Chatting & AMA and stuff
MMaximilian Schwarzmüller
Computing/SoftwareJob SearchTelecommutingAdult EducationInternet Technology
Transcript
00:00:00[BLANK_AUDIO]
00:00:30>> Okay, a bit spontaneous today, but great to be live again. People are joining, I see. So yeah.
00:00:53Hi, everybody. Welcome to another stream. It's been over a month. It's been over a month.
00:00:59I last streamed in early March. And yeah, it's really good to have another stream,
00:01:08give you an opportunity to ask questions and yeah, just talk about stuff. Yeah, so hi, everybody.
00:01:14Hi, everybody who's joining. Hi, everybody who's saying hi in the chat. And yeah,
00:01:18finally after several weeks. Yeah, I'm very sorry for the long absence. There was a lot of
00:01:25personal stuff and nothing super bad or horrible, but just life getting in the way here and there.
00:01:34Also some work stuff. So yeah, so good to see you all. And thanks so much for the excitement
00:01:42of the chat. It means a lot to me. And I do really enjoy these live streams. So the plan is not to
00:01:49have like a six-week break after this one again. It's just yeah, I try to juggle all the things.
00:01:56I'm not a full-time streamer after all. This is really just something I do for fun whenever I can,
00:02:02typically every Thursday. At this time, I try to stream. But if there is something
00:02:10that gets in the way, yeah, then it is how it is. So yeah, the idea for today really is just
00:02:18to hang out, have a good time, talk about a bunch of stuff. I got some stuff prepared. But yeah,
00:02:25also for you just ask any questions, share anything. Let's just talk. That's the plan.
00:02:31If you want to know something, get any questions. It's really also like a bit of an ask me anything
00:02:39thing here. That's the plan. And yeah. And we got the first question. And by the way,
00:02:49so many nice messages. And I'll try my best to answer to them all. I do read them all.
00:02:54So thank you so much for your work. Thank you so much. Thank you all so much joining. That means
00:03:00a lot to me. If you took some of my courses, I hope you got a lot out of them. If you didn't,
00:03:05that's fine too. If you didn't take them, if you did take them and did nothing, and they didn't
00:03:10work for you, that's not good. Of course, academy.com/courses as a shameless plug, released a
00:03:17bunch of new courses over the last weeks, Codex, Cloud Code, VPS, and of course, all the other
00:03:22courses. So yeah, take a look at that. And got the first question. So let's dive right in. Explain us,
00:03:28Max, what should we do if Copilot will go away? So I guess this is referring to something I also
00:03:37created a video about that GitHub Copilot is moving to usage based billing. And I have an entire video
00:03:44about that. So I will not go through everything I say in that video here again. But the gist of
00:03:51it is that GitHub Copilot is switching away from their subscriptions. I mean, you can still have
00:03:56subscriptions, but you will only get the same amount you're paying for the subscription in
00:04:01AI credits. So I don't really see a benefit in those subscriptions. Maybe I'm missing something,
00:04:07but for me, it looks like a weird prepayment if you stick in those subscriptions. But yeah,
00:04:13they're moving to usage based pricing. The obvious alternative for now, of course, would be Cursor.
00:04:18Cursor has subscriptions. And it is an IDE very similar to VS Code, of course,
00:04:28because it's a fork of VS Code. So Cursor would be the obvious choice for now.
00:04:34Codex, Claude Code are other alternatives or open code black, I guess, though I never used that.
00:04:42But I guess eventually they're all going to move to usage based pricing at some point due to the
00:04:48current situation where compute is super constrained. Inference is getting very expensive
00:04:54for these companies because we're all using so many more tokens due to all the agentic stuff. And
00:05:00again, I have an entire video about that on my YouTube channel. I will not go through everything
00:05:04there. But the alternative to GitHub Copilot right now, I guess, is Cursor, or one of these other
00:05:11plans if you want to use Claude Code and so on. Back at mastery, Node.js or FastAPI. So what's important,
00:05:18of course, FastAPI is Python. So it really comes down to which language you personally prefer
00:05:24working with. Obviously with AI, AI can help you there. But unless you're going down the wipe
00:05:29coding route, which I would not recommend, you should really understand the code that's being
00:05:34generated. So if you prefer Node.js, then use that with any framework of your choice, of course, could
00:05:40be Express, could be a more modern one. If you prefer Python, FastAPI is certainly a great choice.
00:05:46If you want to go down the Node.js route, you also may want to have a look at BUN, of course,
00:05:50as an alternative to Node.js. And then maybe at frameworks like Hono, which I personally like a
00:05:58lot. I'm using a lot of Hono for many projects right now. So that could be interesting too then.
00:06:07Will your new VPS course also be available on Udemy? No, this will be an exclusive course on
00:06:12our own platform. How's the life of a developer in the age of agentic coding, etc. So my personal
00:06:19life, of course, I guess like all our lives, or our professional lives, I should say, has changed
00:06:24quite a bit. I mean, a year ago, two years ago, definitely, I was pretty much writing all the code
00:06:31by hand. Now, I would say 90% of the code is generated by AI or something like that. I do step
00:06:40in here and there. Or what I like to do is also often when I set up a project, or when I flesh out
00:06:46a more complex feature, I define the base types, the base interface, and I do review the code and
00:06:53fix the code and tell the AI or fix it myself if I'm not happy with the direction it's taking. So
00:06:59there's a lot of input there. It's not like you just send the AI off and it does its thing. I'm
00:07:05not talking about vibe coding here where you just go with the vibes and ignore the code. But it has
00:07:10changed a lot. And I mean, as I also shared in a video on my YouTube channel a while ago, a video
00:07:17that did quite well for me personally, that took a lot of fun out of development. And I know there
00:07:24are many people that tell you that it was always about building stuff and that you can have a lot
00:07:29of fun because you can build now with AI and you can build so much more. And this is all true. And
00:07:36I do enjoy that building aspect. Don't get me wrong. I'm not switching profession. I'm not moving away.
00:07:41But the the pure process of typing, of getting into the flow state, of writing code was a lot of fun,
00:07:48a lot of joy for me. And I lost that. And anybody was telling me that this is stupid. Okay, that's
00:07:54fine. But that is my perspective on it. That is gone. And sure, I can still write code by hand.
00:08:01Sure. This is my job. I'm not going to do something that's less efficient. So yeah, that is my
00:08:08perspective on it. It has changed a lot. It is changing a lot. I don't know how my work will look
00:08:13like in a year from now. I absolutely don't believe in a near term future where you don't need software
00:08:19engineers anymore or anything like that. But I don't know which tools will have them, how exactly
00:08:25the models will work in a year or so. So I don't know which new skills may be required to efficiently
00:08:32use these models in a year. Maybe spec files are even more important than maybe code reviews are
00:08:39even more important than I don't know. But it's changing a lot. That was a very long answer. But of
00:08:44course, that is a pretty complex and important topic. And I'm sure we all have our own opinions,
00:08:51and you all have your opinions on that. So definitely also share those. I'm very much
00:08:56looking forward to hearing what people think about that and what your experience with AI has been.
00:09:01Are we returning to coding being a valuable skill again with the per token pricing? That is a very
00:09:08good point. I closed it already. With that usage based pricing of GitHub Copilot, and as I said,
00:09:17I think other providers will switch to such a model at some point in the future too. I do think
00:09:23knowing how to code, knowing how to read code is important anyways, because you want to review the
00:09:29code. But knowing how to write code is definitely also a skill that is not totally unimportant,
00:09:36I'd say. I don't know, mid-term, long-term, at some point the usage pricing will very, very likely go
00:09:45down as there is more supply. But right now we're super constrained and it will stay like this for
00:09:51the near term future by the looks of it. So there may be a lot of use cases where actually you decide
00:10:00that writing something by hand, maybe with AI-powered auto-completion like GitHub Copilot
00:10:06in the old days, just a bit better, that may be more cost-effective and more efficient ultimately
00:10:12than delegating everything to an AI agent. We may very well see such a mixture of approaches.
00:10:20I don't know, but I wouldn't rule it out. Yeah, that is definitely something I could see coming.
00:10:25Best instructor on the internet, thank you so, so much. And also then again, you're the best instructor.
00:10:30Thank you all so, so much for all these amazing comments. Really, really amazing. It means a lot
00:10:34to me. Thank you so, so much. Do you think the move is to use local stuff like LM Studio? Well,
00:10:41I'm a big fan of local AI models, which is why I created a course on it over a year ago already.
00:10:50Obviously what I teach in the course still applies because the idea is the same in that course. I
00:10:54cover O-Lama and LM Studio, and I'm a big fan of these local models and of the tools you can use
00:11:01to run models locally. Not for everything, of course. For agentic coding, for example, in my opinion,
00:11:07what I can run on my machine on an M1 MacBook Pro, which is four years old, is not good enough.
00:11:14So for agentic coding, I can't use local models. But when it comes to analyzing text for basic data
00:11:23analysis tasks, when it comes to generating text and stuff like that, I really like these local
00:11:28models, especially if I'm working with data with text, which I don't want to send to the cloud,
00:11:33which I don't want to send to these providers. So yeah, and I think these local models will become
00:11:38even more important as they get more capable. Who knows, there may be technological breakthroughs
00:11:45that make it easier to run more capable models on cheaper hardware, or these models, the cheaper
00:11:52models become more capable or the fine-tuning makes them more useful for certain things.
00:11:57I could definitely see a future where we have a bunch of small models that have each been
00:12:02fine-tuned for specific tasks and you can therefore run them on relatively affordable hardware or maybe
00:12:09small like a maxed out Mac Studio, which is definitely expensive. But in a company,
00:12:18that could be very useful if you can run models on it that are very good at specific tasks. Suddenly,
00:12:23that data doesn't have to leave your machine or your system and that is worth a lot. So again,
00:12:29a long answer. But yeah, I am a big fan of local models and I think they will play a very important
00:12:34role in the future. What after TypeScript and React? I'm still learning and practicing without
00:12:40AI. I want to get my base knowledge right. And that is a very good approach, by the way,
00:12:45because I get asked this a lot for understandable reasons. What should you learn these days and how
00:12:52would you learn something? Is it still a good idea to learn how to code? And what I think is that
00:12:56things will change. You will very, very likely write less code by hand, maybe at some point
00:13:04no code by hand. But in order to truly understand it, in order to be able to review code and instruct
00:13:12the AI properly, you nonetheless must understand the programming language and the technologies you
00:13:19are working with. I, for example, I couldn't build a program in C++ because sure, I can
00:13:26wipe code something. It might do the job I want it to do, but if it gets more complex or if I would
00:13:33want to distribute it and suddenly security issues matter and other bugs matter, I wouldn't be able to
00:13:38do that because I don't know C++. I could probably read some C++ code because I know how to write
00:13:45code in general, but I don't know the best practices, the patterns and so on. So learning
00:13:50how to code, in my opinion, is still important. And I'm not just saying that because I sell courses
00:13:54on that topic. I truly think that is important. So back to the question, what after TypeScript and
00:14:00React? I would say build stuff, build stuff, build stuff, maybe with a bit of AI, but make sure you
00:14:06understand the code it generates. Jump in here and there and build demo projects. Another important or
00:14:12useful step thereafter could be to learn something like Next.js or a 10-stack start, which are meta
00:14:17frameworks. Whoops, what am I doing here? Which are meta frameworks for React. So that could all
00:14:29be good next steps after learning React and TypeScript. Build stuff and maybe dive into
00:14:34those meta frameworks or dive into React Native if you want to build mobile apps. Also a good idea.
00:14:40Do we have any material or courses on Ruby or Ruby on Rails? No, because I'm absolutely not into Ruby
00:14:46or Ruby on Rails development. I did briefly learn it like, I don't know, 12 years, 10 years ago,
00:14:53but yeah, I haven't used it since and I'm absolutely no expert. What language is wisest,
00:15:00not necessarily best to focus on in 2026? Don't say it depends, all of them, or if you do,
00:15:06pick your favorite. So I won't say it depends because, well, obviously it depends, but I think
00:15:13the wisest language to learn is a language where AI is really good at. Now, the modern models and
00:15:22tools in which they run are fine-tuned such that they are, in my experience, decent with most
00:15:29languages. If you give them documentation, the right skills, but out of the box, they're really
00:15:35good at TypeScript, Python. So I would use one of those probably. And because I'm totally in the
00:15:44JavaScript world, I would recommend TypeScript, JavaScript in the end, but TypeScript is good
00:15:50because the types help the AI because that has been my experience. AI is really good at that.
00:15:55Not perfect, far from perfect, but pretty good. So that is the language I would recommend.
00:16:00You're an absolute legend, Max. Thank you so, so much. You're all too kind. Thank you so,
00:16:07so much. Hi, Max. I'm a QA engineer with AI in the picture. I am working more than ever
00:16:13since the expectations almost tripled. I do not think I'm writing test automation anymore.
00:16:18I'm just babysitting AI. Yeah. And I mean, that is my experience too, right? So again, that no joy
00:16:27part here. It's shifting. It's shifting. And of course you can build more if you are in a position
00:16:36where you can build stuff. If you were saying you're a QA engineer, that means you don't
00:16:40necessarily build more. You just get more stuff to test and to review. And that obviously is not a lot
00:16:48of fun also because we have an asymmetry problem. AI can generate an infinite amount of stuff.
00:17:01I'll move it over here. No worries. So AI can generate an infinite amount of stuff
00:17:06and stuff is something like code and so on. But you as a human, you have a limited amount of time
00:17:15to review stuff, right? So let me zoom out a bit. So that is the problem, of course. And that is that
00:17:25asymmetry. We have AI pumping out lots of code, but then we have to review or test that. And that's
00:17:34exactly what you mentioned. This babysitting part is exactly what I'm experiencing myself too.
00:17:38I am in a position where I also build the stuff. I don't just get stuff to review,
00:17:43but of course, I'm kind of capped by the amount I can review. And my job has definitely changed from
00:17:50getting into that flow state and writing the code to coming up with specs, reviewing specs,
00:17:57reviewing code, and so on. So it's a lot of babysitting. And again, I totally get that that is
00:18:03a new role and something to adjust to. And we're still at that point where we're all
00:18:10overwhelmed with AI slop and AI code. And I think that things will adjust and we will probably get
00:18:18tools that help us get through the AI slop quicker. AI power tools too, of course. And we'll learn how
00:18:25to be more efficient with AI if not for the single reason that the price will go up with usage-based
00:18:32pricing. So there may be less slop for a time period. But yeah, I totally get what you mean.
00:18:40What do you think about .NET development in 2026? Not a lot I can share there because I'm not a .NET
00:18:46developer. Sorry. If we have three years of experience in front-end and mobile,
00:18:50and if we want to switch to something else like AI/ML, should we? By the way, no job after layoff.
00:18:57If we want to switch to something else like AI/ML, should we? It's a really difficult question
00:19:05because, of course, I'm not in the AI/ML field. So I don't work in machine learning. I don't work in
00:19:11training AI models. And I'm not sure if the demand for that is so, so high because, of course, AI is
00:19:19all the rage right now. But I wouldn't say that this means that everybody's looking for people
00:19:26that can build AI models. I think companies will be looking for people that can leverage and use
00:19:31these AI models and tools, be that for development or be that for anything else. I mean, all processes,
00:19:39all things that happen in companies will probably be changed by AI or at least
00:19:45CEOs, managers will try to change them, to put it like this. So if you're switching professions,
00:19:54I would look at what can and do you want to do, and the want part is important too, that helps companies
00:20:03make that transition. And that could still be in a development role, just not as a traditional
00:20:09developer. That can be anything related to consulting or stuff like that, obviously.
00:20:15And that can, of course, also be AI/ML, but I wouldn't necessarily focus on becoming an ML expert
00:20:24or becoming an expert in training AI models because, of course, that also doesn't happen overnight.
00:20:29And I wouldn't guarantee that there is an infinite amount of demand for people in that area.
00:20:36I may be wrong, of course, but that is my feeling or my rationale here.
00:20:41So what I'm seeing now is we don't have to retrain as plumbers because knowing how to code will reduce
00:20:48costs. It will reduce costs and it will matter because somebody has to keep the AI slop in check.
00:20:54Make sure that you can really use these AI models and tools to get an efficiency boost
00:21:00because just using them for a vibe coding alone will help certain people in certain use cases,
00:21:07definitely, but it's not what we should collectively strive towards because we can already see that a
00:21:15lot of software is getting worse. We all see that and we don't know if that is because of AI or if
00:21:21it's just random or whatever, but AI is definitely not the way to magically make software better or to
00:21:28produce good software without steering, without reviewing. So yeah, knowing how to code is still
00:21:34valuable. How we code is just changing. I've stopped trying to keep up with AI and coding,
00:21:42hobbyist coder only, and I don't envy you guys out in the marketplace. Yeah, I mean, a lot is changing
00:21:47there, but to some degree, I sometimes think of what was going on in 2019 and so on with JavaScript
00:21:55frameworks and you could feel overwhelmed there, too. And I've never been a fan of people that tell
00:22:01you that you got to do something now or you'll be left behind because the truth is it's all changing
00:22:08so quickly. You're not left behind if you're not up to date all the time. I mean, a year ago or is
00:22:14it two years? I don't even know. MCP was all the rage. And sure, MCP is not gone. MCP has probably
00:22:22its purpose, but I've never really used MCP servers. Only here and there to have an MCP server that looks
00:22:30up documentation. And that may be different for other people, but you're not behind because you're
00:22:35not jumping onto every new tool, every new approach, every new shiny new model. You don't have to switch
00:22:43models every week. It's easy to feel overwhelmed right now. And it does happen to me, too. But
00:22:48there really is no reason if you think about it. It's changing anyway. So yeah, you can take a month
00:22:56off, come back, spend half a day reading Twitter, some blog posts, ask chatgpt what happened over
00:23:04the last month. And you're up to date. You're up to speed again. So yeah, we should probably all
00:23:09relax a bit. What kind of AI power tools or developer products do you think are still missing
00:23:16today? Well, if I had a good answer to that, I would try to build them myself. I think the whole
00:23:23memory thing is still not totally figured out yet. I also think that many tools, we already see that
00:23:31many tools, many services are releasing CLIs so that agents can better work with them. And I think
00:23:37we'll see more of that. And any service, any software you see that doesn't have a good API or
00:23:47a good CLI tool, and they're not working on it either, that may be an interesting opportunity to
00:23:52disrupt or to build an alternative to. Because I do believe, I'm not sure if we'll have a future where
00:23:59it's AI agents only, but I do believe that giving agents a way of using your service, your software
00:24:08is a good idea. And if any service or software is not doing that, that may be an opportunity,
00:24:13I think. And obviously there will always be exceptions, but you get my point.
00:24:17Do you feel like you are learning even more now with LLMs? Yeah, the knowledge is getting way
00:24:25broader. It's way easier to jump into certain topics because you got an infinitely patient
00:24:31mentor where you can ask all your dumb questions. And I have a lot of dumb questions about a lot
00:24:38of stuff I don't know. So you're definitely learning more, but it's of course a very shallow knowledge
00:24:44and you forget it quickly, especially if you're just looking for a quick answer. So I find it very
00:24:49important to also go deep here and there, which is why I personally enjoy reading documentation,
00:24:54building stuff myself. And if something is truly important to me, if I want to really
00:24:59get better at something, I dive deep. But yeah, you can learn way, way more. It's just not a deep
00:25:04learning out of the box, at least not for me. I hear that a lot that developers will still be
00:25:10needed, but how does that go with all the layoffs? That's a very good point. So a couple of things
00:25:17here with the layoffs. There is this chart, which I like a lot, which of course is also,
00:25:24it's just one chart. There is more to it. And this chart of course looks horrible, right? I mean,
00:25:29we were here during the pandemic, now we're here. But in this chart, you can see of course that
00:25:35software developer jobs in the United States on Indeed, so obviously not the entire world,
00:25:42have been rising steadily for the last couple of months. Now, does this mean that everything
00:25:50is good and we're going back to here? No, we won't do that because this also wasn't normal,
00:25:56right? We were coming from here, whoops, from here. So I think it's possible that we go back to here.
00:26:02And of course, that doesn't help everybody and everyone and so on. But I think these layoffs,
00:26:11which we're seeing, there often is the argument that this is because of AI and the companies
00:26:17themselves will say that. And that may be part of the truth, but it still also is the case that all
00:26:24these companies massively overhired during the pandemic and they're still laying off people from
00:26:31back then. And AI is of course a nice reason for these layoffs, because that also sounds good for
00:26:36the investors. So these mass layoffs we're seeing, I don't doubt that AI is a reason, but it's not the
00:26:45only reason. And I think that with AI, more and more companies will realize that they can unlock
00:26:52more potential with AI if they don't lay off everybody, because they need those people to use
00:26:57the AI. We don't know what will happen in 10 or 20 years if it's AI agents only then, but right now,
00:27:03it's definitely not. And I don't doubt that many companies would prefer to only have senior
00:27:09developers that use AI, but they will realize that senior developers don't grow on trees and
00:27:15that they need the juniors in order to become the seniors. So that's why I'm not too pessimistic
00:27:21about that for the near-term or mid-term future. But obviously, I can't look into the future,
00:27:28but these are just my thoughts on that. Which AI tool do you recommend investing in for coding?
00:27:34I switch a lot because I also create courses on them and I want to stay up to date with how these
00:27:39tools work, which features they offer. But what I personally like a lot right now at the moment
00:27:44is I'm using a lot the Pi agent, the Pi agent, Pi coding agent, which you can use with any API key,
00:27:54you can use any model with API keys, or you can use your codex subscription there, which is what
00:28:00I'm doing. And if I'm not using that, I'm using a lot of codex right now, but I also constantly
00:28:04come back to Cloud Code to check how that is. And I like all these tools. I have no clear favorite,
00:28:09but Pi is, if I had a favorite, it's probably Pi right now. And ironically, I don't even have a
00:28:15course on that, though I will probably create one because I really like it. But yeah, I switch
00:28:20constantly, as I said. I went back to the very beginnings of AI development. I write the code
00:28:29myself and let AI generate only what I don't want to write. Yeah. And I think that is not a horrible
00:28:35approach. And again, we're all still figuring out how to best deal with AI. And it's super easy to
00:28:41fall into the trap to suddenly be at a point where you have to ask AI to make a button red because you
00:28:48don't even know, or you're too lazy to look where in the code that color is set up. I mean, it's a
00:28:53one line change, but super easy to fall into that trap. And that is not good either. So you want to
00:28:59find the balance. And we're still, all or most of us, I guess, I definitely am still in that
00:29:05position where I try to find my best way of working with AI efficiently.
00:29:12And again, because everything's changing all the time, the models are changing, the tools are
00:29:17changing. I think that that state will continue to be for the coming months and year or so.
00:29:25JetBrains has the token based pricing a long time and the users can use nearly any AI. That's great
00:29:33stuff. But Claude, for example, consumes 10x more tokens than Gemini with no better results.
00:29:38Yeah. I mean, everybody will have their own favorites. I personally haven't used
00:29:42JetBrain, so I can't say anything about that. I shared my favorites just a second ago. But yeah,
00:29:49and I think, as I said, the usage based pricing thing will come sooner or later for all products
00:29:55anyways. Just passing to say, thank you so much, Alfredo. Thank you so, so much.
00:30:07Why the job market is down in Germany and will it get better soon? It's very hard to find a job.
00:30:12It's too competitive in little positions. So I'm not searching for a job. I'm not looking for a
00:30:17job in Germany. So I can't say too much about the job market. I mean, Germany has a lot of problems.
00:30:23And I would argue AI is the least of the problems we have here. But I don't want to get too political.
00:30:32I think there are many incentives in Germany that go both against working and against hiring people.
00:30:39And that is, of course, a horrible situation. So combine that with the overhiring during COVID,
00:30:45and you have the current German job market. Since Germany is super slow related to everything
00:30:52related to technology and in general innovation these days, AI is not the main reason, I would say.
00:31:01I've taken your Next.js and TypeScript courses. Thank you for the great teaching.
00:31:04Thank you. I'm super happy that you liked the courses. I hope you got a lot out of them.
00:31:09Thank you so, so much. Hey Max, what do you think about Spring Boot framework, which is the most
00:31:16used web framework in 500 fortunes companies? Why did you never release a course about Spring Boot?
00:31:22Simply because I don't work with it. I have no doubt that it's used a lot, that it's great,
00:31:26that it's popular. And so are many things. I'm just not a Spring Boot or Java developer or anything
00:31:32like that. And I don't teach stuff I don't know, which sounds like a good strategy. So yeah.
00:31:39I know there's a lot of popular stuff. And now with AI, things are changing a bit. Coding courses
00:31:48don't sell that well anymore. But six years ago, I could have created courses on all kinds of
00:31:53technologies that were popular. But that was never my philosophy. I want to share and teach
00:31:59what I use every day, what I know really well. And yeah, so there will likely never be a Spring Boot
00:32:05course by me. Hi, been doing your Go course on Udemy. Love it. The way you teach. That being said,
00:32:11I have a question future of Golang and what should be the path forward after completing your course?
00:32:16The path forward, I repeat myself here, but the path forward is always to build stuff,
00:32:21build a web application with any framework like the GIN framework, for example, build a CLI.
00:32:27Not a bad idea in the days of AI. And of course, learn how to leverage AI with Go. What's great
00:32:33about Go and why I think it has a good future is that whilst it's not the most loved language out
00:32:41there, because some people don't like its syntax and so on, AI is really good at it because it's
00:32:49typing system and the way it works, just like TypeScript, for example, make it a good candidate
00:32:56for AI use. And unlike TypeScript, its type system is less verbose, also less powerful,
00:33:02but less verbose. And therefore, in my experience, AI is really good at Go and therefore I think the
00:33:08future of Go doesn't look too bad. As I've taken your Next.js TypeScript courses, thank you for
00:33:18the great teaching. Thank you. Thank you very much. What strategies would you recommend for developers
00:33:23in regions like mine to successfully find and secure remote jobs? Now, I don't know, of course,
00:33:29exactly where you live, but if it comes to finding and securing remote jobs, you, of course, obviously,
00:33:39you need visibility. And how do you get visibility? Well, I think, as bad as that may sound, one good
00:33:48way is social media presence. And with that, I don't mean dance videos on TikTok. Obviously,
00:33:53I mean something like a YouTube channel where you explain stuff, where you share knowledge,
00:33:58stuff like that, or where you build demo projects with or without AI, whatever you like. Anything
00:34:04like that doesn't have to be YouTube, could be Instagram, could be TikTok, just not the dance
00:34:09video stuff. I think that can be one important building block in getting visibility these days,
00:34:15because social media is so important. And I know you don't want to get a social media star,
00:34:19you want to get a job, but that is one thing that I could imagine works, because not a lot of people
00:34:26do it or are good at it. And that is, therefore, something where you can try to get at least decent
00:34:30enough to get some audience, some visibility, and that could help with finding remote jobs and
00:34:35getting that visibility. Morning, I also like telling AI to explain to me what it's writing,
00:34:44kind of obvious, but telling it to break it down to you while it builds is kind of awesome. Yeah,
00:34:50that is true. I also like to challenge AI to explain what it just did or why it did something
00:34:55or confront it with an alternative I had in mind. And I think that is one underrated thing you can
00:35:02do with AI. We maybe used to do it more two years ago, but you still, you can use AI to have
00:35:10discussions, to discuss different approaches to solving a certain problem, different implementations,
00:35:18stuff like that. And I think that is also something where you as a developer can learn and grow. You
00:35:25shouldn't take everything the AI tells you as the truth, it definitely isn't, but it can be a useful
00:35:33discussion because you're talking with someone that just generates tokens, but that generates tokens
00:35:41based on a knowledge base and weights that has seen a vast amount of blog posts and Stack Overflow
00:35:47posts and code bases and so on. So I think talking to AI about your code, discussing alternatives,
00:35:56is a good idea. No AI agent can handle C++, it's just not possible. Yeah, that may be the case.
00:36:03Sorry, not a question, but I think the only variable we have as a dev is how much we know
00:36:08whether there is cheap tokens or not. We can't really control that. Yeah, obviously what you know,
00:36:13what your experience is, opinions you have, that is the value you can add as a developer.
00:36:19Do we have any material on Ruby on Rails? No, I don't. As I said before, I never really worked with
00:36:24Ruby on Rails. I'm a bit late here because I'm going through all the questions chronologically,
00:36:28so I will get to your question too if you just ask one. It just takes a time. I just go through them
00:36:33as they arrived. Do you know Rust Max? I do. I learned Rust, the basics of Rust, and worked a
00:36:43bit with Rust two years ago, but haven't used it since, so it's pretty much all gone again. I liked
00:36:49it. I liked Rust a lot, but I got so much work to do right now, so much stuff going on that I
00:36:58just don't find the time to learn Rust seriously right now. But I would love to get back into it.
00:37:04Definitely a polarizing language, I know, but I personally liked it. What kind of AI-powered tools
00:37:11or developer products do you think are still missing today? Okay, I already answered that.
00:37:16Go is great at readability, so generated code is a lot easier to read. The package,
00:37:20GoFundMe, lets you communicate between the proprietary code of the GPU and go without CEO.
00:37:25Yeah, as I said, Go is really great for humans and agents, as I said, in my opinion.
00:37:32Hi Max, thanks for your great work and content. From a front-end perspective,
00:37:36what's the best way to start getting clients and earning through platforms like Upwork or Fiverr?
00:37:41Yeah, basically what I said before, you want to get some visibility to kind of stand out of the
00:37:46crowd. Obviously on these platforms you then also can build up a client base and ratings and stuff,
00:37:52but anything that sets you apart from the masses is a good idea. And that's why I mentioned the
00:38:01social media stuff. Looking ahead, which path do you think offers more opportunities, full stack
00:38:06development or DevOps? I mean, you could say DevOps because you could argue that AI may,
00:38:20that it's harder for AI to be as good there as companies or people want it to be,
00:38:26to reliably let it do all the DevOps stuff. I mean, for DevOps you probably want to have more
00:38:34humans which at least are responsible if the server goes down, which is not the right take on it,
00:38:41by the way. I think you should care about software quality equally. It just feels to me that many
00:38:46companies think that, yeah, we can patch software problems, but the server is not allowed to go down.
00:38:51Anyways, that's just my thoughts. So maybe DevOps would be my answer here, but personally I think
00:39:00nobody knows what AI will be able or not be able to do in 10 years or so. But right now I think
00:39:07realistically both are good choices, but DevOps is probably the one where more people would tell
00:39:13you that it's the better choice. So yeah, DevOps may be the thing where near-term companies are
00:39:26more likely to hire. Yeah, I think that's what I think about this. I'm starting a company and
00:39:34I'm finding there is no need to use third-party tools since the LLM can just build it really quick.
00:39:41Question is when should I stop building all these tools? Yeah, really only when you hit a wall or
00:39:48once you're dealing with something where messing up has severe consequences. So for example,
00:39:53in Germany you don't want to mess with the taxman. So you don't want to wipe code your own solutions
00:40:00there. At least I don't. But we also have replaced certain tools in our company with our own more or
00:40:08less wipe coded alternatives. I can't change. It's not fully wipe coded. I had to look at the code and
00:40:17tweak it here and there. But yeah, we have internal tools where we're not using external providers
00:40:25anymore. Not a lot because we never used a lot of tools to begin with, to be honest. But especially
00:40:32when it comes to image editing stuff and so on, we're using a lot of hacked together tooling that
00:40:40gets the job done for our use cases. And if you have no reason to switch, don't do it. That would
00:40:46be my recommendation. Unless if something goes wrong, it's a big problem for you. Then you want
00:40:52to go for a professional established solution. Would you not like to create a course about
00:41:00patterns and architecture? Let's say you would make a complete course where you will cover how
00:41:04to architect an app, what patterns to use in React or Node, what gateway architecture is important.
00:41:10I would love to do that. And I definitely plan to do that. I plan to create a systems design course
00:41:15this year. Again, just a lot of stuff going on in work, but that is definitely something I plan
00:41:20to do. Because yeah, I would love to share more knowledge on how to architect software,
00:41:26which decisions to make, about building blocks in software, talk about caches and queues and all that
00:41:32nerdy stuff. So yeah, that is something I have on my shortlist for courses. Great to see you live.
00:41:40Thanks for all the courses on React and JavaScript. They helped me a lot. Thank you all for the nice
00:41:49messages and thank you very much. My company has been pushing for more AI lately and it's shocking
00:41:54how much the code quality in UI/UX has degraded. Yeah. And I mean, we can all see it. I said it
00:42:00before. So many software products feel worse these days. And it's hard to imagine that AI plays no
00:42:08role in that. It may not only be AI, but yeah, software quality seems to be degrading because
00:42:16there is this huge incentive or this pressure many companies are feeling to push out as much code,
00:42:23as many app updates as possible. But there is no pressure to have a high quality, at least
00:42:31that's how it seems to me. So yeah, I think your company is not alone. And I think that will change.
00:42:37And I think indeed it's a good way of standing out and differentiating from competitors right now. If
00:42:46you have a high quality product, if you value quality, I think that will become a comparative
00:42:52advantage in the near future or it already is. But right now, most companies as it seems are going
00:42:57into the quantity direction. Great to see you. Thank you for your work. I wonder to ask if there
00:43:05are any startups to participate in where it would be possible to train Java skills. Well, I'm not
00:43:10really part of the startup culture or a senior in Germany. So unfortunately I can't give any concrete
00:43:18recommendations there. And I don't know how many startups use Java because while Java is popular,
00:43:26and I personally think it's a nice language, my feeling is that it's not the most hyped up language
00:43:32in startups. It's more like a language used a lot in big enterprises, established companies.
00:43:38So I don't think if Java is a language you can train very well when joining startups. I may be wrong,
00:43:47but that is my feeling here. What about codecs?
00:43:57So thanks for your perspective on the layoffs and Pi is great. Thank you so much. What about codecs?
00:44:05What pros and cons do you see? So I think we have to differentiate of course between the codecs,
00:44:13the model and the tool. What I like a lot about codecs is the desktop app, for example. Obviously
00:44:20some people hate it or don't want to use a desktop app. Some people don't want to use a terminal user
00:44:24interface. I like the desktop app. It's not like I use it all the time. As I said, right now I'm using
00:44:29a lot of Pi with my codec subscription. And about the subscription, I really like the model, how it's
00:44:37been fine-tuned and how it works. Not perfect. Just want to emphasize that. But yeah, I like it.
00:44:43I also like Claude Opus though. I'm not in the camp of "Oh, this is horrible. This is great." I
00:44:51like them both, but maybe the GPD 5.5 model specifically a bit more. Now when it comes to
00:44:56the codecs app, I like the desktop app, as I said, and I like the speed and performance of their
00:45:03terminal user interface. And the disadvantages, of course, are... Well, if you don't like the CLI,
00:45:14if you don't like working in CLIs in general, the performance is not worth anything to you.
00:45:19Same, of course, for the desktop app. And in general, they all have a disadvantage compared
00:45:25to Pi, for example, in my opinion, because what I like about Pi so much is that it's super
00:45:32extensible and that it can improve itself. And codecs and Claude code, they can't do it to the
00:45:39same extent. Pi has only a minimal amount of tools built in, and you can ask it to build extensions
00:45:47for itself, and it's all very easy and straightforward. And that is an advantage
00:45:53Pi has and a disadvantage codecs has, for example, just like Claude code, in my opinion.
00:46:00I, Max, kind of know how old you are. Yeah, I'm 37. How do you deal with unnecessary or redundant
00:46:08code generated by AI, and is there any way to prevent it? Well, you tell me. AI,
00:46:14and no matter if I'm using GPD 5.5 or Opus, though maybe GPD 5.5 is a bit worse than it,
00:46:24AI loves to generate stuff I haven't asked it for. And in my experience, you can't prompt this away.
00:46:32You can tell it all day long that it should stick to the plan, that it should not overcomplicate.
00:46:40And still, it likes to add those helper functions. It likes to add stuff I haven't asked for.
00:46:46And yeah, for me, maybe I just suck at prompting, but for me, the only recipe against that is
00:46:54code review. So I take a look at the code and I clean up the code and I ask it to remove stuff
00:47:00I didn't ask for, if I'm too lazy to clean it up myself or if it's too distributed across files.
00:47:07I haven't found a good recipe for avoiding these models sneaking extra code into your code base,
00:47:17other than asking for small changes. So if you have a very, very small isolated change, in my experience,
00:47:26that typically works. But as soon as you have a slightly bigger task, even if it's not a huge one,
00:47:32these AI models really like introducing extra complexity in code. That has been my experience
00:47:38at least. What is your go-to coding AI? Do you mainly use Clod or do you bounce around? I bounce
00:47:44around. Right now, it's a lot of GPT 5.5, but then I go back to Opus 4.7 in Clod code. But it's mostly
00:47:52these two. I haven't really used a lot of other models in the last months. Rust or Go, which
00:47:59language do you prefer? Well, I said before, I learned a bit of Rust two years ago or so.
00:48:07So personally, I know Go way, way better. So I prefer Go. That may change if I had more time
00:48:15to spend with Rust and to learn Rust a bit better, but that's not going to happen anytime soon. So for
00:48:21now, it's Go. Do you know if there's an alternative for Super Maven, feeling kind of overwhelming with
00:48:26all those ancient stuff and want to write code again? Not my area of expertise, unfortunately.
00:48:35I have to say. So not something I have used. So not a lot I can share there. Don't really want to
00:48:47share anything about something I have not worked with. Sorry.
00:48:52Do you have any thoughts on using Langchain or building CLIs for legacy APIs or Langchain in
00:49:04general? So building CLIs is something that, at least for now, is very, very important and a good
00:49:12idea, I think. And I personally did it a lot. I built a lot of internal CLIs for managing our
00:49:20courses and stuff like that so that I can ask an agent to do something there through the CLI,
00:49:25through an API. So I think that is very valuable right now. It may change, but right now it's
00:49:29valuable. Langchain, I played around with it. I used it a bit three years ago and then sometimes
00:49:37here and there in between, but I'm not using it right now. And I think with what I was using
00:49:44Langchain for was to build applications that used AI or tools that used AI or basic agents.
00:49:51And right now there are so many SDKs you can use and so many libraries you can use for building
00:49:58agents. And there's a new one every day that, yeah, I haven't even had close to the time to try
00:50:06them out. And it's not that hard to build your own agent software without a library or your
00:50:15own agent without a library to begin with, especially since you can't ask AI for it and
00:50:20then just fine tune it or steer it in the right direction. So yeah, I haven't used Langchain in
00:50:26quite a while. If only senior devs are going to be needed in a post-AI future, how do you think
00:50:31education should adjust itself for new developers? Well, I think companies would want to only hire
00:50:38senior devs, but obviously you need junior devs first. Now, what I think with education,
00:50:42and obviously I'm asking myself the question a lot, how will education change? Because that is
00:50:48what I've been doing for the last 10 years and what I would like to continue to do. And I think
00:50:56we have to, we have to, it's now easier to get a broad knowledge. As I said before, it's easier
00:51:07to learn a bit about a lot of topics. And I think the role of education will be to steer that process
00:51:15to make it even more efficient for people to quickly gain a broad knowledge so that they can
00:51:20then ask AI if they want to dive deeper. And at the same time, to have these structured deep dives
00:51:28where you can give that extra depth to that knowledge so that people can quickly get an
00:51:37overview of something also with help of AI, but can then dive deeper so that the process from
00:51:43junior to senior is shortened. The time is shortened because of course, in the past where
00:51:49we wrote code by hand. That, or to put it in a different way, we're not really writing all the
00:52:00code by hand anymore. So to go from junior to senior, you can gather more experience in a shorter
00:52:07time likely if you have the willingness to learn, also with help of AI, but if you're willing to not
00:52:14blindly trust AI. And that is where education can come in and can give you these learnings,
00:52:20can show you how to use AI to work with technology A or library B. And you can soak that up and use
00:52:29that in your own projects where you're also using AI, but you don't have to go through the process
00:52:34of learning the entire React syntax or anything like that. So you just need to be able to read it
00:52:38and learning how to read something is easier than learning how to write it. You need to learn how to
00:52:43evaluate it and that will take longer than learning how to read it. But it is also something where
00:52:48education combined with AI can help you. And I think that is how we need to adjust it to make
00:52:53it easier for you to move from junior to senior level to shorten that time span. Because AI is
00:52:59giving us that speed boost and education kind of needs to fill the knowledge gaps that appear
00:53:06because of that speed boost. I hope it's at least somewhat clear what I meant there. It was probably
00:53:12not the best way of explaining it, to be fair. But as I said, I'm still also trying to figure
00:53:18it out myself what the best approach here is. What do you think about Solidity, Rust, Web3,
00:53:24blockchain development? Is it interesting to learn it? Are there enough job opportunities? And do you
00:53:29plan to release a Solidity course? That was never my cup of coffee to or whatever you say, to be fair.
00:53:36I'm not super deep into blockchain. I found Bitcoin and the early days of blockchain, I found that
00:53:43or not the early days, but I found that interesting in like 2016-17. And I spent some time learning
00:53:50more about it there. And I like the idea behind it in general, but I was never on board with the Web3
00:53:56stuff. I was never on board with NFTs and so on. So yeah, I don't have a lot to share there,
00:54:02to be honest. Hi Max, big fan. Can you tell me what is the overall future of development in AI world?
00:54:08I'm talking about the agentic code. It only do the CRUD landing page work, but the future can do complex
00:54:14as well. I think AI can already do complex stuff as well, but maybe unlike with the CRUD stuff,
00:54:28it needs steering and it needs reviewing. And as I said before, I don't think that will go away
00:54:36anytime soon. And I know Dario Amodei, the CEO of Anthropic, is telling us that we're all going to
00:54:42lose our jobs in the next one to five years. And I obviously also can't look into the future,
00:54:46but from how these models work today and the code I get out of them and how I have to use them to
00:54:54at least try to be productive with them, that is still far away from just letting them go. So I
00:55:03think they can do complex stuff, but they need steering, they need reviews. And I think that
00:55:10will stay that way for the near future. And I don't know what's going on in five or 10 years,
00:55:17obviously, but that's my take on it right now. Have anyone ever told you you look a bit like Seth
00:55:27Rogan? Yeah, I heard that a lot. Actually, it was way more frequent like five years ago,
00:55:33probably because I got older, but I have heard it before. I don't know if that's good or bad for him.
00:55:41How do you structure your day so you can have time for a latnoff? What is a latnoff?
00:55:47Oh, I don't get it. I'm happy to answer it if you explain to me what a latnoff is.
00:56:01Hi Max, as a beginner wanting to build a SaaS, what language framework would you choose today?
00:56:07Something that is good for SEO, but also providing a good development experience is important.
00:56:11Yeah, what is that before? I would recommend TypeScript because AI is good at it and because
00:56:16that is my stack, so I know that it's good. Framework, React, Next.js or TanStack Start
00:56:23give you server rendering, which is important for SEO. And yeah, I guess that's the short answer.
00:56:28That would be my recommendation. And no, that's not the only option. That's just the quick
00:56:33recommendation I have because it is a tech stack with which I have worked a lot.
00:56:38Do you have any plans to create an AI engineer course? I've been hearing a lot about it and I've
00:56:47been reading about it online, but as a software engineer, it's still confusing for me. Yeah,
00:56:51I would love to. I would love to create a course where I show you how I work with AI. The problem
00:56:56just is things are changing so quickly and I'm still figuring out so much myself there
00:57:03that what I would teach today would probably be outdated in a month or so.
00:57:08So a course kind of feels wrong to me. I'm thinking about that and maybe it would be some
00:57:15cohort-based learning, like a live class with video recordings, but not a video on demand course in the
00:57:25traditional sense. I don't know. And maybe I also still have to wait a year or so for the dust to
00:57:31settle a bit more, though I'm not sure if one year is enough for that. But yeah, it feels to me like
00:57:36what I would share today would probably be outdated in a month. That's not a good feeling. It's
00:57:43difficult enough with AI-related courses because the software and the models change all the time, too.
00:57:49But at least for my Claude Code course, for example, the fundamentals I teach there,
00:57:54they still apply. The tool still works the way I teach it. It's not outdated. It's fully
00:57:59up to date. And I have kept it up to date with recent developments. But when it comes to my
00:58:04general approaches and how I think about things, that of course can quickly change in these days.
00:58:10That's why I have no course in that. Will the VPS course be available on Udemy? No,
00:58:16that is exclusive for our own platform. How do you structure a normal working day of your life?
00:58:22A normal working day for me is really I get up at 6 a.m. I have a coffee or I shower, brush my teeth,
00:58:30have a coffee, drive into the office, have a quick breakfast there, and then I work. And then I
00:58:38essentially get home at like 7 p.m. So I do spend roughly 12 hours here in the office. And then I,
00:58:48I don't know, watch some TV or some movie or just play a board game, whatever, with my wife.
00:58:55First we cook, then we spend time together on the couch or anything like that. And then I go to bed.
00:59:03So really not exciting. In an AI-first world, is learning new frameworks like Lynx.js still
00:59:11valuable or is prompting the new skill? Well, I think you have to know the stuff you're prompting
00:59:17for. You have to understand the technologies, the frameworks you're working with. So learning new
00:59:23stuff is still valuable. I think the only thing that's changing is how we learn. You don't need
00:59:30to learn all the syntax anymore, I think, because you're going to write less code, but you absolutely
00:59:38need to understand the idea behind that technology, its philosophies, its best practices, its features,
00:59:44and of course the fundamentals of its syntax and its API structure and so on. That is something
00:59:49that is still valuable and where it still can make sense to learn a new framework. Because if
00:59:54that framework solves a certain problem in a better way than other frameworks did, that's still
01:00:00valuable. I think the only thing that's less valuable now than it used to be is if a framework
01:00:06just offers better developer experience. Because in the past, of course, that was a thing. Developer
01:00:14experience mattered a lot. And that is arguably less important today because AI is generating more
01:00:22of that code. But on the other hand, AI experience may become more important. If a framework, a
01:00:28language is built such that it's particularly easy to work with for an AI model, that may be an
01:00:37advantage and a reason to learn and use it. And I know there are already frameworks and
01:00:42libraries out there that are purpose-built for AI. AeroJS, I think, is one. I haven't used it
01:00:47personally, but I think that is a framework that is built to be very usable by AI models. I know that
01:00:54Remix version 3 is another example here. It's not out yet, but it's also built such that agents can
01:01:00easily understand it and write code for it. So yeah, I think there still is a lot of value in
01:01:06learning new technologies. The reasons just have shifted and how you learn has shifted too.
01:01:10The challenge I see is that you now need to be an expert to judge whether AI-generated code is
01:01:20actually scalable or not, and companies expect this level of skill. But then how do we deal with
01:01:26that? Yeah, that's exactly what I meant. It's easier than ever to get a broad knowledge,
01:01:31but you need that deeper knowledge to really get use out of AI and to be useful and to have it produce
01:01:38good code. And I think that is something companies will learn that too over time. But of course,
01:01:46during the transition period, it's super annoying because companies expect something from employees,
01:01:52which just isn't there. It's impossible. You aren't suddenly an expert at everything,
01:01:59and neither is the AI. The AI just makes it seem like it is. But that is something that will change,
01:02:06that will improve, I'm certain. And the only way forward is to embrace this broader knowledge,
01:02:15which you can get, that is very useful too, but to then go deep on selected areas. And I think
01:02:20going deep is also easier now because AI can help there too. It can speed up the process of learning
01:02:28something, but you still need to spend the time and do the work of going deep. Yeah.
01:02:35When are you going to publish system design course, please? We are excited. Yeah, I'm excited too. I
01:02:43don't have a release date yet. I definitely plan to do it this year and it won't be this or next
01:02:50month or so. Yeah, maybe in the summer, in late summer. I don't know. I don't have an exact date
01:02:58yet because I want to produce a really amazing course there. Don't rush it and therefore, yeah.
01:03:04But it is on the shortlist. I said it before and it definitely is.
01:03:12Are you familiar with the Hermes agent? Any thoughts on using that? I was messing around
01:03:17LM Studio with QAN 3.6 for testing, but Hermes seems cool. So my understanding is
01:03:23that the Hermes agent is kind of an alternative to OpenClaw.
01:03:31I may be totally wrong there because as you can tell, I haven't used it.
01:03:38I haven't used it, so I can't really say anything useful about it. But my understanding is that it
01:03:45is an alternative to OpenClaw, that some people like it. But yeah, I haven't used it. And I'm
01:03:51also not using OpenClaw, by the way. I played around with it, didn't really find any great
01:03:56use cases for it. I'm using agents a lot for a lot of stuff outside of coding. But then again,
01:04:01I'm just using Pi and I tell it, here's a PDF document, extract the content, or here's a folder,
01:04:08find me the three PDF files that do X, Y, Z, whatever. So that's kind of what I do. And
01:04:14I don't need a daily morning summary briefing at 9am. I mean, these are always the kind of
01:04:22examples which I think sound nice until you actually see, well, I don't care about that. So
01:04:28yeah, that's why I'm not really using these agents for that.
01:04:31Do you plan to release a course about NestJS? No, because the last time I worked with NestJS was in
01:04:422019, I think. And whilst I did like it a lot back then, I haven't used it since then. So I'm not an
01:04:49expert. And so no courses planned. Worried about burnout. I recently received a job offer where I
01:04:56should be first and only dev at a small company. They want to lean into AI and think one dev can
01:05:03handle the entire app, including DevOps. Well, they will learn that this is not going to work. But yeah,
01:05:10burnout is a real danger here with AI for so many reasons, because you can feel the pressure
01:05:17to do more because you have this powerful tool and you have to get the most out of it, right? And
01:05:21suddenly you don't have any pauses anymore, or you work, you send a prompt, you go to X, scroll
01:05:29around. Oh, the AI is just done. You have to review it. You're constantly context switching. You're
01:05:34overloaded with work. Managers expect that one developer can do the work of 10 because that is
01:05:40what they read on the news or heard on the news. So I think that is unfortunately part of this messy
01:05:46transition period in which we currently are. And yeah, companies will learn that it doesn't work
01:05:55like this. And pushing back is the only thing you can do in the end. But it's just not a great
01:06:02situation. Let's face it. It's just not a great situation right now. It's this messy transition
01:06:07period. And people, a lot of people are expecting things from AI, which it just can't deliver. That
01:06:13is the sad truth. And even if we're not talking about employment, it's easy to feel the pressure
01:06:21that there is AI. You have to do something with it. You have to use it. You have to spend your tokens.
01:06:26So, so dangerous to get burned there, which is why I said before, I think it's super important
01:06:33to relax. You're not being left behind if you don't use the latest model. If you're not spending all
01:06:38your tokens, play around with stuff. Try to find out how you can use AI efficiently. At least that
01:06:44is what I'm trying to do. But kind of relax if you, yeah, it's really important. Nobody's winning
01:06:52if you burn out. Which job is secure future if compared to Devs? Yeah, I also kind of answered
01:07:00this before. I think the suggestions here are data engineering, scientists, DevOps, cybersecurity,
01:07:06cloud. So as I said before, DevOps, cloud, you could make the argument is a bit more secure because
01:07:13again, if we're talking about jobs here, companies are probably a bit more afraid of their service
01:07:18going down or their bills exploding. So they probably will not let an AI agent manage their
01:07:25cloud infrastructure. So from that perspective, it's secure. Cybersecurity will be a huge topic.
01:07:31But of course, if you're not interested in those jobs, you don't gain a lot because you will not
01:07:38be good at something you absolutely are not interested in. If you are interested, I think
01:07:42these are interesting areas. But again, I think developers are not as screwed as some people make
01:07:51it seem. The role is just rapidly transitioning. And it's absolutely possible, by the way,
01:07:57that we'll also move into a direction where all these roles are kind of blending and merging
01:08:02into each other. So a developer also does more DevOps. If you're working on your own, like I do,
01:08:08you do all that stuff already. I manage our servers. I have all these AWS certifications also because I
01:08:15personally always was interested in that. And I think more developers will acquire some cloud
01:08:20skills, some DevOps skills, and vice versa. Doesn't mean you have to become an expert at everything.
01:08:26You can't. But I think this will all blend a bit and new roles will emerge eventually.
01:08:31Before you even start any project, you should focus thinking how to prompt the model. It saves you a
01:08:40lot of hassle. In general, working with specs and spending time to craft good prompts. And with that,
01:08:49I mean, provide the right context and useful information and think about what you actually want
01:08:56from the model instead of just throwing in some random thoughts. That is important.
01:09:00The company I work for pushed us to use AI saying if you don't, that would be taken as a refusal to
01:09:10keep developing your career. And that will be noted in your performance review. So we all naturally
01:09:15started using it after they very aggressively pushed the deadlines because we now have AI.
01:09:20A lot of problems with the code and a huge bill they are blaming all on us. Fun times. Oh, man.
01:09:25Yeah, I obviously can't relate because I don't work in such a company. I'm self-employed. I have
01:09:31my own company. We are also using AI, but we're not pushing it on everybody here. And I do understand,
01:09:38indeed, if companies want to incentivize the usage of AI that they want employees to try it out. I do
01:09:51indeed understand that. But of course, forcing people and then telling them now you have AI,
01:10:00you can do more is horrible. And unfortunately, you are certainly not alone. I mean, good for you that
01:10:07you're not alone. But unfortunately, a lot of companies are doing that. I mean, we've heard
01:10:14reports about Meta doing the whole token maxing, token leaderboard stuff and so on. And I think for
01:10:20a lot of companies, they're just afraid that they're missing out, that all their competitors are now
01:10:25pulling ahead. And therefore, they have to use AI and have to use a lot of AI. And that's also what
01:10:30I said before. Quantity over quality. Right now, we're absolutely in a market where it's all about
01:10:36quantity. It's all about using as much of the AI as possible and not about quality. And whilst
01:10:42the opposite is also not good, it's really the balance you want. And right now, it's way more
01:10:47quantity. So things will change, certainly. But again, it's now that messy transition period.
01:10:53Unfortunately, there's nothing useful I can say. It sucks. It just sucks.
01:11:04AI is not perfect, even if the model has gotten better. No, yeah, it's absolutely not perfect.
01:11:08What do you think about Pi? Yeah, I think I already answered this. I like Pi a lot. It's a really
01:11:13awesome agent, I think, or agent harness, as we say. Do you find yourself more likely to try building
01:11:20products or services you could sell now that AI can dramatically reduce the time from idea to market?
01:11:25Yes. And I have built products, like built my graphic, mostly stuff that we use internally,
01:11:33like infographic building stuff and so on, which we then kind of published as products.
01:11:37I think it's also very deceiving to build a lot of stuff and then you actually don't publish it,
01:11:44because the first 90% are easy. And they were harder in the past, maybe,
01:11:51but they were never the hardest part. The hard part starts once the basic prototype is done. And yeah,
01:11:59AI can help with that. But then once it's about security, once it's about performance,
01:12:04once it's about deploying backups, if payments, once it's about the legal stuff and about marketing,
01:12:12all that stuff, AI can help there too, definitely. But it's still hard. It's still hard. And therefore,
01:12:20I think it's super deceiving that you want to build more stuff than ever, but it hasn't necessarily
01:12:26gotten... It got easier, but the hard stuff still is hard. Let's put it like this.
01:12:32But I know the feeling that you want to build more. What local model do you run and with what? Right
01:12:40now, recently, I played with the JAMA 4 models. I always liked the JAMA models by Google for the
01:12:46kind of work I do, which is about basic text generation, text summarization, extracting
01:12:51information from images and so on. So yeah, I'm using the JAMA 4 models right now.
01:12:56It kind of feels like AI takes away the stepping stone jobs. So you can't really learn on the job
01:13:04now as a junior developer. Yeah. And that's obviously a huge problem, but it's exactly
01:13:09as you said. And again, I think I'm certain that will change because senior developers don't come
01:13:16out of thin air, but right now, yeah, definitely. It's super, super hard as a junior and companies
01:13:23will pay the price for that, I think, if they totally stop hiring juniors. But that entire
01:13:30learning on the job thing definitely got way harder. And that's, as I said before, why I'm
01:13:35also thinking about how does education have to change? What can I do to kind of fill that gap?
01:13:41Because that period right now, definitely not easy as a junior. I feel like the AI agents took the
01:13:52joy of programming. These days, we're just code reviewing the AI code and code review was never
01:13:57the fun part of programming. Yeah. As I said before, got that video. What I say in that video,
01:14:03I still stand 100% behind it. I'm not saying that working with AI is horrible. I do like
01:14:10that it empowers you, that it makes a lot of things easier, that it makes it easier to build stuff. But
01:14:18the joy part of programming for me was really the flow state you could enter when you wrote code and
01:14:24that is gone. And I don't enjoy the code reviews either. Why didn't you update your Svelte course
01:14:32to Svelte 5? Because I have limited time and resources and I can't update all my courses
01:14:39all the time. So naturally, I update what's selling well and the Svelte course isn't unfortunately.
01:14:45And I also dive into all these new topics, create new courses. It's just impossible to keep up with
01:14:53everything and keep everything updated all the time. I would love it if it were different, but
01:14:58the only way for me to update everything all the time would be if I would AI generate everything
01:15:04and that is absolutely not something I will ever do. Yeah. About current TypeScript meta frameworks,
01:15:13what is your opinion on SvelteKit, my favorite framework, right after 10-stack start?
01:15:19I'm not that much into Svelte development these days. I've heard a lot of good things about it,
01:15:25about the remote functions especially, which is RPC done really well, as I understand it.
01:15:32I personally use a lot of 10-stack start these days. I like it a lot, so I can relate to that,
01:15:39but I can't say much about SvelteKit. Learning is always valuable. Don't be blind to think
01:15:45that you can speedrun your life using AI. 100% agree. What's generally better? Being in a niche
01:15:52field like networking cyber with few switching options or being in a high-dominated field like
01:15:59that of full-stack web AI engineer? If you can get the jobs and clients, the niche field
01:16:07will very likely be better. And that was true before AI too, of course. But yeah, it's exactly
01:16:15the trade-off you're mentioning. One field has lots of jobs, but lots of competition. The other field
01:16:22has less jobs, less competition. I mean, right now, if you're a cybersecurity expert, that is probably
01:16:27a good spot to be in because now we're all realizing how big of a problem cybersecurity will be in the
01:16:34near future. But maybe you could have predicted that three years ago, maybe not. But in general,
01:16:40it's of course hard to predict what will be in demand in three years or so. So I always say,
01:16:46you have to do what you enjoy because you'll only be good at things you enjoy. And if you think
01:16:51cybersecurity is awesome, but you hate it, you will not be good at it. That's kind of my stance on that.
01:16:59Did you see the hanker rank video on lead code is dead? What's your take on technical interviews in
01:17:04the future? I didn't see it. And indeed, I'm very interested in seeing what will happen to lead code
01:17:10style interviews in the future. You could make the argument that companies still are interested in it
01:17:16because more than ever, they want to see how you think, because that is the thing that matters,
01:17:23that you can think like a software engineer, that you can come up with solutions, that you can
01:17:29architect software. So that would be an argument why these interviews will continue to be conducted,
01:17:36but they may change. I mean, in the past already we had interviews where you only wrote pseudocode.
01:17:43Writing actual code I think will become less important there, but I definitely see companies
01:17:49still being interested in that kind of knowledge. But I don't know if they will stick to this
01:17:56interview format or if it will be a format where they want to see how you can apply that knowledge
01:18:01to working with AI tools so that you have to solve a concrete problem with help of AI so that you can
01:18:06see, that they can see how you prompt, how you work with AI. Hard to tell for me, but I don't necessarily
01:18:14see lead code dying right now just for a different purpose. I think the way you think still matters to
01:18:21companies, to some companies. Personally, I was never a big fan of lead code interviews
01:18:29because I totally get that you want to see how a candidate thinks that they have a certain way of
01:18:39approaching problems. But I think the lead code problems are sometimes very, very specific and
01:18:46not necessarily what you will be dealing with in your day-to-day work. And I would be more
01:18:51interested in seeing how you tackle those day-to-day problems. Which software development language and
01:19:03framework will you use in future system design course? It will probably just be TypeScript because
01:19:10that is what I'm good at. But the course will absolutely not be about the language. I don't even
01:19:16know how many code examples will be in there because it will not be about writing lots of code. It will
01:19:23be about understanding the different building blocks you have in systems. How they work, why they exist,
01:19:28when you need them, when you don't need them, when you may need them but you actually don't because
01:19:34for your constraints, your use case, it's the wrong decision. I think that is the value such a course
01:19:39could add and it won't be about the programming language or a specific technology. How do you
01:19:49master a complex technology fast? Docs first, small demos or solving a specific problem?
01:19:56Yeah, generally what you said especially before AI it was for me it was like dive into the docs,
01:20:02build small dummy apps, then build something bigger. Now with AI I may skip the small demo app
01:20:08part. I still dive into the docs. I really want to understand it. I don't just want to trust the AI
01:20:15summary or ask the AI. I will do that too but I still personally dive into docs, read them and then
01:20:22I may just tackle a concrete maybe a bit more complex problem. Could still be a demo app,
01:20:28a dummy app but it may be a bit more complex and I may do that with help of AI but then I really want
01:20:33to dive into the code, understand it, compare it to the docs, see if that's in line with the docs,
01:20:39challenge AI, talk to AI about it. That would probably be my approach there.
01:20:51Everyone who is using AI is getting left behind because new AI would render old techniques
01:20:56obsolete anyway. Well, I don't necessarily agree with that. I don't think you get left behind
01:21:01because you do use AI but you probably have a disadvantage if you use AI and then you close
01:21:10your mind and you don't look at what's happening and you just stick to what works for you right now
01:21:18or to the opinion you formed at a certain point in time. I mean, as far as I know or as I understand
01:21:27it, what we can see is that I'm obviously in a bubble and many of us are. We are in a bubble
01:21:35where we read about every new model and every new tool and every update for a tool that is
01:21:43being released. That's a super small bubble. The vast majority of people out there haven't
01:21:52heard about Claude Code. For many people out there, AI is the free version of chatgpt as it worked two
01:22:01years ago. That's getting left behind. Now, also not fully left behind, you can still catch up.
01:22:10Obviously, it just takes you a couple of days to get up to speed but that's my definition of being
01:22:16left behind, I would say. Will you have a course about BUN? That is also something that's on the
01:22:26idea shortlist. I would love to do a course about BUN but for all these tech framework or runtime or
01:22:36language related courses, I'm still figuring out how to best teach that now in the age of AI because
01:22:42it's not the old course style. Not that this is necessarily bad but I of course also want to evolve
01:22:49how I teach and that's why I'm still trying to figure out how I best do that going forward.
01:22:54But I would like to do a course on BUN, yeah. When Opus 4.5 came out, I was able to create two
01:23:00projects I always wanted to make. At the end, I became the only user and I still use it. However,
01:23:06I'm now demotivated to work on another app. Advice? So I don't think you have to work on another app.
01:23:12So are you demotivated because you have no other users? That is of course the hard part.
01:23:21Marketing is super hard. It's extremely, extremely hard. I don't think it's necessarily right to just
01:23:29spit out more and more apps if you haven't at least tried stuff to make the other apps visible.
01:23:36Could be through social media marketing, not necessarily paid marketing, but obviously you
01:23:41can build a YouTube channel where you talk about how you build those apps, like a tutorial style
01:23:48channel. That's just always my example because YouTube is obviously what I do. Stuff like that.
01:23:56Obviously traditional SEO work, blog post articles. Of course, that's all a bit changing due to how
01:24:04Google works and AI summaries, but it's not like it's not working anymore at all. So I think that
01:24:11is something you can and should try because just building more and more apps and all those apps
01:24:17failing because nobody knows about them is of course not really motivating either. But I can totally
01:24:24relate. You have a lot of ideas, you build stuff, which you personally like, and then it can really
01:24:30feel bad that nobody else is using it. But for me, for the projects I built, like this
01:24:36build my graphic thing I mentioned before, we have paying users there, quite a few of them,
01:24:42because I have of course a certain amount of reach. And it's not like I mentioned it every time in all
01:24:48my videos, but here and there. And therefore that is something that can work here in which then gives
01:24:53you extra motivation to maybe then also work on other stuff. But obviously that's not an easy
01:24:58route. That takes time. I'm fully aware of that. But marketing is unfortunately the annoying part
01:25:04that is difficult for me too. It's not like it's super easy, unfortunately. Where was I?
01:25:14Do you only stream once a month? No, typically not. It's been a super long break, unfortunately.
01:25:22Six weeks or so. I try to stream every Thursday at this time, 5 p.m. Central European Summer time.
01:25:34But I'm not a full-time streamer. I really only do this for fun, and I don't earn any money,
01:25:43or not a significant amount of money here. I really just like doing these streams,
01:25:47so I try to do them every Thursday. But if there is something that gets in the way, something
01:25:52personal or something very important, work specific, then I will not stream. But the
01:25:58plan is not to stream once per month, but every week or at least every two weeks.
01:26:04Today you prefer Tanstack Start or Next? Do you have a plan to do a full Tanstack ecosystem course?
01:26:19I use Tanstack Start a lot. I don't use Next.js a lot. Maybe just because of the shiny new things
01:26:29stuff. It's not like I have a problem with Next.js. But yeah, I like Tanstack Start. What I didn't like
01:26:37about Next.js, and I know that it's better now, but what I didn't like was how slow the development
01:26:43server was, and how slow some things were, and certain bugs I was facing. I know the development
01:26:49server is way quicker now, and bugs have certainly been fixed, but that is when I started looking for
01:26:56alternatives, and I just like Tanstack Start right now. Do I have plans for courses on it?
01:27:02I don't know. Maybe a Tanstack Start course, but not on the entire ecosystem, because I don't use
01:27:09everything from the ecosystem. I really only use Tanstack Start, therefore naturally I use Tanstack
01:27:14Router, a bit of Tanstack Query and that's it. I'm not using Tanstack DB, Tanstack AI, I'm not using
01:27:20any of that. What do you prefer? Angular, React or Vue and why? These days I work a lot with React.
01:27:31Main reason is that it's AI's favorite. As I said, I use Tanstack Start a lot, so that's a good fit.
01:27:39Angular will always have that special place in my heart, because that was my first important
01:27:46Udemy course back then, and I still like Angular, but for the reason that I'm a lot into Tanstack
01:27:53Start right now, and that React is really all fine-tuned towards React, I'm using a lot of React
01:28:01right now. If it were just the syntax or how the framework works and about its API, Vue.js may be
01:28:13number one actually. I always like Vue and the simplicity of Vue, especially Vue 2, but for the
01:28:21reasons I mentioned right now, it's React. Do you plan to release any courses about SAP, the German
01:28:27ERP software and ABAP? I don't know ABAP, I know SAP, but only what it is, but I'm absolutely no
01:28:35expert on it, so no plans on it right now. When you are hearing CEO of AI companies like Dario talk
01:28:43about programming going away and everything being done by AI, how much do you think it is hyping and
01:28:48marketing for themselves? Yeah, I don't fully get what exactly Dario's motivation is. Now, you have
01:28:55to give the man credit. He wasn't totally wrong regarding the percentage of code written by AI,
01:29:02with the important exception that definitely not all developers out there have 90% of their code
01:29:11written by AI, but there definitely are developers where this is the case. Now, when it comes to AI
01:29:17wiping out all these jobs, I have a very hard time with that. Obviously, Dario knows more about
01:29:29future model capabilities than I do, because he can see how certain models work that haven't been
01:29:38released to the public yet. From what I can see, we're far away from AI doing it all. Very far away
01:29:46from it. Now, will that magically change with new models in half a year or a year? Maybe.
01:29:52Just doesn't seem like it to me from what I can see from the existing models.
01:29:58So, I don't know what his motivation is there. It sounds very much like a big part of his motivation
01:30:10is that he wants to get attention onto Anthropic, because we're in this year where SpaceX,
01:30:16Anthropic and OpenAI all want to go public. They all want to have their IPO and they're all valued
01:30:24at such eye-watering sums of money that I think it will be hard for all of them to just collect
01:30:35that amount of money they want to collect. So, obviously, you want to get attention,
01:30:41you want to be the company where investors want to put their money, so you want to have a big
01:30:45market share, especially in the enterprise segment, something where Anthropic is really good. And if
01:30:51you sound like the company that's responsible, that's warning people, that's withholding models
01:30:57like the Mythos model, and you're the company that's reasonable, and at the same time you're
01:31:04the company that has the powerful models that will disrupt everything. That's not the worst framing,
01:31:09so that may be part of his motivation. That's my explanation at least. And we will see what future
01:31:16models will bring. But that is kind of how I think about it. I'm a full stack developer and I'm
01:31:25thinking of switching in AI. Where should I start? Are you bringing any AI cores? I'm not sure what
01:31:30you mean with AI. Do you mean switch to building AI models? I talked about this before. By the way,
01:31:36the live stream will stay online on YouTube after I'm done, so you can still watch it there.
01:31:41I don't think that switching to AI just because AI is the hot stuff right now is a good idea.
01:31:50You want to switch, if you want to switch, to a role where you help companies adopt AI. And that
01:31:55can still be as a software developer. Being a software developer that is able to leverage AI,
01:32:02that's definitely useful to companies. Any consulting stuff related to AI can be useful
01:32:08right now in that transition period. And that transition period will take a while, so that can
01:32:13all be interesting. And really, I would go with what you're good at, what you like, and where you
01:32:21can help companies make that AI transition. And that can really be any role in the end.
01:32:27Just if you call yourself a traditional software developer totally without AI or stuff like that,
01:32:33that will probably not be super appealing to most companies.
01:32:38Have you ever had a chance to use some Chinese LLMs like GLM 5.1 for coding? Do you think they
01:32:49can be comparable with Claude? Yeah, I do play around with these models here and there.
01:32:52I'm mostly interested in the open models from there, which I could host myself because I
01:32:59find the idea quite intriguing that I could have a capable coding model run on a Mac Studio. I don't
01:33:11have a Mac Studio, but potentially in the future. So that's something I would be interested in,
01:33:15and I absolutely see that they could be comparable with Claude at some point. Not there yet from my
01:33:23experiments, but close enough. And all these models are being fine-tuned right now to work well in
01:33:31these agentic harnesses, to work well in these agentic coding tools and situations. So why not
01:33:38also some open models? And especially with the token prices and the compute constraints, having
01:33:47local models and decentralized clusters or running these models on your own machines in your company,
01:33:57your own small data centers or on a Mac Studio or whatever. That could be very, very interesting as
01:34:04we see those token prices go up, whereas we maybe see those subscriptions go away. Now I think these
01:34:11frontier models will very likely stay ahead. And of course, even if you had an open model that's
01:34:23pretty much on the same level as the frontier models by Anthropic and so on,
01:34:30that would be an open model that probably requires hardware that's so expensive
01:34:36that I'm not sure if it's the better deal compared to paying the token price.
01:34:42But still, yeah, I said it before, I'm a fan of open models and I definitely see open models being
01:34:48good enough or very good at agentic coding in the future. And to some degree already today,
01:34:57to be honest. Do you plan to release a course about Kubernetes? I have a Docker course where
01:35:02I also cover Kubernetes. I have that. Here it is. But Kubernetes is just a small part of that course.
01:35:15A dedicated Kubernetes course is not planned right now. Again, definitely a topic I would like to do
01:35:22more, but eh, time. How to level up my coding skills? Build stuff. Really build stuff with AI,
01:35:30but make sure you understand it and you steer the AI. But it's always been about building stuff and
01:35:37that hasn't changed. I'm a junior developer relying a lot on AI like chatgpt or codex. Could this hurt
01:35:43my learning? What's a healthy balance? It can definitely hurt your learning because the danger
01:35:48with AI always is that you just ask it, you ask it, you take its answers. You don't really question
01:35:53those answers. You don't really look at the code. Maybe just briefly so that you have a good feeling.
01:35:58That is dangerous. Challenging the AI. As I said it before, having discussions with the AI
01:36:05about the solution it proposed or implemented, reviewing the code, pushing back, understanding
01:36:11the code, looking at the docs of the framework or technology you're using, thinking about the code
01:36:17yourself. Yes, that will be slower than just prompting, prompting, prompting, but that will lead
01:36:23to learning and will give you personal growth, which will make you more productive with AI in
01:36:29the future. And even if you are already an expert, of course it would be faster if you just keep on
01:36:34prompting, but taking the time to review and to push back and to optimize the code, that saves you time
01:36:42in the future, makes you more productive in the future, leads to a better product. At least it
01:36:47increases the chances of getting a better product. So it's a valuable trade-off in my opinion. And
01:36:52especially for learning, it's important. I'm a junior developer and I rely a lot on AI. Like,
01:36:58oh yeah, I already answered this. For the person feeling demotivated, always build, build. And
01:37:06about the app not getting Angry Birds was created in 2009, which was on the brink of bankruptcy after
01:37:1251 failed games. That is very motivating, by the way. Yeah. I think it's a trade-off. You can't
01:37:18just build, build, build and never try marketing because there may already be a hit in those apps,
01:37:25but if nobody sees it, that's worth nothing. But of course, it's very true that most successful
01:37:33companies or founders weren't necessarily successful with their first or second product.
01:37:40It may take a lot of attempts. Next question for all the devs on this chat. Don't you feel
01:37:46demotivated when your work that took several days now can be done in 10 minutes using AI
01:37:52or the fact that writing code which makes you feel nice now isn't the most important? Yeah.
01:37:59I already answered it before, so I'll keep it brief. Whoops, that's the wrong one.
01:38:05I have a video, of course, for me. Yeah. It was more fun before AI when it comes to the coding part.
01:38:14The 10-minute part, I mean, the code AI gives you after 10 minutes is not necessarily the code you
01:38:21want to accept. So you can and you should still use your knowledge and your brain and so on. But
01:38:30our role is changing and I totally get and I feel the same that some parts that were very joyful
01:38:38are taken away. Do you still use Redux? No. I mean, most React projects I work on these days
01:38:49are full-stack projects with 10-stack start or something like this. So I haven't found myself
01:38:55in a situation where I need a state management library very often. If I do need one, I often use
01:39:02Zustand, which is another great library. Not that Redux is bad necessarily. Zustand is so difficult
01:39:11because it's a German word, but I don't want to say it in German. But Zustand is a very simple and
01:39:18good state management library. So I use that sometimes, but often I don't use any at all.
01:39:26Love your work. You've been a huge impact on programming over the years.
01:39:30Thank you. What impact has AI been on buying your courses? Is it true tutorials for humans are dead?
01:39:37So it's difficult. Coding courses sell worse. AI courses sell very good. I mean,
01:39:50the Claude code course, for example, is our best-selling course ever. Ever. It sells very,
01:39:59very well. The React course still sells very good, but then there are some coding courses which
01:40:09don't sell very well anymore. And especially on YouTube, you can see that tutorials don't perform
01:40:16very well anymore. So yeah, overall course sales are down. Also because of those software developer
01:40:26jobs. I mean, until like two years ago or three years ago, I obviously only had coding courses. So
01:40:34that was good here. Less good here. Now I have a bunch of AI courses, too. And of course,
01:40:41these are all AI coding-related courses because I'm still a developer, but those sell better.
01:40:47Way better now. But overall, it's still less.
01:40:51Any plan to release a COBOL course which is still highly present in the banking industry?
01:40:59I do think COBOL has its place, and you can probably make a lot of money out of knowing COBOL,
01:41:04but I don't know it. So no course plans right now. I joined late to the streaming, so do you have a
01:41:09course on system design? No, but planned. A course I absolutely would love to do and I will do, and I
01:41:16plan for this year, but I don't have it yet. But it will be great. It will be a great course. I'll
01:41:22do my best to make it an awesome course. Thank you so much, Max. Sending my appreciation and
01:41:27warm regards from Ethiopia. I hope you're doing well in life, and may God bless you and your
01:41:32family. You've been incredibly helpful in my journey. Wow. Thank you so, so much. Thank you.
01:41:37And the same wishes to you and your family and loved ones. Yeah. Don't know what to say. Thank
01:41:43you so, so much. It means a lot to me. Thank you. I would like to see a course making behind-the-scenes
01:41:50vlog from you one day. Today I have all your English Udemy courses in both Reactbooks. Wow.
01:41:55Don't know what to say. That is just amazing. Adebow88, thank you so, so much. I hope you got
01:42:03a lot out of the courses in the book. And regarding the suggestion or the books, I should say,
01:42:08and regarding the suggestion about the vlog, I played around with that idea. I just feel really
01:42:17awkward when I do stuff like this. I'm not like that traditional YouTuber. I'm not the hyped up
01:42:25kind of person. And I'm not sure if a boring vlog would be interesting. I'll take the idea with me.
01:42:35But I don't know yet because I would want it to feel authentic. And the truth is my days are
01:42:47pretty boring. As I said before, I get up at 6 a.m., drink a coffee after showering and so on,
01:42:55go to the office, spend my days here, prepare courses, work on courses, work on other projects,
01:43:00support and so on, drive home, cook, do something with my wife, go to bed, rinse and repeat.
01:43:09But yeah, I'll take the idea with me. How much time does learning Terraform
01:43:16take to be able to say that you can use it? I'm no Terraform user or expert.
01:43:25These platform as code solutions like AWS, CDK, Terraform, Pulumi, I think now with AI,
01:43:34like everything, are a bit easier to learn. But if you want to dive deeper, what you want,
01:43:40if you really want to learn it, that still takes time. And I think the syntax or the logic behind
01:43:47those platform as code solutions, I never found it too hard to get that, though I haven't used
01:43:53Terraform, so I don't know specifically about that. But for Pulumi, I found it okay.
01:43:58The difficult part is to then know about all the services your favorite cloud provider offers,
01:44:04the best practices of combining these services. And then sometimes these platform as a code
01:44:11solutions have their own glitches and problems. For example, the AWS CDK doesn't support all services
01:44:18equally. So suddenly you want to use a service that's not well supported and things get awkward.
01:44:23That's the part I found challenging about the AWS CDK and it may or may not be similar for Terraform.
01:44:30Wow, I just saw this, missed most of it. No worries. The stream recording will stay online.
01:44:38And I had a long pause. That's not the plan. Typically I try to stream every Thursday
01:44:44at this time from 5 p.m. Central European Summer Time to 7 typically.
01:44:52I'm not able to make it every week, but I try my best
01:44:55and I definitely don't plan on having six month pauses all the time.
01:44:59When was your node course last updated? That was a couple of years ago, I would say.
01:45:08Two years ago, three years ago, something like this.
01:45:12I definitely haven't rerecorded it in the last years or anything like that, if that's what you mean.
01:45:20Again, definitely something I would love to do, but just like with the BUN course,
01:45:24I'm still also trying to figure out how to best teach stuff like Node.js or frameworks
01:45:31now in this new world, which is made easier by the fact that everything's changing so quickly.
01:45:35So yeah, that's part of the reason. And of course, time. I would love to
01:45:40update all my courses every day. That's just not something I can do.
01:45:46If I could, I would, but I will not do it by having AI generate videos or audio or anything like that.
01:45:53I will never do that. Can you build and deploy an app to production
01:46:00that can be used by people using only AI? I think I know a bit about coding and keeping the app secure.
01:46:06Can you build and deploy an app to production that can be used by people using only AI?
01:46:16I think that will have its problems, but of course it will depend on the app. Something very simple,
01:46:22probably. But there is a high danger that the app will have bugs. There may be bugs that are
01:46:31security relevant or where people have a bad experience, they pay but they don't
01:46:37get access to the service, stuff like that, which absolutely must not happen.
01:46:43I personally would be too afraid. So if I were to use, let's say, a language or a framework I don't
01:46:51know anything about, if I would just wipe away, even though I have a lot of knowledge about
01:46:56programming and so on, I would probably not feel comfortable deploying that.
01:47:00Maybe that's the wrong stance on this, but that's my stance on it.
01:47:10Coursemaking with your quality take a lot of hard work and I think many people would just love to
01:47:13see behind the scenes. Yeah, thank you so much. I'll think about it. Maybe I can do something
01:47:20interesting there. I'll take that with me and try to do something there. Thank you so, so much.
01:47:25How many languages do you speak? So not programming languages, you mean? English, German and a bit of
01:47:33French. Three. Just jumped into the chat, quick question. A lot of people say you shouldn't bother
01:47:41learning programming anymore and should focus more on architecture, system design and being an AI
01:47:45driver. But don't you think that if token costs and subscription prices keep going up, programming
01:47:50as a skill might make a comeback? I talked about this earlier and the recording will stay online,
01:47:54but my short answer is yes. I think with rising token prices, we may indeed, maybe, I don't know,
01:48:01maybe have a transition period where more code is being written again, maybe in conjunction with
01:48:07smart autocompletion like GitHub Copilot before it became an AI agent. So that may be a thing. And
01:48:13either way, I think programming knowledge will matter because somebody needs to steer and control
01:48:19and review the AI output. So that's not the same as writing all the code. That I think will not come
01:48:26back, maybe only to a small degree during that transition period as mentioned. But in general,
01:48:32knowing about it and knowing about the code, not just the architecture, I think will also still be
01:48:37a valuable skill and will be important. But I do think that the entire architecture part, the
01:48:44patterns part and so on will play a more important role in the future. I think that will be important
01:48:50for most developers. That is my personal opinion here. Can you tell us approximately when will you
01:48:55release your system design course? I think it will. I don't want to promise it. That's important
01:49:02because I will only release it once I'm happy with it and I won't rush it just to make some date. But
01:49:08I think a realistic expectation hopefully could be late summer, August, but again, no promises there
01:49:21because I want to create an awesome course there and that just takes time and then sometimes life
01:49:26gets in the way or other work stuff. So yeah, but the plan is definitely this year, maybe as early
01:49:33as late summer. Too much rely on AI. It is bad as developer. It must be balanced. Yeah, you shouldn't
01:49:43rely on just AI. That's vibe coding and that has its place. I mean, some internal tools and stuff
01:49:50like that. You don't need to have the best code ever there, but just relying on AI. I mean, if you
01:49:56just rely on AI, you're only as good as AI and that may be fine if you're self-employed or if it's just
01:50:03a fun project, but if you're looking for a job, why would I hire somebody that isn't better than AI or
01:50:10that doesn't add value to the AI? I don't need people that don't add value to an AI from if I
01:50:16were an employer, so yeah. "Code with Antonio is doing live training courses now. Have you thought
01:50:22about it?" I have thought about it and I might do that. No super concrete plans yet, but again,
01:50:29still figuring out a lot of stuff. There is just a lot going on obviously with AI and all these tools
01:50:34and so on, but yeah, definitely something I'm also thinking about. "Is Vercel the best hosting
01:50:39platform with a generous free plan? Are there better alternatives out there for hosting AI
01:50:44applications?" The problem with Vercel is that it can get expensive if things are maybe not
01:50:52configured perfectly. It is a great platform though, great for getting started. I personally
01:51:00really enjoy using my own VPS, also because I like the server admin stuff and so on, but that's of
01:51:07course not for everyone. Netlify is a great alternative to Vercel. Cloudflare, if you want
01:51:12to go down the Cloudflare path, but there you have to adjust your code a bit often,
01:51:18but that's a great alternative too. But yeah, nothing wrong with Vercel. You just probably
01:51:25want to spend some time on configuring stuff and maybe set some budgets and some alerts and make
01:51:30sure that you don't suddenly spend more than you would be willing to spend. "Thank you Max for your
01:51:38courses and dedication for spreading programming knowledge. Your work has had a real impact on my
01:51:42career and I'm sure you're not the only one who feels this way. You've helped shape the paths of
01:51:46countless engineers and your passion for teaching continues to inspire. I truly appreciate everything
01:51:51you do." I appreciate that so, so much Pavel. Yeah, I don't know what to say. That's just so amazing.
01:51:58There were so many nice messages today, this one. So thank you so, so much. It means a lot to me.
01:52:05It truly, truly does. Thank you so very much. "I just started following your Next.js course.
01:52:11Vine is lagging so much, especially when designing a web app. What's causing this when it comes to
01:52:17React counter? It runs smooth." So the app is lagging. Normally that shouldn't happen. I mean,
01:52:24Next.js in development has its issues. Some stuff is slow there indeed. Page navigation can be slow,
01:52:30unfortunately. I mentioned before, this was part of the reason why I switched to TAN stack start,
01:52:35but I never encountered horrible lag, to be honest. From the top of my head, I don't have a good idea
01:52:40what could be causing that. Sorry. If it's the course that's lagging on Udemy, that is something
01:52:48for their support. There is nothing I can do about that, unfortunately, because I don't own the
01:52:52platform and I have no influence on the platform itself there. "I learned JavaScript React and Next
01:53:01with your courses and today I have a software house with 30 devs started in 2021 in Milan because of
01:53:06your hard work." Wow. Thank you. That's just amazing. Hearing what people do with the knowledge
01:53:11I share, that means a lot to me because that is ultimately why I started this all. I wanted to
01:53:16share knowledge so that other people can get something out of it and do their own stuff,
01:53:21build their own companies and stuff. Hearing that really makes me super happy. That's so amazing.
01:53:28"Join late, however, to this. How about local coding LLMs that I think are coming,
01:53:33which you can run on your machine that will offset the external token prices?" Yeah, I reply to that.
01:53:38Local large language models, I like them a lot. For coding, absolutely. Could be interesting when token
01:53:48prices rise, but of course, it depends. If you need a super expensive machine to run those local models,
01:53:56hard to tell what is cheaper. Pay the higher prices, pay for the machine because it's not like
01:54:02one Mac studio or whatever can serve 10 engineers. So that could get very expensive too. Nonetheless,
01:54:11I'm a big fan of local models and I definitely would love to see one that is capable. I mean,
01:54:18we would all love to see that. That is capable and good and has been fine-tuned for coding and
01:54:23can maybe run on a MacBook Pro M4 or something like this or another Linux or Windows equivalent,
01:54:30of course. Thanks for the answer. You're 100% right. Here's an interesting thing. Our project
01:54:38was taken over because my company is expensive and the new team had permission to use AI. Let me tell
01:54:43you, the kind of over-engineering miracles happening there are absolutely insane. It scares me to think
01:54:48how many apps like this may be going into production and the fact that it works means there is nothing
01:54:52to complain about on the surface, but underneath it's a total mess. And that's the big problem.
01:54:56AI can produce working code that may have bugs, but they may not always appear immediately and it can
01:55:04over-complicate things. It's important to keep AI in check. It's important to review and steer it.
01:55:10I can only repeat myself there. And I know that many companies have a different philosophy right
01:55:14now. And I think the price will be paid and we can already see that many software products
01:55:21are degrading in quality right now. Will this architecture course be long? I don't know yet.
01:55:27I want to create a good course and it'll be as long as it needs to be. I won't arbitrarily bloat it
01:55:32or something like this. You should make a course on payload CMS. I got no experience on that,
01:55:38unfortunately, so no course planned. Do your triangular lights in the background still work?
01:55:44I haven't seen them on in some time. Or have you just gotten bored? No, I haven't gotten bored. The
01:55:49problem is if I turned them on, I wasn't able to calibrate everything such that they wouldn't
01:55:55flicker. They would constantly flicker. And I found that super annoying in the recordings.
01:55:59So I just kept them off. That's the reason. And I wasn't willing to spend the time to make it work.
01:56:06Because I spent a lot of time, it wasn't able to make it work. So yeah, also because I'm switching
01:56:11between recording setups and I got some other stuff that changes here and there. That's the reason why
01:56:16they're off. Since AI is not reliable, I think I'm going to learn React and Node.js. Do I need to
01:56:24learn JavaScript first? Yeah, I would learn some JavaScript basics. The base syntax, the base logic,
01:56:29I would learn that first. I wouldn't learn every niche feature, but I definitely would learn the
01:56:34basics first. Just finished the VPS Essentials course. It's absolutely amazing. Thank you very
01:56:40much. Thank you very much. Super happy you're liking the course. Yeah, I try to share the
01:56:46essentials that I need for every new application I deploy on a VPS essential. Essentially super happy
01:56:53that you got a lot out of the course. In case anybody's wondering and maybe interested,
01:56:57talking about this VPS Essentials course, which you can find on academy.com. It's a brand new course
01:57:04where I teach you how to properly set up a VPS and run a web application there or OpenClaw or
01:57:10some workflows. Just wanted to say thanks for your work. You are sadly one of the few internet
01:57:18developer personalities that I think we have lost over the years. The programmers got out of control
01:57:23these days, so I appreciate you, the chill disposition. Also, thanks for the call out,
01:57:27what I think we're all thinking lately. Yeah, thank you so much. And I'm definitely not going
01:57:32anywhere. I'm staying. And of course, I'm changing too. And how I work changes because it's changing
01:57:38for all of us. And I'm not creating all that AI content just because it sells, but also because
01:57:45it sells because that is my business after all. And I don't get anything out of just
01:57:50creating old school coding courses that barely anyone buys, unfortunately. But at the same time,
01:57:56I'm also doing that because I use AI. We kind of have to and have to adapt and have to learn how to
01:58:03efficiently work with these tools and get a lot out of them and also get more joy out of them. That's my
01:58:08goal, at least in my mission. So yeah, I'll answer these last messages, which are there right now,
01:58:16because I got to leave soon. I believe in five years, local models will be more efficient.
01:58:21So small models will be able to do the same job today. I don't think token price will
01:58:26matter in the future regarding adding value to AI. Yeah, I think for local models, I definitely
01:58:31see a future maybe where we would have a lot of very specialized small models that can do
01:58:36certain tasks extremely well. And that could be super valuable. Do you still believe that
01:58:42it will still exist, that software developers exist in two years? Yeah, I absolutely believe
01:58:47that. I can't look into the future, but I absolutely believe that. From what I see with models today,
01:58:53how I work with them, what you still need to know, which value you add as a developer,
01:58:58when I look at all these things and then you have other factors like companies not transitioning
01:59:03that fast and so on, I absolutely see developers still existing in two years. But our role is
01:59:09changing and evolving for sure. I agree with you. I feel that in close future good models
01:59:16with wide contacts could be run on average PC or Mac Studio without spending thousands of dollars.
01:59:20So, yeah. Thank you all so, so much for joining. I have to go. So, I got through all the messages
01:59:28I can see right now. Again, I'll try to do this stream every week. In the last year, I did it
01:59:37almost every week. And the long pause I have for now, thank you.