely has some potential problems, but depending on your scale, may be good enough yet.

01:37:15But, at some point, I would 100% learn about Redis, learn how it works, and consider using it.

01:37:22Do you have thoughts about LibSQL, Terso versus SQLite?

01:37:25Yeah, I talked about it a bit earlier.

01:37:28SQLite is amazing, no matter how you use it.

01:37:31If you use raw SQLite, or if you're using Terso, but Terso LibSQLite, or LibSQL as it's called, is also great, in my opinion.

01:37:43It's essentially a re-implementation, which comes with certain nice features built in.

01:37:49The normal SQLite is missing.

01:37:52That being said, for some reason, maybe just because I'm used to it, I'm mostly using just SQLite in most of my applications.

01:38:00But, yeah, Terso, definitely super, super interesting, extended re-implementation, that is worth the consideration when looking for a database solution.

01:38:27Do you emerge of the product owner and dev roles?

01:38:31Yeah, I mean, a lot of stuff is changing there, right?

01:38:36All these roles will be evolving, and there will be new roles.

01:38:42I think I have a hard time looking into the future here, to be honest, because I don't want to make – I'm already making a lot of predictions or sharing my thoughts simply, but I don't want to make too many predictions about how job roles will evolve.

01:39:01Because, for one, companies, enterprises especially, tend to be slower than one thinks they are.

01:39:09Then again, it's hard to evaluate – I mean, AI is very capable, but it's hard to predict how AI usage will look like in a year or two, especially in those different companies and across different company sizes.

01:39:29But all those roles are, of course, evolving, and as I mentioned before, the skills of developers are changing, in my opinion, the skills you've got to have.

01:39:38The time where you just need to know React really well, and you just spit out React code, those times are gone or are coming to an end.

01:39:53That being said, there may be demand for developers that don't type the code anymore but spend the entire day reviewing it.

01:40:02So, is that fun?

01:40:02I don't know, for me not, but that may be a valid job we see in the future.

01:40:08It may also change such that, as I mentioned before, developers need to become better at architecting software, all developers, not just some of them.

01:40:18And yes, we absolutely may have that merge with product owners, but that is super hard to predict for me, and therefore, I don't want to make any predictions there.

01:40:30Other than that, of course, it is changing.

01:40:35Can you write out your choices?

01:40:37That would take a bit too long, but I don't have those clear winner choices.

01:40:41And the stream will stay online.

01:40:44You can re-watch it after it's done, so that will be no problem.

01:40:49Do you think writing code by hand is obsolete?

01:40:51Yeah, I think it is.

01:40:52Unfortunately, I've shared that before.

01:40:54I liked that.

01:40:55I liked the flows that you could enter there.

01:40:59And I still write some code by hand, but it's mostly fixes, tweaks, or some initial code to have something there to get the agent started.

01:41:09But I think, yeah, writing code by hand on a large scale is over.

01:41:15In most situations, there may be niche scenarios like FMPEG, which is like handwritten assembly, and they may still be doing that because AI may not be good enough at that.

01:41:26But I think in many cases, it is obsolete.

01:41:28Perhaps a course explaining code rules, architecture, code standards.

01:41:36Yeah, I got a lot of ideas there.

01:41:38And impossible to do all of that, but I definitely want to create courses on those fundamentals, on those architecture and systems design decisions, core concepts every developer should know.

01:41:50Basic stuff like understanding, hashing, encryption.

01:41:54I got a lot of ideas there.

01:41:55Time is the biggest issue, as always, but definitely want to create more there.

01:42:04Seems to be a great session.

01:42:05We'll watch from beginning later.

01:42:06Yeah, definitely do.

01:42:07The recording will stay up, as mentioned.

01:42:09So definitely re-watch later or tomorrow or whenever you want.

01:42:13Yeah, let me get back to the list before I then jump back into the chat later.

01:42:30So I think I tackled most things, and I guess the general theme is pretty clear.

01:42:34Now, last but not least, the AI agent part here, it's not just about which agent.

01:42:39I mentioned before that I personally like Pi a lot, and I got a course on that coming up in case I didn't mention it yet.

01:42:46But it doesn't matter in the end, right?

01:42:49So, and obviously, you may even not be using any AI agent at all, just because I think that the future is about AI generating vast amounts of code and about AI-assisted coding.

01:43:06Not Vibe coding for real software, but AI-assisted coding does not mean that everybody has to believe that, of course.

01:43:13And nobody can tell how good these agents and models will get if we're going to hit a wall, because we're definitely not at a point yet where they can do it all on your own, no matter what some people write.

01:43:26But yeah, it will be important, and you can use Cloud Code, you can use Codex, you can use Pi, you can use AMP, you can use the Google tools, whatever the recent one is.

01:43:36I can't keep track of all the tools they're launching and killing, but that is not important.

01:43:43I think the only thing that is important here is that you give it a try, that you give it a try, because I know that some people just aren't convinced by AI.

01:43:54They had a bad experience one or two years ago, and those tools improved for sure.

01:43:59And again, you don't have to use them for everything, but you can definitely get useful stuff done with them, and I would encourage people to give them a try.

01:44:09You will need a subscription, you will want to use the good models, that being said, of course, I also mentioned that before, we now see models like GLM 5.2 coming out, which you can easily consume from various providers via Open Router, for example, which is way cheaper than the Anthropic models.

01:44:27And I think that is the important part here.

01:44:29I'm definitely using AI agents for coding these days, for the most part, and I think it's important to give it a try.

01:44:39And I think it's important, no matter which agent you're using, that you read up on it, or you take a course.

01:44:45Again, I got courses on Cloud Code, I got a course coming up on Pi, but you don't have to take a course, but you should spend a couple of minutes, at least, to understand which features that agent offers.

01:44:57So, for example, Pi has that extensibility as a core feature, the fact that it can build its own extensions to add new capabilities to it, and that it's very, very good.

01:45:09And that makes Pi quite powerful.

01:45:11It does not have a desktop application, for example, Codex has one.

01:45:15If that's important to you, you may want to use that instead, but for Pi, there are open-source solutions you could look into when it comes to that.

01:45:23But, yeah, it's worth learning about it.

01:45:26It's worth taking a look at the settings you can control to really tweak it to your likings, because just like IDEs in the past, and by the way, I still use IDEs or code editors, but differently now.

01:45:41Mostly as div viewers and editors, but anyways, it's important to build an environment you feel comfortable working in, and this environment is, unfortunately, evolving so quickly that it is kind of important, I would say, to stay up-to-date and try out different things there, here and there.

01:46:01Try out a different coding agent, read through the change logs every couple of months to see what changed, and it's also worth noting, that's the other side, to not feel pressured to use everything right now.

01:46:16So, for example, let's take Claude code, they are spitting out new features almost every day, not just in Claude code, but anthropic in general, we got Claude design, we got Claude tag, which is about using Claude from inside Slack, and for Claude code, you got like 10 different ways of controlling it remotely, and I covered them all in my course, but it's super easy to feel overwhelmed, and you don't need to.

01:46:45So, you have to realize many of these features are really just niche features or nice for marketing, the core features are that you use the AI, that you learn how to provide the right context, that you can control as good as possible, that it does the right stuff, and if you then can control the AI from your mobile phone or not, yeah, that may sound fancy, I'll be honest, it's not something I ever do, for example.

01:47:12So, yeah, it can feel overwhelming, and therefore it's important, don't feel overwhelmed, yeah, that I think is important.

01:47:25Well, let's see, what do we have, do you have a jujitsu course, no, I don't, haven't worked with it, sorry, the elk stack, elastic search, also no expert on this, not a lot I can share there, sorry.

01:47:44I want to release my own project, but I'm scared of the responsibility, security attacks during users' data, especially if I plan to add payment.

01:47:51Yeah, oh, payment is a good point, by the way.

01:47:53So, I like Polar for payment, that, or Stripe, of course, but Polar is a merchant of record, which helps with handling sales tax, which may not matter to everybody, but believe me, if you're in Germany, it matters.

01:48:17And it's also a bit easier, and I guess nowadays, even in some situations, cheaper than Stripe.

01:48:23So, that can be great for integrating payments and offloading some of that responsibility.

01:48:29So, Polar is a recommendation I have, for sure.

01:48:32And regarding publishing stuff, just do it.

01:48:36I totally understand that it can feel overwhelming, just, you can use AI for some security audits, doesn't give you 100% protection, of course, but it's a nice extra layer.

01:48:46Make sure you back up your databases so that you can't lose customer data, or at least not too much of it.

01:48:54Definitely use a well-established payment provider, like Polar, where they, or Stripe, where they also store the transactions, so that you can't lose that data, because it's not on your server, it's on theirs.

01:49:06And then just get started.

01:49:09You'll have no users initially anyways.

01:49:11So, the main problem will be getting users and not losing user data or being attacked.

01:49:17Regarding attacks, consider using Cloudflare, not for hosting, but at least as a CDN in front of your service that gives you some DDoS protection, and then just go for it.

01:49:30A lot of things can go wrong.

01:49:33A lot of, very few things do go wrong.

01:49:36So, yeah.

01:49:37These are my two cents there.

01:49:39Yes, systems design course is planned.

01:49:43Converted from Convex to SurrealDB.

01:49:47I heard about that, but have not used that at all, to be honest.

01:49:51So, not a lot I can say about that.

01:49:53Nowadays, we have so many database or managed database offerings.

01:49:59And that side seems to crash my tab here.

01:50:03Not sure.

01:50:04But, yeah.

01:50:04I haven't heard about it yet, or not used it yet, I should say.

01:50:09Hey, Max, I wanted to say thank you.

01:50:10Started web dev about nine years ago from your courses.

01:50:12Yesterday, I was offered a job at Microsoft as a senior full sticker.

01:50:15Wow.

01:50:16That's amazing.

01:50:17Amazing, Darth Vader.

01:50:18Congratulations on the job offer, and thank you so much.

01:50:21I'm super happy that I could be part of the journey and that my courses could help with that.

01:50:25So, thanks so much for the shout-out and for the feedback, and congratulations on your job.

01:50:30Do you think Chinese open-source AI models will win the AI race?

01:50:34Not sure if they will win, but if the Trump administration restricts who can use the top models of their companies, so like Fable 5, it sounds a bit like it's coming back, but it will only be usable by verified U.S. citizens, so not by me, for example.

01:50:53If they go down that route, the Chinese open models will win for sure, because which models am I going to use if I can't use the top U.S. models, right?

01:51:04I'll use the top Chinese models at some point when they are better, and so will many other people in the world.

01:51:11If the Chinese then also lock down their models, I'm in deep shit, because Europe sucks regarding that in some other respects.

01:51:19But, yeah, for now, I think the Chinese can win by just letting the Americans do their thing, at least if they continue restricting model access.

01:51:30But it's too early to tell, to be honest.

01:51:34It's the era of software engineers, not for coders.

01:51:37What do you think?

01:51:40Yeah.

01:51:42The pure code generation days are definitely over, as I said before, I think.

01:51:46So software engineering, on the other hand, is important and will stay important.

01:51:51Cheers for the Insights Max.

01:51:52Always great to listen.

01:51:53Thank you so, so much.

01:51:54And thank you all so much for the nice feedback and for being there.

01:51:58It means a lot to me.

01:52:00Websites will be obsolete because Google search decreasing and AI chats are increasing.

01:52:05Web dev will change.

01:52:06Yeah.

01:52:07So websites will, of course, not be obsolete, I think, because I mentioned way earlier, I don't believe in a future where we'll only interact with services through a chat,

01:52:17because a chat is simply not the best interface for all things you want to do.

01:52:24For a lot of interactions, having a purpose-built interface is better than a chat.

01:52:29If I want to sum up two numbers, being able to just type, or maybe that's a stupid example, but building an Excel sheet is definitely better,

01:52:43if I have a row of numbers I want to add up, than explaining this to an AI model, I would say, just as a stupid example.

01:52:51Building a slide can be better if I care about the positioning, if I am able to manually drag the boxes instead of describing it through a chat.

01:52:59So websites or apps in general, including web apps, will stay important, I believe.

01:53:05It may change, and content is definitely changing.

01:53:08I 100% agree there, because, yeah, Google search volume is decreasing.

01:53:12It's all ending up in those summaries, and the people are not spilling over to the websites, and that's a huge problem also for me.

01:53:19And we'll have to see how that works out in the future, but web apps will continue to exist, I would say.

01:53:26What's my age?

01:53:27You always glow with knowledge.

01:53:28Thank you so, so much.

01:53:29There's a lot of stuff I don't know, obviously.

01:53:33This is just what I work with every day, but I'm 37.

01:53:39Please don't stop the effort to upload videos and do live streams, please.

01:53:42You are amazing.

01:53:42Thank you so, so much.

01:53:44And, yeah, I will keep it going.

01:53:45I will keep on creating courses, videos, also on the Akadamine channel.

01:53:49It's not that.

01:53:51Just need to find more time again.

01:53:53But, yeah, I'll keep on doing that.

01:53:54So, thank you all so much, and I'm not going anywhere.

01:53:58Also not regarding the live streams.

01:54:01As always, I plan on doing live streams every Thursday at this time.

01:54:05I don't make it every Thursday, but I try my best.

01:54:09I'm afraid paywalls will create two worlds, two species of coders.

01:54:12Yeah, definitely.

01:54:14The cost factor and the fact that now you generate code with money and not with time, at least to some degree,

01:54:23is definitely, just as those restrictions I mentioned before, is something that will be interesting to see how it plays out, I'd say.

01:54:39Do you feel that AI code could be a potential license violation issue in the future?

01:54:45Well, not in general.

01:54:47But, obviously, if you use AI to just blindly copy something and people can prove that it, in the end, is a copy, it's a copy, with or without AI, you can definitely run into problems there.

01:55:03Local first is the future.

01:55:06Regarding AI, I mean, I'm a big fan of local models.

01:55:09They're just not there yet, regarding coding, and we'll need some technical breakthroughs for them to be either small enough or local hardware to be good enough.

01:55:18But, yeah, I'm a big fan of local, 100%.

01:55:21Being able to do something offline is amazing.

01:55:30Polar is amazing.

01:55:31Made my first sales ever with it.

01:55:32Congratulations.

01:55:36Thanks for the Flutter course.

01:55:38Thank you for taking it and happy it was helpful.

01:55:48So many nice comments.

01:55:49Thank you all so much.

01:55:51Yeah, last I checked, Polar was much cheaper than Stripe.

01:55:54That's important.

01:55:55Stripe has a lot of hidden fees.

01:55:57They have the sticker fee of 2.9% plus 30 cents or something like this.

01:56:03But then you pay extra for everything else.

01:56:06And suddenly, you're paying like 6% or 7%.

01:56:08And Polar, I think, they recently switched their pricing model.

01:56:12I'm still on the old one for my services.

01:56:14But I think there are, with Polar, you're paying around 4% to 5% or stuff like that.

01:56:23Or maybe even less, depending on which monthly plan you got now.

01:56:26But it often is cheaper than Stripe.

01:56:29That has been my experience, too.

01:56:31Hey, from France, thanks for all the advice.

01:56:35Greetings back to France.

01:56:36And thank you so much.

01:56:38And, yeah, thank you all so much, as mentioned.

01:56:45And actually, I got to go now.

01:56:48So, I will wrap up just answering the last questions here.

01:56:53PostHoc, Datadoc.

01:56:54I don't have a favorite there.

01:56:57Sorry.

01:56:58Because I know all these services.

01:57:00And I'm not really using any of them.

01:57:02So, not a lot I can share there.

01:57:07Which thing encouraged you in the early days to learn web development?

01:57:10I kind of stumbled into it.

01:57:12Back then, I wanted to build a website for Warcraft 3.

01:57:15I liked Warcraft 3.

01:57:17And I wanted to build a website for it.

01:57:18And that is how I kind of got started with HTML and inline styles.

01:57:22And it looked ugly.

01:57:23But these were my first steps back in 2002.

01:57:28I don't know.

01:57:29One.

01:57:31That were the first web development steps.

01:57:39Okay.

01:57:40So, thank you all so, so much for being part.

01:57:44For all the nice comments.

01:57:45For all the opinions.

01:57:46This was an amazing stream.

01:57:47I enjoyed it a lot.

01:57:48I hope it was helpful.

01:57:49The recording will stay up.

01:57:51And I'll be back next Thursday.

01:57:54Yeah.

01:57:54Looking good.

01:57:55Nothing in my calendar.

01:57:56So, I will be back next Thursday.

01:58:00Thank you all.

01:58:02Have a great day, night, evening, morning.

01:58:04Whatever it is for you.

01:58:06And yeah.

01:58:08I'll see you in future streams.

01:58:10And if you don't want to miss anything.

01:58:11As mentioned before.

01:58:12Consider joining the Discord community.

01:58:14Or signing up the newsletter.

01:58:17On Discord, I send out all my streams.

01:58:20And yeah.

01:58:21So, see you in the future, hopefully.

01:58:24Bye-bye.

Key Takeaway

In 2026, the essential developer skill is no longer just writing code but architectural judgment and steering AI to produce maintainable, performant, and secure solutions using tools like Bun, BetterAuth, and SQLite.

Highlights

Understanding the underlying code remains critical in 2026 because AI agents frequently introduce bloated, defensive, or insecure code that requires human oversight.
Framework choices, such as Next.js, React, or Vue, are less critical than previously thought because modern coding agents can be steered to use libraries they were not trained on by providing documentation access.
Building own internal building blocks has become highly feasible due to AI's speed advantage, reducing the need to rely on external library ecosystems and minimizing exposure to supply chain attacks.
Bun offers a performance advantage and built-in features like a SQLite client, making it a viable alternative to Node.js for server-side development.
Authentication can be effectively handled using libraries like BetterAuth, which provides extensive support for OAuth providers and advanced features like two-factor authentication, avoiding the need for expensive managed services.
Modern CSS and browser capabilities have significantly reduced the need for complex UI frameworks, allowing developers to build sophisticated interfaces using vanilla CSS or utility-first approaches like Tailwind.

Timeline

Does the tech stack still matter?

Framework choice is secondary to understanding the product and maintaining the ability to steer AI-generated code.
AI allows for easier exploration of unfamiliar technologies by generating understandable code even if the developer lacks deep expertise.
The emphasis has shifted from relying on library-heavy ecosystems to building custom, lightweight solutions.

While AI can generate entire codebases, relying purely on 'vibe coding' creates risks for commercial products. Developer experience and personal preference for syntax have become less critical, but the ability to read, debug, and optimize AI-produced output is vital to prevent performance degradation and security vulnerabilities.

Runtime and Framework Selection

TypeScript remains the primary language for full-stack web development, while Bun is favored for server-side performance and built-in utilities.
Next.js is the popular default for AI agents, but developers can successfully steer models to use other meta-frameworks like TanStack Start.
The distinction between front-end frameworks like React and full-stack meta-frameworks is crucial for managing interactive client-side and database-connected server-side logic.

Developers often confuse front-end libraries with meta-frameworks. While AI models lean towards React and Next.js, providing documentation to the AI agent enables its use with newer or niche frameworks, ensuring the choice remains one of developer preference and project needs rather than tool capability.

Hosting, Database, and Auth Strategy

SQLite is a highly efficient database for starting new projects, offering simplicity in development and deployment on VPS environments.
Managed hosting platforms like Vercel offer convenience but restrict architectural choices, whereas VPS deployments provide full control at lower costs.
Authentication is manageable to build or use with lightweight, well-maintained libraries like BetterAuth, reducing reliance on expensive managed third-party services.

Choosing a deployment strategy early influences database and infrastructure decisions. Managed services like Vercel or Cloudflare offer ease of use, but VPS setups with tools like Docker provide greater flexibility and cost-effectiveness for long-term growth and data ownership.

Future of Web Development and AI Integration

The software development lifecycle is shifting from code creation to validation, review, and architectural governance.
Web applications remain essential because purpose-built interfaces often outperform generic chat-based interactions for complex tasks.
Effective AI usage involves learning to customize coding agents like Pi through extensions and maintaining human oversight over architectural decisions.

AI significantly collapses the cost of writing boilerplate, moving the bottleneck to human judgment. Success in 2026 relies on mastering system design, managing trade-offs between scalability and simplicity, and knowing when to build custom solutions versus using managed providers.

Community Posts

No posts yet. Be the first to write about this video!

Write about this video