Transcript
00:00:00Thank you.
00:00:30Okay, finally back online or live streaming again, I guess.
00:00:58Give me a second.
00:01:06Okay.
00:01:07Yeah.
00:01:08Hi, everybody.
00:01:09Hi.
00:01:10Welcome to another stream.
00:01:12I think it's been three weeks.
00:01:14Is that possible?
00:01:15It's been a couple of weeks since the last stream.
00:01:19Life got in the way, as it sometimes is.
00:01:22Normally, I try to stream every Thursday, but yeah, I wasn't always able to do that.
00:01:27Oh, thank you so much for the subscription.
00:01:30That is an amazing start.
00:01:31Thank you so much.
00:01:32And yeah, welcome everybody to the stream.
00:01:35The idea really just is like, obviously, since it's been three weeks, a bunch of stuff happened.
00:01:42And so we can hang out.
00:01:45If you got any questions, anything you want me to comment on, let me know.
00:01:50I also got some ideas, of course.
00:01:52And yeah, we may also build some fun stuff along the way.
00:01:59It's super warm here.
00:02:00I got my air conditioning running full throttle, but still super warm.
00:02:05So please forgive me if I'm maybe a bit not super sharp all the time, but yeah, should be fine, I think.
00:02:15So right now, I just read on X that looks like a bunch of people are waiting for a Kimi K3 to drop a new model.
00:02:28I mean, we had a bunch of models over, especially the last week, right, with the GPT 5.6 models, GROC 4.5.
00:02:36And I don't want to turn this into a long discussion about models, but it's been some exciting weeks, I will say.
00:02:43It's been some exciting weeks, lots of AI-related drops and stuff.
00:02:47I have been super Herder-pilled.
00:02:51In case you don't know, Herder is a terminal multiplexer.
00:02:56And today I released a video on my other channel, Academind, where I dive a bit deeper into what it is and what you can do and why I think it's pretty great.
00:03:07So, yeah, there is that.
00:03:11And since I just saw it in the chat, by the way, if you have issues watching my Udemy courses, Udemy support is the right address because I am totally not involved in that platform.
00:03:22I have zero influence on it, especially also not the technical aspect.
00:03:26So if you're facing any issues there, I'm very sorry, but I can't do anything about that.
00:03:31Udemy support is the right address there.
00:03:34So, yeah, a bunch of stuff has been happening.
00:03:37And definitely, as I mentioned, let me know what your thoughts are about the last weeks, what's happening, how the entire development world is evolving and changing.
00:03:51And, yeah, I just read in the chat that Kimi K3 is released.
00:03:55I read it on X2, but I'm waiting for their official announcement.
00:04:01And on their blog right now, I only see 2.6.
00:04:06So, at least I think that is the right place here, right?
00:04:14Not sure.
00:04:16Yeah, but a bunch of stuff is dropping indeed.
00:04:19And it's weird times, I will say.
00:04:22It's pretty weird times.
00:04:25Personally, as I mentioned, I have been pretty active playing around with my herders set up with multiple agents.
00:04:37Yeah, and good point, by the way, also in the chat, that the Academy dashboard is also buggy sometimes.
00:04:44I am very, very sorry about that, too.
00:04:47The problem here is that this, despite the branding, is also not our platform.
00:04:51We are working on our own thing, though.
00:04:54So things will hopefully get better.
00:04:56But I have no release date for that yet.
00:04:59But, yeah, I know it's horrible.
00:05:02And it's really sad.
00:05:04So for our courses, in case anybody's interested, we're using Teachable, which is essentially like a white-label course hosting platform, right?
00:05:13You can use it.
00:05:15So it's not a replacement for Udemy because on Udemy it's a platform with a bunch of different creators.
00:05:21And Teachable is basically a white-label platform where you can host your own course platform on top.
00:05:26But they provide the technical infrastructure.
00:05:29And, unfortunately, there are some long-standing bugs.
00:05:31And we keep on surfacing them to the Teachable team.
00:05:36They just don't care too much, I'm afraid.
00:05:38So, yeah, we are building our own stuff there, too, so that we have more control over the course viewing experience and stuff like that.
00:05:45But I don't know yet when we'll be done there.
00:05:48But things will hopefully get better there because I'm super not happy with how things work there, especially since our name is on it.
00:05:55So very sorry about that.
00:05:59And thank you so much for the very nice message, Gabor.
00:06:04It's really cool to read that you bought the Angular course seven years ago and that it gave you that solid foundation.
00:06:13So, yeah, thank you very, very much for that.
00:06:16And thank you all for all the nice messages.
00:06:19Also, Guzseska for the nice feedback on the GitHub Actions course.
00:06:24Glad to hear it's helpful.
00:06:28I use GitHub Actions for all my CI, CD workflows, for all my projects.
00:06:33It does have – there are some problems related to GitHub Actions, but I like the convenience of having it right there in GitHub.
00:06:40So, yeah, happy to hear that the course is helpful.
00:06:45And then the question, I learned React, Angular, Springwood, CI, CD, Docker, et cetera, et cetera, and now creating a big personal project and planning for open source control and learning DSA as my plan.
00:06:57Good in current timeline?
00:06:58I think in current timeline, what you should absolutely do – and I published a video about that, too, on my channel today.
00:07:04A bit controversial, maybe, but it's not meant to be that controversial.
00:07:10I label it Time to Wake Up.
00:07:11It's on this very channel on YouTube, if you're watching there.
00:07:15Because I think as a developer, it is really important to use AI.
00:07:23That is not new.
00:07:24I've said that before.
00:07:25You've heard that a dozen times.
00:07:27I will not bore you with it.
00:07:28But since the question was if that roadmap is right, yeah, I would absolutely build stuff.
00:07:32I would absolutely build stuff.
00:07:34That was always a good idea, even before AI.
00:07:37But now with AI, of course, I would do it with AI.
00:07:40And I would try to find out how you can be productive with AI.
00:07:43Because I'm convinced you're not productive by letting AI do everything or by just wipe coding away.
00:07:51You can do that for throwaway software, for little tools where you don't care about.
00:07:56I do that, too.
00:07:57If I just need a quick dashboard for something, for kind of – I know for presenting some internal data, right?
00:08:05I'm happy to wipe code that.
00:08:06I'm not publishing that.
00:08:07I'm not selling that.
00:08:09Totally fine.
00:08:11But using AI really just means finding a way of using it efficiently.
00:08:15And we're all still figuring out how to do that.
00:08:18You will find people that try to sell you how to do it.
00:08:22But there is a reason why I, for example, don't have a course where I try to tell you how to use AI.
00:08:31I have courses on cloud code and so on.
00:08:32And I want to share how you can use these tools.
00:08:35But we're still all figuring out how to effectively work with AI.
00:08:40And that changes every day almost because new models, new agent harnesses are released.
00:08:45And for me currently, for example, it's that Herder plus Pi thing.
00:08:49Pi, in case you don't know, is essentially an open, very extensible coding harness.
00:08:54And that works well for me.
00:08:56But I'm still also figuring out how to bring skills into the mix, which skills, how these skills should look like.
00:09:03Extensions, because with Pi you can build extensions and you can have Pi build its own extensions, which is a really nice feature.
00:09:11But that is all stuff I'm figuring out.
00:09:13And that is really what I think every developer should be doing.
00:09:16Figure out how to work with that stuff, how to work with AI.
00:09:19And of course it'll change.
00:09:21Of course it'll be different in a year or so.
00:09:24But the journey is super important, in my opinion.
00:09:29That is something I think every developer should do.
00:09:32Unless, obviously it's totally fine if you're just doing this for recreational work or if you're working at a company that absolutely bans AI.
00:09:41So if you're in an environment where you can totally ignore AI and you want to ignore it, of course that's fine.
00:09:48But I think, yeah, it's the future if you like it or not.
00:09:53And therefore I think every developer...
00:09:55Yeah, I did say kind of honor, by the way.
00:09:57Yeah, I fell back to German there.
00:10:00So yeah, I think that's important.
00:10:03But yeah, I shared this opinion over and over again.
00:10:06I just feel like there still are people that act like it's all just going to pass and it's totally not useful at all.
00:10:18And whilst it can be useless or since you can definitely use it too much or in the wrong spots, it's not useless.
00:10:25And that's just my point.
00:10:32Yeah, I haven't written a line of code this month.
00:10:36Yeah, for me it also has drastically shrunk, shrunk, whatever.
00:10:43Most of my time is now spent...
00:10:45I do write annotations.
00:10:48I do spend a lot of time crafting plans and so on.
00:10:51And occasionally I do jump into the code.
00:10:53But yeah, it has decreased a lot.
00:10:56What do you think about Elm-Elixir combination?
00:10:58In some sense, Elm doesn't fail in production while Elixir has similar properties as well.
00:11:03I can't say a lot about that simply because I haven't used it.
00:11:06And I don't want to comment on something where I have basically zero experience with.
00:11:12Sorry.
00:11:14I learned JavaScript from your tutorials on Udemy.
00:11:17You learned a lot.
00:11:18Super happy to hear that.
00:11:19Thank you so much.
00:11:20And do I speak German?
00:11:21Yes, I do.
00:11:23I want to learn everything that this new tech can do just so I can still find a job.
00:11:28Yeah.
00:11:29And join your be like Linus video this morning.
00:11:31Great take.
00:11:32Just trips fun for me.
00:11:33Yeah.
00:11:33I mean, the fun part is important.
00:11:36Some of you may know that I created a video a while ago already, which is still 100% true, by the way.
00:11:46There's no joy video where I talk about how AI sucked the joy out of coding for me.
00:11:54And that is totally true.
00:11:56I did enjoy that part where you just wrote code, the flow state you could enter there.
00:12:02And that is gone.
00:12:03That is simply gone.
00:12:04Now, I am, I think, I hope, finding new joy in the building stuff.
00:12:10Right now, I'm finding a lot of joy in figuring out these different tools and how to set up a productive environment.
00:12:16And in building projects with AI and figuring out how to do that in a good way.
00:12:21That's a different kind of joy.
00:12:23I'll also be honest.
00:12:24If you wouldn't have AI, I wouldn't be sad.
00:12:26But it is here.
00:12:27So you can just move on or move away.
00:12:29But, yeah, it's not coming back.
00:12:32But I totally get that feeling of the fun being stripped away.
00:12:42Which subscription do you think is a better value for money, Claude or ChatGPT?
00:12:48Definitely the ChatGPT one because they reset limits almost every day right now.
00:12:52They also give you limit resets, which you can use on your own.
00:12:58So, yeah, the value for money, the usage you get out of the subscription is definitely better for ChatGPT, I would say.
00:13:06I have the $200 subscription.
00:13:08I don't know how it looks like for the cheaper ones, I will say.
00:13:11But I have currently, I have both because I wanted to play around with Fable and so on.
00:13:15I have both the Anthropic and the OpenAI $200 subscriptions.
00:13:19And you get more out of the OpenAI one for sure.
00:13:26Yeah, I had vibe codings.
00:13:29Are you saying integrating AI into a project or using it for development?
00:13:33I mean using it for development.
00:13:36You can, of course, also build products on top of AI.
00:13:38But I mean using it for development.
00:13:40But, and I don't know.
00:13:42That is really important.
00:13:43But I keep repeating myself here.
00:13:45Vibe coding is one thing you can do with AI.
00:13:49And it depends very much on how you define that.
00:13:52For me, and that is my definition.
00:13:54It doesn't have to be yours.
00:13:55But my definition, when I say vibe coding, means don't care about the code.
00:14:01And that means I don't look at it.
00:14:02But I also don't care about it in any way.
00:14:05I just build stuff.
00:14:07I care about the outcome.
00:14:10Fix errors if wrong outcome.
00:14:16And I also don't plan, no specs, etc.
00:14:21That is my definition of vibe coding.
00:14:24And you can do that.
00:14:25And as I said, it's fine for anything non-serious, I would say.
00:14:30Totally fine.
00:14:31But that is not the future.
00:14:34That is not the future.
00:14:36Let me be very clear.
00:14:37The future is something which was called,
00:14:40I'm not sure if we still call it like this,
00:14:42agentic engineering, right?
00:14:43So that is the future.
00:14:48And I don't care too much about these terms.
00:14:50But what I mean here is you care about the code.
00:14:54That does not necessarily mean that you read all the code.
00:14:58You probably read some of the code.
00:14:59But you will definitely care about the code.
00:15:02Which means you try to build a system
00:15:05where AI and you, the combination of both,
00:15:09produce good code, review the code.
00:15:11You can trust that you have a project that works well.
00:15:15You do create plans, specs, testing strategies,
00:15:22and so on.
00:15:22All that stuff.
00:15:24That is, I think, the future.
00:15:26And I think that is where developers should look at
00:15:29or should get it.
00:15:31That's my definition at least, or at least, or my plan.
00:15:38What do you try to finalize in the plan?
00:15:41Is it just requirements or you try to fix technical implementation?
00:15:44It's definitely also technical implementation.
00:15:46It's details about the tech stack.
00:15:48It's also details.
00:15:51I tend to create very detailed plans together with AI.
00:15:57So, basically, my flow, and maybe we can jump into a project later.
00:16:01But, basically, my flow is like a back and forth with AI,
00:16:05discussing, then I have it create a plan.
00:16:09I have a skill that gives AI guidance on how that plan should look like.
00:16:16And then I review the plan, fine-tune it over time.
00:16:20And I'm still figuring out how to best do that.
00:16:24Now, don't get me wrong.
00:16:25But the plan for me is to get AI to build a plan that's not too large,
00:16:30not too small, and that has been fine-tuned by me.
00:16:34And then I let it implement this.
00:16:36Of course, it depends on the complexity of the task,
00:16:38of how much time I spend on that.
00:16:40And then we have review cycles, AI reviewers.
00:16:44And then, of course, also I review the critical parts.
00:16:48That is how I try to approach this in the end.
00:16:54Yeah, do I have plans for courses about DevOps techniques?
00:16:59A DevOps course in general would be interesting.
00:17:02But I think the next big course, I have smaller courses.
00:17:05I got a Pi course coming up.
00:17:07The next big course will probably be a systems design course.
00:17:19The Discord private message, I didn't even know that I have turned those on, to be honest.
00:17:25I never read them.
00:17:26Maybe I remember and I'll take a look.
00:17:28But I can't promise it, unfortunately.
00:17:33But I'll try to remember.
00:17:37K3 is the same price as Sonnet 5, Big L.
00:17:40It depends on the capabilities, I guess.
00:17:42But I haven't read anything official yet.
00:17:44But maybe something occurred whilst I was streaming.
00:17:48I struggle with starting new projects using AI.
00:17:51The bootstrap process is strange for me.
00:17:53It takes me time to find me in this generated code.
00:17:55Maybe it's personal.
00:17:57Yeah, I mean, I'm struggling too.
00:18:00I think most developers, many developers are struggling.
00:18:03And it's super easy to fall into the Vibe Coding trap.
00:18:06Don't get me wrong.
00:18:07I mean, it's super easy to start with the intention of controlling stuff, setting up everything nicely.
00:18:14And then you find yourself prompting away and getting stuck into, getting sucked into that Vibe Coding trap.
00:18:21Or what I had, as I said, I'm experimenting a lot.
00:18:24I tried to build, right, the latest rage is loops, that you should be designing loops.
00:18:32So basically entire systems where the agent can work for a very long time on its own.
00:18:36So I figured, okay, I want to build a loop, an implementation loop, where I have a plan and I give it to an agent.
00:18:42And then the agent should spin up sub-agents, do the work, review the work, and keep on going until I get back a nice piece of implementation.
00:18:51Now, what happened is that because of the way I wrote my skills and my prompts, and maybe also because of the model, I don't know.
00:19:00It went down into a spiral where it kept on finding issues and issues, and they became more and more esoteric and niche.
00:19:11But I let it do it because I was interested in seeing where it would end.
00:19:16But I stopped it after, I think, four hours because I saw that it was just digging deeper and deeper.
00:19:21And I was building a very basic product.
00:19:24I was building essentially my own service for sending out emails on top of AWS SES, so for newsletters, for marketing emails.
00:19:35I was building my own service there.
00:19:37And AI kept on digging deeper and deeper.
00:19:40And I ended up with that project that was way too complex, that covered a lot of niche cases, which could never occur in that project.
00:19:49Because, again, it was my self-hosted email sending service, which I may open source, which may be interesting to other people then, too, for self-hosting, which may at some point be opened up for customers.
00:20:03But that's not what it was or what it is right now.
00:20:07And I think that is one of the traps you can – there are so many traps you can fall into.
00:20:11Because suddenly you get that employee, if you want to call it like this, that is to some degree smart and to some degree not smart, that is very eager to do a lot of work, at least the more recent models are.
00:20:31But that doesn't always think through how to tackle certain tasks, to put it like this.
00:20:42And also an employee that when you ask it, hey, are there issues with that code, will always say yes.
00:20:49And working in an efficient way with AI, therefore, is complex.
00:20:54Now, I still think, though, that it's important to figure out that stuff because clearly you can be faster if you have something, an AI agent, that can write code as quickly as it can.
00:21:08That can't be worse.
00:21:10But, of course, it can be worse if you don't control it in the right way.
00:21:14But that is a learning process.
00:21:15This is a brand new technology.
00:21:17This is all evolving so quickly.
00:21:19It's normal that it's not totally figured out.
00:21:22That's my take on it, at least.
00:21:24And I just don't see how the old world of writing everything by hand is going to be better going forward.
00:21:33Again, I also don't see the world where AI does everything going to be the winner.
00:21:40That doesn't sound right.
00:21:43But, yeah, using it efficiently as a tool is something I'm aware and I think most developers are trying to figure out.
00:21:52Yeah, and since I read in the chat, I love to have full knowledge of the code.
00:21:59Yeah, I mean, that is one of the weird parts, too.
00:22:02You get detached from the code a bit.
00:22:05Even if you're not vibe coding, I feel like, yeah, you're a bit detached because I do, as I mentioned, review code.
00:22:17I look at some critical parts, but still it's not the same as if you've written it from scratch on your own.
00:22:22And especially if you start working maybe on two projects at the same time, which is kind of dangerous.
00:22:28But, yeah, you can lose track of what's happening in the code base.
00:22:33You may have reviewed stuff and you may be happy with the code quality or with where the project is going.
00:22:40But still, it's not the same, at least for me, as if you would have written it from scratch on your own.
00:22:47But also my, maybe my brain is just not good enough at memorizing stuff.
00:22:53That may also be it.
00:22:58Agentic engineering is too complicated to be a good SEO keyword.
00:23:01Maybe, yeah, I don't know.
00:23:04What's needed in addition for forward deployed engineering?
00:23:07Yeah, so I'll be totally honest.
00:23:11All these weird terms that have been coined, I can't comment too much on that.
00:23:20I don't even know what exactly that is meant to be now, especially in the AI age.
00:23:27And I don't care too much about all these hype terms.
00:23:30I think you just want to be a software engineer that's good at building software products.
00:23:36So products with code, digital products with code.
00:23:40And that involves AI, in my opinion.
00:23:44You want to be good at that.
00:23:45That means, I also talked about this before.
00:23:49I think that means that we'll have a shift where, for one, developers will become a bit more generalist
00:23:56because it's easier to work with technologies where you don't have a deep expertise, I believe.
00:24:03And I also believe that systems design and code architecture questions will matter more.
00:24:12Sorry.
00:24:13So that is what I believe will change a bit.
00:24:18So you could argue that software engineering was always about that, and that is probably true.
00:24:24I don't know.
00:24:25But of course, the term software engineering was used very broadly in the past and to this day.
00:24:32And it definitely involved coding.
00:24:34And the coding part, the writing code part, that is going away to a large extent, maybe entirely.
00:24:42But that does not mean that you don't interact with the code anymore, I would say.
00:24:46But in a totally different way.
00:24:49Bun rewrote the entire project using AI, which is crazy.
00:24:53Yeah.
00:24:53I mean, that has been super interesting, I believe.
00:24:57Obviously, a bit easier if you get free access to the anthropic models.
00:25:04But I found it pretty impressive.
00:25:06And I know from the comments below my YouTube video that not everybody did.
00:25:11But I found it impressive.
00:25:13You don't have to find it good or anything like that.
00:25:16But I found it impressive, the entire process that was set up there.
00:25:23I found it impressive.
00:25:31Is there value in having personal projects when applying for jobs?
00:25:34Or employers will just assume it's five-quoted slop?
00:25:37Yeah, good question.
00:25:39I mean, for employers, they obviously get flooded in applications.
00:25:42I think personal projects still can matter.
00:25:45But what I also said before on similar streams, when applying for jobs, you need to find a way of standing out.
00:25:52And that is harder than ever.
00:25:53And I think a big or an important part of standing out can be social media presence.
00:25:59Which does not mean that you have to become a YouTuber or anything like that.
00:26:03It can be frequently posting, sharing interesting stuff on X, on LinkedIn, or on YouTube, or whatever you like.
00:26:09But having a presence, I think that also helps besides having a portfolio or personal projects.
00:26:18I created my own spec-driven development framework.
00:26:26I hardly check any code.
00:26:29Yeah, I mean, as I mentioned, I think that is all something which will evolve and where we're still all figuring out how exactly that will look like.
00:26:39And caring about code does not mean reading all the code.
00:26:42Definitely not.
00:26:43And if you reviewed code, you didn't necessarily read all the code before, too.
00:26:47The data hasn't changed.
00:26:49But, of course, still, how you approach it does change.
00:26:54Any ETA on a system design course?
00:26:57Definitely still many weeks ago.
00:27:04In the future.
00:27:05I think interviews will also change.
00:27:18Right now is a super annoying time.
00:27:20Right now is a super annoying time.
00:27:22Because this technology is changing rapidly.
00:27:27Then you've got people like me who have the luxury of being able to always work with the cutting-edge models
00:27:34and tools and I can try out Pi, I can customize it, I can try out cloud code.
00:27:40Then you have companies which may not provide the latest models to their employees,
00:27:46which most likely will not allow employees to freely choose their harness, their coding agent.
00:27:52You have some companies that don't use it at all.
00:27:54You have companies that use bad models, but the managers expect miracles because they read from people like me
00:28:00or they see on videos from people like me what you can do with AI.
00:28:03They just don't realize that the AI in their companies is not what other people use.
00:28:07Then I'm sure you have interview processes that totally haven't caught up, that still act like it's 2020 or whatever.
00:28:17I think with every technology change, and this one is of course moving fast, there is a lot of friction and that is the super annoying part or one of the super annoying parts of this AI stuff or of this entire evolution that we're currently a part of.
00:28:37So I totally get the frustration.
00:28:39Unfortunately, there's not a lot of smart things to say, especially since it will probably differ from company to company depending on where they are in this journey.
00:28:55I'm a final year and learn React from your course, so please give me some tips and tell me what next to do that make my career bright in the AI era.
00:29:02Yeah, so basically what I said before, and by the way, this recording will stay online, so you can watch it again if you missed the beginning or anything like that.
00:29:10You should dive into AI and try to learn how to effectively work with it, and that is a process, that is not one afternoon, that is really figuring out or trying to figure out how to work with these tools, how to then produce good results.
00:29:26And I'm spending entire days and weeks trying to get better and figuring this out because it's all moving so quickly.
00:29:33My company supports Vibe Coding, but I do agentic engineering and work at the same speed as my Vibe Coder colleague, but he always sucks at fixing bugs of his own code, but I don't.
00:29:44Yeah, I mean, again, I don't think Vibe Coding is a good idea for any serious project.
00:29:51It is definitely okay for every, as I said, for throwaway software, but if you have to maintain something, I believe Vibe Coding is just a nightmare.
00:30:01I haven't figured out fronted dev and AI and back-end work is like mathematics.
00:30:05If you write good proofs, you are good and the code will work, but with UI, I haven't closed the gap, thoughts, ideas.
00:30:10Yeah, I mean, the disadvantage about everything UI involved, I mean, is exactly what you write.
00:30:16It's a bit harder to validate.
00:30:18Now, AI, of course, can use your browser.
00:30:23You can use tools like Playwright or Agent Browser by Vercel.
00:30:28There are ways for that too, but my experience is that there still is a lot of manual testing involved.
00:30:38There are huge differences in models regarding how good they are at UI work.
00:30:44If we're really just talking about UI work, if we're talking about the logic behind the UI, that can, of course, also be tested with unit tests and so on.
00:30:53So, that is what I'm also doing there.
00:30:56Are we going to see more employers using a degree filter to filter out those that just build with AI?
00:31:03Yeah, that's possible.
00:31:04I don't know.
00:31:05I don't know.
00:31:05It's possible that employers will reach for degrees and certifications to filter out the swamp of applicants.
00:31:13I definitely see that danger.
00:31:15And, of course, it is a danger because I was obviously never a fan of just degrees and certifications being self-taught myself.
00:31:23I definitely see that as one possible option of what could be happening.
00:31:29At the same time, we all know that's, of course, not necessarily a good filter, but it could happen.
00:31:39Is there a lack of trust from clients now compared to before?
00:31:42I mean, if I want something from some company, I just wouldn't believe them because they are using AI.
00:31:46What is your experience?
00:31:46Yeah, I mean, that's a very interesting one because I have exactly the same feeling.
00:31:54If I would outsource something to some contractor, I would assume that they would probably use AI.
00:32:02And whilst that is not necessarily a problem, I also use AI, you kind of assume the worst, I guess, or I would assume that they're not necessarily doing it in the good way.
00:32:16Because as a contractor, you are incentivized, of course, to quickly handle many clients.
00:32:23So I think there will be some trust issues here.
00:32:25And I also think right now we're at a point in time where I know personally from some friends that you can make decent money by leveraging the fact that there still are businesses out there that don't know or don't want to build their own stuff with AI.
00:32:48So they need a new website.
00:32:49Let's say a dentist needs a new website.
00:32:53And they're still outsourcing this to contractors.
00:32:57And contractors keep on charging their old hourly rates and maybe charging 10 hours but only need two because of AI.
00:33:06So I think there is this opportunity right now.
00:33:10And it's not very ethical, I think.
00:33:13But I get why it's there and why it happens.
00:33:16But that, of course, will close.
00:33:18And the other side of the coin, of course, is that there are trust issues for that exact reason by companies, by people that have figured it out.
00:33:29Is DSA and open source contributions still relevant after AI?
00:33:42No, wait a second.
00:33:43I skipped something, didn't I?
00:33:49All right, here.
00:33:50I'll get back to the other question.
00:33:51And by the way, I see there is a question.
00:33:54Yeah, I'll get back to those questions.
00:33:56So the Pi course will hopefully come out next week.
00:34:02Do you have experience with Cream as the payment provider you have mentioned using Polar?
00:34:07No, I have no experience with Cream, only with Polar.
00:34:10Sorry.
00:34:13I'm learning system design and AI systems design but don't want to dive into AI, ML, all the mathematical stuff.
00:34:19But is it enough?
00:34:20So I have no AI, ML background.
00:34:22So when it comes to really the AI, ML stuff.
00:34:27So I can't comment a lot on that.
00:34:30For systems design, talking from an application development perspective, the mathematical stuff doesn't matter.
00:34:38For AI, ML, it may.
00:34:41But ultimately, I'm no expert there.
00:34:43So I can't judge.
00:34:47Pretty similar answer to the question.
00:34:49My goal is to become an AI, ML engineer.
00:34:52But I am lost on where I should start.
00:34:53Do you have any tips to what I should focus on?
00:34:57Not my path.
00:34:59I'm not an expert there.
00:35:01So I'm probably the wrong person to ask.
00:35:04Sorry.
00:35:05Is DSA and open source contributions still relevant after AI?
00:35:13Right now, for some companies, it probably still is relevant because they are still looking for it.
00:35:20And I personally was never a big fan of how some interviews were conducted and the importance of that DSA stuff.
00:35:29But I know that there definitely are developers that would disagree.
00:35:33I think it will become less important.
00:35:34What will stay important is your ability to prove or show that you can logically think about a problem.
00:35:42That you can design a solution, a software solution.
00:35:47And data structures and algorithms may help with that or maybe something employers reach for.
00:35:56But I think ultimately employers will accept or realize that you can't really expect to hire developers that are good at leveraging AI.
00:36:06And at the same time, be super good at writing all those data structures and algorithms by hand.
00:36:12And of course, on the other hand, you could always make the argument that you should be able to do that in pseudocode, even if you don't write code in general every day.
00:36:24But I don't know what the long-term shift will be like there.
00:36:28I would definitely not worry about that at all.
00:36:31If I were hiring, I would not care about that at all.
00:36:34I would want to see how a person works with AI.
00:36:38If they're just wipe coding away or if they have a plan, if they have a strategy, if they really tackle a problem in a way that makes sense for our business.
00:36:47That is what I would care about much more.
00:36:49But that's just me, of course.
00:36:51Stromnitz Berlin has just removed the certificate requirement in favor of work experience.
00:36:56Yeah, I've always been a big fan of just proving that you have some experience, that you work with something before.
00:37:04I think that's worth more than any certificate, but certifications, a certificate will always matter for certain companies.
00:37:15That is just the thing.
00:37:18Since I use agentic coding, I feel I need to think deep and make constant decisions and judgments on what to do and how did it get implemented.
00:37:29Prior to this, it was 20% of work and coding itself was relaxing and fun.
00:37:34As Jeff Bezos said, you can make only so many good decisions in a single day.
00:37:38What is your experience?
00:37:39Shouldn't devs who do productive agentic work get more work breaks, more off days?
00:37:44This is an excellent question and I think observation.
00:37:48And it's exactly the same for me.
00:37:51You absolutely nailed it, midi-broadcasting.
00:37:56It's exactly the same for me.
00:37:57The writing code part was the relaxing part for me.
00:38:03The flow state.
00:38:04Now, the nice thing was that whilst writing the code, you also made decisions.
00:38:09Not big ones all the time.
00:38:10Sometimes you did.
00:38:11But that really was, as you said, the relaxing part.
00:38:14Now, it's just the tough decisions.
00:38:19And especially if you start maybe working on three projects at the same time, you're constantly getting pings by your agents that there is some plan to review, some implementation to review.
00:38:32Maybe the feedback of some reviewer agent to review.
00:38:36And you can, of course, just brush them off and say, yeah, yeah, implement the plan.
00:38:40Yeah, yeah, create a plan for fixing those issues that were found by the agent.
00:38:46But then, of course, you're entering vibe coding territory.
00:38:51Now, that's just my take.
00:38:52Maybe in a year we work with complex loops where we just operate on a high level because tools and models are so good.
00:39:02I don't know.
00:39:02But right now, that doesn't work for me.
00:39:05But as you wrote, you're constantly facing those complex decisions.
00:39:09You suddenly need to start reading a plan again.
00:39:11You think, okay, what am I working on here?
00:39:14You need to challenge the plan because if you just accept it blindly, well, then you don't need to look at it at all, right?
00:39:21So I totally agree there.
00:39:23And I think the right thing would be to take more breaks or to maybe blend in some other work.
00:39:30But I'm super bad at that.
00:39:33And I find myself really exhausted quite frequently because I just get all these agents doing that stuff.
00:39:41And I'm just constantly facing that stream of complex decisions.
00:39:48So I'm probably doing it wrong.
00:39:50But, yeah, that's a very good point.
00:39:55If you bought a course on Udemy, you can't access it on Academind, no.
00:39:59And vice versa.
00:40:00These are independent platforms and we have no controls over enrollments or anything like that on Udemy.
00:40:07Plus, and I'll be very honest here, anybody who buys a course on Udemy, of course, we pay a significant share to Udemy.
00:40:22Once we bring somebody on our platform, we are responsible for all the support and the technical support and so on.
00:40:31On Udemy, that's Udemy's responsibility.
00:40:34So, of course, if you buy something on Udemy and then you would come to us, we would have gotten a fraction of the amount we would have gotten from you if you bought it on our site.
00:40:44But now we have all the responsibility.
00:40:46So, even if we could, we would not be super pumped about migrating people from Udemy to our site for free.
00:40:54But the question isn't in the room because we technically can't.
00:40:59So, just a piece of honesty here.
00:41:01But that's not something we could do.
00:41:07I'm a huge fan.
00:41:08Thank you for what I've learned.
00:41:10Thanks to you, I've learned React, NextChase and Go.
00:41:12Thank you. Thank you so much for the very nice feedback.
00:41:14Super happy that the courses were helpful.
00:41:18I'm a full-stack engineer with seven years in Laravel, Node.js, Vue, and so on.
00:41:22What should I focus on next to maximize my career growth?
00:41:26Yeah, essentially what I've said before, so I'll keep it short to not bore everybody else.
00:41:29But really, how to work efficiently with AI.
00:41:33I think it's the thing every developer should try to figure out right now.
00:41:37Not 24-7, obviously, but seriously trying to figure out how to effectively work with it, which is harder than it may sound, at least for me.
00:41:47But that, I think, is something we should all try to do.
00:41:56Do you see WebAssembly fitting into front-end development?
00:41:59Will it eventually become as common as JavaScript for application development or stay limited for specific performance-critical scenarios?
00:42:06A couple of years ago, I would have said that WebAssembly will probably stay limited for performance-critical scenarios.
00:42:14I don't think that's necessarily the case anymore.
00:42:19I don't think it will overtake JavaScript or become more relevant than JavaScript for front-end code execution.
00:42:28But due to AI, I think many barriers of writing code and compiling that code to WebAssembly and debugging issues there are crumbling or are just not there anymore.
00:42:45And therefore, I could see that being more relevant in the future, but still kind of niche, I believe.
00:42:59I wait every Thursday for you.
00:43:01Yeah, I'm back.
00:43:02I'm sorry that I wasn't there the last Thursdays.
00:43:04I said it initially.
00:43:05Sometimes life gets in the way, but I'm also very happy to be back on stream.
00:43:11A course on LangGraph and LangChain, excuse me, not planned because I barely worked with it, worked a bit more with it three years ago, but I'm no expert on it.
00:43:25So no course planned.
00:43:27Do you think there are a lot of tech support, site reliability, engineer types of jobs, people debugging and fixing software rather than building?
00:43:34Yeah, I think that is maybe something.
00:43:38So if we're just talking about.
00:43:41So fixing software, of course, is also part of building it, I would say.
00:43:45If we're just talking about debugging and analyzing, I feel like with AI, that may be something AI is going to do more and more.
00:43:58Cheers for the vids maximum.
00:43:59I appreciate your insights.
00:44:00Thank you.
00:44:01I appreciate the positive feedback.
00:44:03Much, much appreciated.
00:44:04Thank you so much.
00:44:08GitHub Copilot takes endless money, even with the 7000 token pro plus.
00:44:13Yeah, I'm not using GitHub Copilot anymore.
00:44:17For me, the best value you can get right now is the chat GPT subscription.
00:44:23So the OpenAI one, because they reset the limit so often.
00:44:28The Anthropic one is okay.
00:44:30It's not as bad for me, at least, as some people make it sound.
00:44:33But you definitely get a better deal with chat GPT.
00:44:38And Grok is also not bad, especially if you got an X subscription, which I happen to have.
00:44:45It's actually, you get decent usage out of it.
00:44:53When I build my own projects with AI, it happens so that I have some really advanced code, which is really good.
00:44:59But it is hard for me to understand it.
00:45:01What would you suggest?
00:45:02Yeah, so discuss the code with AI.
00:45:04I mean, AI is not just good at writing the code.
00:45:07You can also ask it, hey, what's that code doing?
00:45:09And then you should ask it, okay, how could we build that in another way?
00:45:13Why did we, and I mean we, it was the AI.
00:45:16But why did we, why did you build it like this?
00:45:19Challenge it, let it explain to you, and not just the code, but also why that code was written, what it does, what could have been done instead, and why that other approach wasn't used.
00:45:28I think that's a great way of learning more about the code AI wrote and about learning coding in general, if you're still in the learning part.
00:45:37And even if you know a certain technology, a certain library or language already, it might still be a good idea to occasionally challenge AI and ask, hey, why did we build it like that?
00:45:48Why didn't we do it in another way?
00:45:51If you know about another way, you could ask, why did we not use, why did we not use this approach?
00:45:56Now, in my experience, often AI will say, oh yeah, we should have done it like this.
00:46:01But you can always ask, okay, but why should we have done it like this?
00:46:04Why is this approach bad?
00:46:05Why did you pick it originally if it was bad?
00:46:07And I mean, AI still has a tendency of always or very often trying to please you, but you still get some insights there, and you can definitely learn more about the code this way, I would say.
00:46:24Do you have to be a developer programmer to be an agentic engineer?
00:46:27I think and hope so.
00:46:30Yeah, I mean, I think it's just the next evolution step of a developer, agentic engineer.
00:46:38So nowadays, I wouldn't learn coding anymore in the way I learned it before.
00:46:45Instead, I would learn it for the AI age, which means less code written by hand, but still an understanding of what's going on in your software, in your code.
00:46:57Still an understanding of patterns that you may use or of building blocks that make up software and how to plan good software.
00:47:07All that stuff, an understanding of which issues or potential problems to look out for, but also which problems to ignore.
00:47:17Because if you just ask AI, hey, review that code, be super critical.
00:47:23What's your opinion?
00:47:24It will surface a lot of problems that may not be problems for you.
00:47:27And having that judgment, I think, is super important.
00:47:31Because otherwise, you'll end up with complex software that becomes more and more complex over time.
00:47:37And that's rarely the right call, I would say.
00:47:43The one thing that annoys me in the agentic coding is that most of the time, I forget to let AI know that I made manual changes to the code.
00:47:50And next prompts completely ignore my changes and erase them.
00:47:54Oh, yeah, I know that.
00:47:56I've been there, too.
00:47:58I constantly, you need to remind it.
00:48:00Or what I do is, if I make changes to the code, I typically, if it makes any sense, I just commit it.
00:48:05Because it doesn't erase or revert commits, in my opinion, in my experience.
00:48:11But, yeah, I know that if you just change something and it's uncommitted and you then prompt something, it'll just, it's very likely to discard your changes.
00:48:21Do you have any suggestions for skills?
00:48:23Have you tried skills from Matt Pokok?
00:48:25No, I have not.
00:48:26I am a, I'm building my own skills.
00:48:31So, I'm not a big fan of installing pre-built skills.
00:48:36Building skills is not trivial if they should be good.
00:48:43It's easy to build skills that don't add a lot.
00:48:46But it's also not super hard.
00:48:48A good skill describes a certain problem or solves one specific problem, I would say.
00:48:54And then it's relatively concise, clearly structured and actionable.
00:48:59And you can use AI to let it write a skill, but you need to give it guidance.
00:49:04And I think what's also super important is that when you have your agent skills, you must evolve them over time.
00:49:10And I think the danger is very high that if you install a skill, it's just there, you forget about it.
00:49:17And even if it maybe seemed to work well initially, the results you get out of it may be worse in the future.
00:49:25Because you're working with another model that's running in a different harness, has a different system prompt, whatever.
00:49:32And suddenly your skill is not doing that well anymore.
00:49:37If you built it your own, maybe there is a higher chance of you remembering it that, hey, you should update it.
00:49:44So, I have my own repository with my own skills, which I keep updated.
00:49:48And which I refine over time because all the time I realize I'm not happy with how that worked out.
00:49:54I'm not happy with how that plan was built.
00:49:58And therefore, I try to refine those skills.
00:50:02And that's why I want to build my own skills.
00:50:04I want to really tweak them.
00:50:05And that's part of that learning process, learning how to give my AI agents the right tools.
00:50:11And skills are like tools in the end for it to be effective.
00:50:16So, that's my take on skills and why I don't use skills by other people.
00:50:21You can use them for inspiration.
00:50:23And I don't mean the skill content, but really just the name or the goal of the skill.
00:50:27You can use it as inspiration, but I'm a fan of building my own then.
00:50:34My CV is honest, but how to compete with many candidates who often lie a lot in their CVs in the current challenging job market.
00:50:40So, I think one thing that's important about CVs, and that has always been the case, and also not just in development, is yeah, you want to be honest, but you want to be bold.
00:50:49Maybe that's already the case for you, but that's important.
00:50:52You want to be bold and not defensive.
00:50:55Now, lying is not the goal, but bold.
00:50:58And then, as I mentioned before, to stand out in the current job market, I think you will need some presence somewhere.
00:51:08X, LinkedIn, YouTube, doesn't matter.
00:51:10You don't have to become an influencer or a content creator.
00:51:12But being active, being visible there, I think can also help because the best thing that could happen to you, of course, is that you get in touch with someone there that then can help you land a job.
00:51:27That's not necessarily the goal, the visibility alone is worth a lot already, or can't be worth a lot combined with the other things, but it could happen, and that's just another chance or another shot you have at getting a job.
00:51:40And I know that this is not a great suggestion, it's just that the current job market is simply brutal, and I'm just trying to give my honest thoughts on what you could do there.
00:51:50How are we avoiding overworking and burnout?
00:51:53Yeah, I think deliberate breaks is important.
00:51:57Deliberate breaks is important, and finding joy in what you do.
00:52:02So I mentioned before, for me, some joy has been taken away by AI because I liked writing code by hand, and I'm trying to find joy in building with AI and tweaking those agents and getting good results out of them and learning that.
00:52:17And I think that is important because if you enjoy what you do, the danger of burning out is way lower than if you don't enjoy it.
00:52:24So that is my plan.
00:52:26And I know it sounds dumb, but it's not always easy, you can't force yourself to enjoy it, but you have to find joy in some aspect of what you're doing, I feel like.
00:52:38Thank you so much for all your input.
00:52:39Thank you so much for all your input, you're welcome, I'm happy that I could help.
00:52:45Download the courses from torrents, then paid for you on Udemy.
00:52:48Yeah, thanks for paying.
00:52:50I mean, yeah, pirating, not that great.
00:52:54Very happy that you enjoyed them and then decided to pay.
00:52:58So that's much appreciated.
00:52:59Thank you.
00:53:01Do you recommend React or Angular for a big web application build with many little teams and why?
00:53:06Probably React these days because it is more popular, it has higher usage.
00:53:13AI is very good at it, though AI can really be good at any technologies nowadays if you give it the right sources and skills and so on.
00:53:21But, yeah, I think React is just a safe choice, not a bad choice, yeah.
00:53:27How to work efficiently with AI, indeed, because personally I use AI sometimes because I feel dumb when using it a lot.
00:53:33Yeah, I mean, that's exactly the trap.
00:53:35You can easily use AI too much, have all your agents doing stuff, and suddenly you entirely lose track and so on.
00:53:43And, I mean, what's a number issue, what I also don't have a solution for yet is skills change.
00:53:52And I definitely had way better skills when it comes to writing code, when it comes to, let's say, React, for example, since we just had it in the chat a couple of years ago than I do now.
00:54:05That has definitely changed.
00:54:06Now, that does not mean that I don't know React anymore or any best practices or patterns, but it changed.
00:54:13It definitely changed.
00:54:14And I think that's just part of that evolution.
00:54:18Our skill set changes, I guess, but it feels super weird and also a bit wrong, to be honest.
00:54:29Thank you so much, Vlad.
00:54:31Thank you for your years from Angular 1.3.
00:54:34I learned from your course and have pro jobs and earned very well in Serbia.
00:54:38That means a lot to me.
00:54:39It's always exciting to read from someone who thanks or also partially at least thanks to my jobs.
00:54:45Of course, it's got a nice job and now has a good career.
00:54:48So that means a lot to me.
00:54:49Thank you so much for sharing it a lot.
00:54:51Really amazing.
00:54:52Thank you very, very much.
00:55:02So a couple of statements.
00:55:12Do you feel like with all this AI hype, we forget at all about old stuff like what's new in React, Next.js?
00:55:20It simply gets blurred.
00:55:21Yeah, I feel like this.
00:55:23In the past, every new React or Next.js or Angular release was a big thing.
00:55:29And nowadays, you care way less about it.
00:55:32That is definitely the case for me.
00:55:35Because, for one, you can just point your AI agent at the changelog and let it update the code base or figure out if there's anything in there that matters.
00:55:44But, B, because at least some of the changes that were introduced in the past by new versions of frameworks and so on were introduced to solve certain problems for developers.
00:55:58Now, of course, not all changes.
00:56:00There also are security changes, performance changes, and so on.
00:56:02But it was not uncommon that a new feature was added to make a certain thing easier for developers.
00:56:10Think of Angular signals have a performance impact, too.
00:56:17But think about maybe something like signal forms.
00:56:22I'm not sure.
00:56:23There may also be a performance impact there.
00:56:26But definitely there are these features that are being released that are there to help developers build an application in a better way, in a more efficient way.
00:56:37And that simply doesn't matter as much anymore.
00:56:41That simply is not as important anymore.
00:56:44And in theory, I believe, you could nowadays release a framework where writing code is really clunky, but it produces amazing, high-performance applications.
00:56:57And it would maybe be better than what we had before because the fact that writing the code is cumbersome and that maybe it's not enjoyable as a developer
00:57:10just doesn't matter anymore because you're not writing a lot of that code.
00:57:14So, I think that has shifted.
00:57:18And therefore, it's less exciting when a new version of something is released.
00:57:22Because, let's be honest, all those performances, all improvements, yeah, we took those for free.
00:57:27That was great.
00:57:28But it was the new syntax, the new API, the new feature that was exciting.
00:57:33And that definitely changed for that reason, I think.
00:57:41It's just good to know that I'm not the only one feeling more mentally exhausted and annoyed after AI coding session.
00:57:48Yeah, I think I would imagine that many developers feel like this because it is mentally taxing and definitely challenging, depending on how you do it.
00:57:57If it's just a slot machine style where you just pull and pull and you don't really do anything else than just say, hey, make it blue, make this go bigger.
00:58:08Maybe then it's not exhaustive.
00:58:10But if you make decisions, if you read plans and so on, yeah, definitely that can be exhausting.
00:58:17How about agentic-based apps expert like LangChain, LangRef?
00:58:22If I'm an expert there, I'm not.
00:58:24I'm not a LangChain, LangRef expert.
00:58:27Now, definitely one area which I see become potentially likely important in the future is being able to build AI agents.
00:58:38Like, for example, there is the EVE framework by Vercel.
00:58:42There also is the FLU framework by, I don't know.
00:58:46And these are frameworks that help with building AI agents.
00:58:51So now here the goal is not to have AI agents that build you a website, but instead you build a finance agent that is good at pulling financial documents from your Dropbox and preparing summary reports and stuff like that, whatever you ask it.
00:59:08And I think that will be an important area too in the future.
00:59:12So you don't just build web apps, mobile apps, low-level software, but you also probably build such agents in the future.
00:59:21Because I imagine that many companies, all companies have use cases where they want their own expert agents, which have their very specific tools and instructions and maybe skills and maybe a knowledge base and a sandbox to run in so that they can do certain tasks.
00:59:40That could be customer support, that could be, yeah, what I just said, I built an agent with EVE actually, because I am considering a course on that too.
00:59:50I built an agent for us internally at Acatamine where we can easily chat about all our financial documents, all our accounting documents, all the invoices and so on, because we obviously have to do accounting and so on.
01:00:04And every month we have to prepare some documents there and we have to check them for completeness.
01:00:09And that's now done also with help of such an agent where we can just ask, hey, accounting is due for this month.
01:00:15Can you please check the documents in this month's folder for completeness?
01:00:19Then it does its thing.
01:00:21It gets back to us and maybe says, hey, looks like this invoice is missing.
01:00:24And then we can say, okay, no, that's missing on purpose or yeah, good catch.
01:00:28Here it is.
01:00:28And I think agents like this for all kinds of purposes will be a thing in the future and building them, of course, also is something AI can do.
01:00:40But again, with guidance by developers, because it's just another type of application.
01:00:45And I definitely see that being an interesting area, along with all the other areas we already have in the future.
01:00:56That sounds like running your own agent harness like Pi makes it worth it.
01:01:00So for me, using Pi is really, really fun because you can extend it so well.
01:01:07And that is part of where I also get my fun out of right now.
01:01:12Are skills sensitive to the AI model you use?
01:01:15Have you found that you need to adapt skills to a specific model?
01:01:19So in the past, I definitely had to try to adapt, for example, the description,
01:01:25because some models were way more likely to use a skill than others, but that has changed.
01:01:31They're all pretty good at calling and invoking skills now.
01:01:35But yeah, I definitely tweak them all the time.
01:01:37I definitely tweak them all the time, not just to the mod, but also to the harness.
01:01:43So as I said, for example, I am using Herder to have multiple agents potentially running in there.
01:01:52And I built my own skill where I instruct agents to spin up sub-agents using Herder because Herder has built in support for agents and so on.
01:02:06So I built a skill for that, and I adjusted it to work well there.
01:02:14Claude Code and Fable therefore, of course, on the other hand, has its own sub-agent logic.
01:02:21So for Claude, I built a skill that helps it decide when to use sub-agents because I don't want to spin up an endless amount of pointless sub-agents,
01:02:34but I don't tell it how to do it because it has its own built-in logic.
01:02:37And so stuff like that is kind of, yeah, I do adapt those skills to different, mostly to different harnesses, not so much models anymore.
01:02:52Vibers are just NFT bros.
01:02:54Yeah, there may be some overlap there.
01:02:58I've fallen out of love with web development recently and looking to transition into DevOps.
01:03:02Any insights on how AI is affecting that side of the industry?
01:03:06So my take on that and from what I see on X and what I've heard is, obviously, it's affecting that too, depending on how you define DevOps.
01:03:16I mean, if you, how do you put it?
01:03:21So in DevOps, just like in coding, I would say, knowing what's happening is worth a lot.
01:03:29That being said, of course, you can use AI to build your Docker images, to maybe even help you set up your VPS or configure stuff on your VPS.
01:03:40I'm doing that.
01:03:42But it's pretty nice if you know what to ask for, what not to ask for, and to understand what the AI did there.
01:03:51It's interesting to take a look at those Docker compose files and Docker files, because again, AI may maybe conjure something which is just way too complex, which fails to do what you want it to do.
01:04:05And I'm not talking about errors where maybe building the container fails, the image, but instead where you end up maybe with an infrastructure, with a setup that's way too complex or that has certain issues.
01:04:18I personally ran into an issue where on one VPS, I had like a Docker setup, and I had AI work on that, and I didn't review it as well as I should have.
01:04:30And a month or so later, I realized that my hard drive on the VPS was completely full, 100% usage of the hard drive.
01:04:41And the reason for that was that the way logging was set up in that Docker compose file, those log files never got cleaned up.
01:04:50There was no rule, no logic for removing them after a certain amount of days or a certain total size.
01:04:58And that's what I mean.
01:04:59That's not the worst thing, but it's also not a great thing for you to happen, for your hard drive to be fully used.
01:05:06But knowing what's going on is worth a lot, but of course, AI has an impact there too, just as it has everywhere else.
01:05:17Because ultimately, if you break it down, everything is just code.
01:05:22So everything digital, it depends on code.
01:05:25It is powered by code in the end.
01:05:28So in the end, AI can help with all of that.
01:05:31But again, in conjunction, in combination with a human, with an expert, in my opinion.
01:05:39On skills, your views make very much sense.
01:05:41I built skill for a code review, fine-tuned it over a month.
01:05:44It was using Sonnet, Opus, now I'm updated for five.
01:05:49Yeah.
01:05:49So skills really need to evolve and you want to update them and you don't want to overdo it.
01:05:54Or I don't want to overdo it.
01:05:55I don't want to have like 40 different agent skills.
01:05:59I only have like four, five, six skills maybe.
01:06:03And essentially, a good proxy is whenever you find yourself asking AI for the same thing over and over again,
01:06:10or you're annoyed by the same thing over and over again, that may be a good opportunity to add a skill.
01:06:16Yeah.
01:06:17Or if you wanted to give it a certain behavior that's missing, good opportunity for a skill.
01:06:24But then again, that is a process.
01:06:25It's not like, yeah, write me a skill that does X, Y, Z.
01:06:28No, it takes effort to craft a good skill.
01:06:31And you should take the time to build a good skill because I totally get that it's not the most exciting thing ever.
01:06:37It's not the most exciting thing for me either, but it's a thing that gets used potentially by hundreds and thousands of agent sessions you're about to run over the next weeks, months, whatever.
01:06:52So you should spend some time crafting good skills that work well for your projects, for your code style, for your personality, for your models, for your harness.
01:07:02I think it is worth the effort.
01:07:08I feel that especially with the recent 5.6 release and the perma resets,
01:07:14I got about 20k in API inference downs the last week.
01:07:20I actually now took a few days off from work.
01:07:23Oh, so the exhaustion, you mean?
01:07:25Yeah.
01:07:26I mean, that's also a very interesting point here.
01:07:30I see that too, and I read it on X too.
01:07:33With all those, so OpenAI, they are constantly resetting usage limits, which is nice.
01:07:39You get more out of your subscription.
01:07:40But of course, it's easy to feel pressured to use that usage, right?
01:07:45So suddenly, you got 100% usage available again.
01:07:49So you got to use it.
01:07:50And I totally get that.
01:07:52But don't.
01:07:53Just don't.
01:07:54Don't let OpenAI dictate how you work, how much you work, and what you work on.
01:08:00The danger is that you burn out or you just get burned in general, even if it's not a full burnout.
01:08:07But B, it's very likely that you're just producing slop because you're in such a hurry to burn those tokens that you're probably not really thinking about what you're doing there or that you're not carefully reviewing what you're working on and what the output of the agent was.
01:08:23So chances are very high that you're going to throw away whatever you build with all those tokens anyways.
01:08:30So what's the purpose then?
01:08:32Sure, you used some capacity by OpenAI.
01:08:36You burned some of their money.
01:08:38Might be a good feeling.
01:08:39But ultimately, it's not worth to let OpenAI tell you how much you have to use it.
01:08:47Work on your stuff and don't let those usage resets dictate how to work on it.
01:08:54Love to see a course on agentic pi from you.
01:08:56A pi course is coming soon.
01:08:58And I also got more plans on sharing how I work with AI and showing that too and not just talking about it.
01:09:08How would it affect new developers beginning their journey where they get almost no need to learn fundamentals in order to build something right?
01:09:15I think current high-level languages will be assembly.
01:09:19How would it affect new developers beginning?
01:09:23Yeah.
01:09:25So if I understand the question correctly, you mean how it will affect new developers that now that they don't need to...
01:09:33So how can they get good fundamentals if they don't like hand code and so on?
01:09:39Is that what you're asking?
01:09:40Correct me if it's not.
01:09:44Learning how to code with AI, I think, is pretty, pretty challenging.
01:09:51And I don't have a great response for that yet.
01:09:54I'm obviously trying to figure that out as well.
01:09:58Because I totally see the problem that the danger just is that you're using AI, but you're not learning what's happening.
01:10:06It just feels nice to see results until you don't.
01:10:10So the way to go is really, and that's the hard part, to still learn what that code, that AI generated, is about.
01:10:22To still learn what makes up software.
01:10:26To still learn about building blocks of software and trade-offs.
01:10:29Now you may not have to make all decisions.
01:10:34There may be some smaller ones, which AI makes for you, which don't really matter.
01:10:39But still understanding what's happening there is important.
01:10:42Now you can use AI to talk about the code.
01:10:44I mentioned that before.
01:10:45And that, I think, is one part of how you should learn those fundamentals.
01:10:49To ask AI why it did something in a certain way.
01:10:51To ask for alternatives.
01:10:53To ask for explanations.
01:10:55Another building block, I think, and honestly not just because I create them, will be tutorials courses.
01:11:02Can be free ones.
01:11:03Can be paid ones.
01:11:04Because, of course, suddenly you've got someone else telling you something about a certain technology, a certain approach.
01:11:12And you don't have to ask for it because with asking, the problem is you might not know what to ask for.
01:11:18So, I think for getting good fundamentals, it's important for experts to share that knowledge.
01:11:25And then you can consume it.
01:11:26Then you can challenge your agents.
01:11:28You can ask your agents for explanations.
01:11:31And I think that is how you can still build good fundamentals.
01:11:34But it's a totally different learning process from before where you, of course, wrote code by hand.
01:11:38And you naturally ran into certain problems.
01:11:40And that was, like, a good way of learning.
01:11:43But that is not really how it works anymore, I feel like.
01:11:48How many agent skill files do you usually have per some moderate-sized project?
01:11:52Only a few.
01:11:52So, I have, like, four, five, six main global skills.
01:11:59And then I have per project maybe two to three skills.
01:12:05So, if I'm working on a project that's maybe using tanstext start and the fact library, and that's meant to be deployed on Cloudflare, then I may have a Cloudflare skill and a fact skill and a tanstext start skill.
01:12:18And I create them on my own.
01:12:21So, for example, maybe I already have one in one project.
01:12:24I would just tell my agent, hey, I got a skill and a fact skill in that project.
01:12:28Use that as inspiration.
01:12:30But also do some extra research on top to see if that skill really covers a fact appropriately.
01:12:36Dive into the latest change logs to see what changed.
01:12:40And then give me some ideas on how we could improve that skill.
01:12:43Then I take a look at those ideas.
01:12:44I add my own ideas.
01:12:45And then I maybe take that new skill and add it to the project in addition to the global skills.
01:12:50But it's only overall global and local combined maybe just, like, ten skills.
01:12:55I really try to keep it down.
01:12:57Not just because those skills consume some tokens.
01:13:00That's not a lot.
01:13:02Doesn't matter too much, I would say.
01:13:04But because I don't want my AI agents to constantly activate some skill, which sounds like it could be helpful.
01:13:11Because, for one, that will then consume more tokens.
01:13:14And, B, it may steer the model in the wrong direction.
01:13:17So, it's really, for me, it's a trade-off.
01:13:19Skills are super useful.
01:13:21But you don't want to have too many of them.
01:13:24At least that's my current take.
01:13:26Maybe after more experimentation or with better models, that will look different in a few months.
01:13:31You don't know what will be the case in a few months.
01:13:34Oh, I also, I think I skipped it.
01:13:37I'm very sorry.
01:13:37I saw the question before if devs are safe.
01:13:41And I think the answer is, so, for one, obviously, it's definitely bumpy out there.
01:13:48There have been way better times.
01:13:50That is obvious.
01:13:52The job market is horrible.
01:13:55That is all obvious.
01:13:57Only a small part of that is related to AI, I think.
01:14:00But AI will definitely have an impact.
01:14:03That is totally clear.
01:14:04Now, I think developers that leverage AI will be safe.
01:14:09Yes.
01:14:09I don't see a future anytime soon where companies only use AI and have basically no humans involved.
01:14:21Obviously, I can't look into the future.
01:14:22And maybe these models and tools improve so much that this is all wrong.
01:14:28But if that would be the case, it would not just be devs.
01:14:31Then all professions, most professions, would be in deep trouble.
01:14:36But I don't necessarily see that.
01:14:38I feel like it's the combination, like with pretty much all technologies, jobs evolve.
01:14:44But it's the combination of that technology and human input that can produce great results.
01:14:49And I think the developer of the future is simply a developer who knows how to efficiently work with AI.
01:14:56And the job will very likely look very different from what it is today and definitely from what it was four years ago.
01:15:02It's already different.
01:15:04But there still is a job.
01:15:05It's a different job, though.
01:15:07And again, that's what I mentioned before.
01:15:09For me, some fun definitely was lost.
01:15:11But I think I'm finding new joy in that new world.
01:15:17Would I have needed AI?
01:15:18No, but it's there.
01:15:27How do you catch up with all this stuff?
01:15:29It is hard for me to do so with some technologies.
01:15:32And I can't even imagine how hard it is for you as a teacher.
01:15:34Maybe you have some special source notifications.
01:15:36X is a really good source.
01:15:38I spend maybe a bit too much time on X, but I'm on X.
01:15:42There is a lot of noise there, but also a lot of valuable information.
01:15:47And then I really just build the stuff.
01:15:49I do work long hours.
01:15:51And I build a lot of stuff.
01:15:53I experiment a lot.
01:15:56And that's how I try out all these different things.
01:15:59And, of course, I can't try out everything.
01:16:01That's also part of the truth.
01:16:04But, yeah, I really just try to play with as much or experiment with as much as I can.
01:16:10I try to build serious projects.
01:16:12And I use X as a really good source for learning about new trends or technologies and opinions of people.
01:16:22I mean, sometimes there is stuff I don't have to try out because it's easy to tell after a few days that it was just hype.
01:16:28And now nobody's talking about it anymore.
01:16:31I may then still take a look myself.
01:16:33But, yeah, you see what sticks and what doesn't stick.
01:16:41If you hired a person right now for a junior role with zero years experience, what would you expect them to know and do on a day-to-day basis?
01:16:50I would expect them to know.
01:16:52So let's say I'm hiring for that web development, app development, including back-end development stuff we do.
01:17:03I would expect them to know about the core technologies.
01:17:07Let's say Node, Bun, React, PHP, whatever.
01:17:14Not all of them necessarily, but definitely in our case, for our stack, Node, TypeScript, React.
01:17:20They should know what these technologies are there for, how they work together, and what some core features are, some core concepts for React, components, state.
01:17:31Just the basics.
01:17:32They don't need to be able to write the code necessarily, but I expect them to know about it.
01:17:37I would expect them to know about auth.
01:17:41Which options do you have for adding auth?
01:17:44What are some challenges there?
01:17:46And you don't need to be a super deep expert there, but you need to show me that you have a good understanding of how you would build applications, which building blocks make up an application, what some potential trade-offs and challenges are there.
01:18:00And then, as I mentioned, the AI part.
01:18:02I would want to see how you work with AI.
01:18:06And I don't expect perfect results.
01:18:07If you're a junior, I don't expect senior-level experience.
01:18:11But I want to see that you're not just vibing away.
01:18:14I want to see that you challenge the output of AI.
01:18:18If you're not sure what the best way forward is, I want to see you discuss it with AI.
01:18:23And then you can still come to me and present me the options and say, hey, I discussed this with AI.
01:18:30These are like the three different options we came up with.
01:18:33That is my personal preference for these reasons.
01:18:39But I'm not sure.
01:18:40That is fine.
01:18:40But that would be my expectations.
01:18:43One thing about Claude usage, for some reason, the usage percentage resets for me early.
01:18:48I have the max plan and sometimes the weekly usage resets daily instead of weekly.
01:18:53That's a nice feature.
01:18:54I unfortunately don't have that.
01:18:57Kimmy K3 is released in web.
01:19:01Yeah?
01:19:03Is it?
01:19:04I would like to see an official announcement blog post.
01:19:12I know that I can probably use it there.
01:19:15I don't have an account, though.
01:19:16But I would love to see an official announcement blog post with all the benchmark numbers and so on.
01:19:29So, yeah.
01:19:32I have never had a problem with Gemini.
01:19:35Yeah, for coding, I'm not convinced by Gemini, I will say.
01:19:38Greetings from Belgium.
01:19:40I love your channel and your personality overall.
01:19:42You're awesome.
01:19:42Thank you so much.
01:19:43And thank you all for the nice messages.
01:19:45Thank you very, very much.
01:19:46It means a lot to me.
01:19:47Very, very kind.
01:19:51Thanks, Max, for everything.
01:19:53Thank you.
01:19:54Thank you all, as I just mentioned.
01:19:56These streams are fun.
01:19:57I really enjoy discussing these things, answering questions, hearing about different perspectives.
01:20:02Because that's a great way for me to learn, too.
01:20:05Because, as I mentioned, we're all just figuring this out, right?
01:20:07And I'm no exception, obviously.
01:20:10Am I using Open Router?
01:20:11Yes, I am.
01:20:12For example, for playing around with the GLM 5.2 model, I'm using Open Router.
01:20:19Yeah.
01:20:22I'm really curious why Google Gemini is struggling with their competitors in LLMs,
01:20:26considering the amount of data and experience they have in this industry.
01:20:30Yeah, I'm wondering about this, too.
01:20:34I don't know what Google is doing there.
01:20:37They have so much data.
01:20:39They clearly have a lot of very talented engineers and also data scientists, but ML engineers.
01:20:48They got all the experts.
01:20:50They got lots of data.
01:20:51I'm not sure why the Gemini models are so much worse, in my experience, and as it seems in the experience of many people,
01:21:01than the other top models.
01:21:05They do okay-ish well in benchmarks, but, yeah, they're just not on the same level.
01:21:11Maybe it's just not a priority for them to optimize them for coding because my understanding is that all that coding expertise is mostly post-training, fine-tuning, reinforcement learning,
01:21:25where you really get the AI model to be good at calling tools, using skills, and so on, and being aware of certain coding patterns, practices, and so on.
01:21:39So, yeah, I would definitely jump into building with AI, but you should, side-by-side also learn what the code AI generates is about.
01:21:53As I mentioned before, you should really understand the code that is being written, not every line, but you should have an idea of the application that's being built.
01:22:12It is definitely changing what you have to review, and I'm not saying that you have to review every line.
01:22:21I don't do that either, but far away from it.
01:22:24But you have to, I think you should be able to understand what the software, the AI built for you, is doing in general, which features, which APIs are being used in there, and not just blindly let it do its thing.
01:22:42I think that is important.
01:22:43You don't have to challenge every tiny decision.
01:22:46You don't have to enforce your code quality standards.
01:22:49I don't do that anymore.
01:22:50But you should have an understanding of the general structure or approach of your software, of your application.
01:23:03That is my opinion, at least.
01:23:06Do you write docs for a project so AI can read it?
01:23:11No, I don't write docs.
01:23:13I instead create, you could call it specs or maybe some, yeah, but you could call it docs, too.
01:23:22What I like to do is when I'm working on a new project, but also for existing projects, I like to discuss the project with AI, or if it's an existing project, I let it analyze the project.
01:23:33I discuss different ideas and options, and maybe things I didn't anticipate, or I should consider, and then I also push back, because just as with code reviews, if you ask the AI, what else should I consider, it will always give you more to consider.
01:23:50So I don't blindly accept it, but I have a discussion with AI until I feel like I have a scope and a project description I'm happy with, and then I tell it to put that as a summary with all the technical details into a markdown file in the project.
01:24:05That's often, or in most cases, my starting point, and then from there, we work our way through it.
01:24:10So I might then tell it, okay, so next we'll work on authentication for that.
01:24:15I need you to create a plan based on that project, blah, blah, blah, blah, and so on.
01:24:19So that's how I approach this.
01:24:22How do you know once you've already tried out enough breadth of topics and are determined that it's time to specialize, pick and specialize?
01:24:29You mean when discussing with AI?
01:24:32There really is no hard rule there.
01:24:34When we're talking about learning, I would really just have that discussion until you feel like you understand what's going on in a certain area of the code.
01:24:45If we're talking about planning, that really comes with experience.
01:24:51I know it's not a very satisfying answer, but it comes with experience when you're happy with the state of the discussion and the scope of the project.
01:25:02But let me know if I misunderstood you.
01:25:04I have to express my gratitude while I have the chance.
01:25:10I managed to become a developer all thanks to you.
01:25:12Thank you so much.
01:25:14I said it before, but I'm not getting tired of saying it.
01:25:17It means a lot to me.
01:25:18It really means a lot to me when I hear that my content, my courses were helpful in getting a job, making a career transition.
01:25:25That's why I created it.
01:25:26And so thank you.
01:25:27Thank you very, very much.
01:25:29It means a lot to me.
01:25:30Very, very, very much.
01:25:33Yeah.
01:25:34Agree with you on X.
01:25:36Too much noise indeed there.
01:25:37But sometimes there are posts talking about IT, dev, DevOps, and so on.
01:25:42And even on Instagram where I can discover some good tips.
01:25:44Yeah, I'm totally not on Instagram.
01:25:46So that may be a good source, which I'm just totally missing out on.
01:25:49But yeah, X is a good source, and I feel like I'm at a point where I'm following just the right people so that the algorithm is pretty good for me, and I get some nice insights and news from there.
01:26:05And again, of course, like on all social media, you have to ignore like 30, 40% of posts.
01:26:10But over time, you gain some experience on how to do that, I feel like.
01:26:17Regarding collapse, I don't do collapse.
01:26:19Sorry.
01:26:20What do you think about learning and using fast API for freelance work and building my own AI solutions as a business?
01:26:27I would say whatever you feel comfortable with.
01:26:29I think the tech stack is less important now than it was in the past because AI can be pretty good at working with most libraries, languages, and so on.
01:26:42So I would use whatever you personally enjoy and find interesting and are good at.
01:26:49That is my decision.
01:26:49So there is a bit more freedom there than in the past, I would say.
01:26:54Thanks for keeping us updated in this AI roller coaster.
01:26:57Thanks for being part.
01:26:57Thanks for being part of the stream, for watching the videos, and for being part of that overall journey with me, I think.
01:27:04Do you find GLM 5.2 to be as good as Opus 4.8?
01:27:09It was pretty good, but I will say since I didn't have a subscription for GLM 5.2, I was a bit too cheap.
01:27:17And I didn't use it closely as much, even close as much as I used Opus.
01:27:27So I used Opus way more because it's part of my subscription and GLM was not.
01:27:32So I have not gained as much experience on using it as I did with Opus.
01:27:39But where I used it, I found it pretty impressive.
01:27:43And yeah, I have a hard time.
01:27:45You sometimes hear people post, oh yeah, this model is better than Opus 4.8.
01:27:52And I personally have a hard time making such judgments.
01:27:56It's more the general vibe that kind of becomes visible over time, where you get a feeling for what a model is good at and what not.
01:28:10And again, it's also important, don't underestimate the hardness in which a model is running.
01:28:15It does have an impact.
01:28:16And the tools you give it and the skills, that all, it's that combination of things.
01:28:21And there are so many moving parts in there that for me, it's hard to say, yeah, this is better than that.
01:28:28I can say that Fable is a really good model.
01:28:31And it has a lot of taste also, oops, excuse me.
01:28:35It has a lot of taste when it comes to creating good-looking frontends, in my experience.
01:28:40GPT 5.6 is way worse at doing that, at least when used in Pi.
01:28:45I don't know about codecs, but in Pi, it's not that great.
01:28:52But it is really good at working its way through a plan.
01:28:56It's good at analyzing your code base.
01:29:00Again, when giving the right skills and tools and so on.
01:29:03So that's more how I approach this.
01:29:06It's hard for me to really say this is on the same level as this.
01:29:11Sometimes you see clear differences.
01:29:13Sometimes it's more blurry.
01:29:17Google has talented engineers and bad decision makers.
01:29:21Yeah.
01:29:22There may be some truth to that.
01:29:25Thank you for answering the question.
01:29:27Your content is great and really useful.
01:29:28Thank you so much.
01:29:29Thank you very, very much.
01:29:31I'm learning C Sharp now.
01:29:32I really like the way you teach, especially how you implement the features with covered topics and refactor later.
01:29:37Do you have any suggestions for C Sharp critics and instructors?
01:29:40I do not because I haven't taken any C Sharp course.
01:29:45I'm actually relearning C Sharp now.
01:29:48I learned C Sharp as one of my first languages.
01:29:52Well, no, not as one of my first languages.
01:29:55But when I got back into programming after university or got more serious about it again.
01:30:04So like 16 years ago, roughly, I started 24 years ago.
01:30:13And then I had a little break whilst I went to university.
01:30:18Where I didn't program as much.
01:30:20And then I got back into it and I learned C Sharp back then.
01:30:23And I'm relearning it now.
01:30:25But I'm doing that with a good old book.
01:30:27Because especially with all the AI stuff and so on, I sometimes enjoy just sitting on the couch, reading such a book.
01:30:34And yeah, maybe a weird way of winding down.
01:30:38But that works for me.
01:30:39So no, I have no instructor suggestions, unfortunately.
01:30:43Someone having worked many years in the world of SAP.
01:30:46Does it make sense to switch to web dev since so much is going on due to AI as a newcomer?
01:30:51I'm not sure.
01:30:52I mean, web dev is a pretty tough world right now.
01:30:55Because the job market is definitely horrible in dev in general.
01:30:59But especially web dev, of course, had a huge influx of developers in 2020 and the following few years.
01:31:07And also before that already.
01:31:09So there's, it's a pretty crowded job market.
01:31:12So for that, I think web dev has since, has seen better days.
01:31:17One big advantage web dev has always had and still, of course, has is when you're building web applications, you are building for a platform, the web, that is pretty much totally open.
01:31:36If you're building iOS apps, you have to go through Apple's review process.
01:31:40They may block your app.
01:31:41They may pull your app.
01:31:43What I always liked about web development is one, during development, you instantly see something on the screen because it's in the browser.
01:31:50That's not the case for other programming languages.
01:31:53If you're building something with C++, you see it in the terminal.
01:31:56But graphics programming is a bit more complex there.
01:32:00So that's nice.
01:32:01And if you're building a product, you can theoretically launch it and it's there for the entire world.
01:32:07The problem is nobody will know about it.
01:32:10Marketing is hard.
01:32:11But that's something I always liked and which I still like about web dev.
01:32:13But the job market has seen better days.
01:32:17I think that Google has already lost in this AI models for development war.
01:32:22And they try to focus more on general use for the majority, like Google questions, researches, et cetera.
01:32:28Yeah.
01:32:28I believe what Google is doing, I think they would like to be better in that entire AI for development race.
01:32:37Because otherwise they wouldn't be launching dozens of coding CLIs and tools where nobody knows what still exists and what was already shut down.
01:32:48But I agree that Google's endgame may very well be that they just, they have AI summaries, they have their AI chatbot right in Google search.
01:33:00They make all their products AI searchable.
01:33:04And they just follow their mission of making, as they would say it, knowledge easily accessible by everyone in the world.
01:33:13By also including ads there, of course.
01:33:17By the way, I'm pretty sure they're still also trying to figure out how to monetize this stuff.
01:33:21But yeah, Google may not care too much about coding in the end.
01:33:27Though I think there is one important aspect here.
01:33:30There is the theory, which does sound quite right to me, that of course developing, building better AI models will become easier if you have good models that are good at writing code.
01:33:45Because of course the part, the steps that are required to build a good model, preparing the research that goes into that and the actual model building, the actual code that is in the end still written for that.
01:34:08And everything related.
01:34:10That of course is something a coding model can probably do quite well.
01:34:14So of course you could see a flywheel where a good model makes it easier to build better models in the future.
01:34:22And if you don't have access to a good model, you can't enter this flywheel or you'll have a huge delay and your competitors will always be ahead of you.
01:34:32There is this theory and it doesn't sound totally bonkers to me.
01:34:36So that may be something which could become a problem for Google.
01:34:44Love your BUN course.
01:34:45Glad even with AI we still get informative courses from you.
01:34:48When is the systems design course coming out as well?
01:34:51Later this year is the plan, but it's still definitely quite a couple of weeks away.
01:34:58It's still in early stages right now.
01:35:01Have a nice stream, MIMX.
01:35:02Until next time, thank you for joining.
01:35:05See you in the next stream too.
01:35:07Thank you for being part.
01:35:09Do you try the grill me skill?
01:35:10It's perfect for creating specs.
01:35:12No, because I have my very own approach.
01:35:15And again, I want to figure stuff out myself there.
01:35:19And I want to find my own approaches, build my own skills.
01:35:23I'm not using any skills by anybody.
01:35:26I'm doing that myself.
01:35:28I take inspiration, sure, by understanding what certain skills do, not by copying them, obviously.
01:35:34But then I want to figure it out myself.
01:35:39Because ultimately, creating specs, deciding on the scope of a project, of the features, of the tech stack, the technologies, the approaches, and everything that goes in there, is not something I've...
01:35:56I have a pretty clear vision of what it takes to build good specs or good project descriptions, whatever you want to call it, whatever you want to hand to your AI.
01:36:07But I want to get there myself, depending on what I'm building, because that, for me, works really well.
01:36:16And I don't need a specific skill for me.
01:36:19Not saying that it can't be helpful for anybody else, though.
01:36:22And no course on LangGraph and LangChainPlane, because I have not used those technologies in quite some time.
01:36:29So, no expert in that.
01:36:35What a beautiful gift you have given to so many of us.
01:36:37Greetings from Leipzig.
01:36:38Thank you so, so much.
01:36:39That's just amazing to read.
01:36:41Don't know what to say.
01:36:42Thank you so very much, Jesus Maria.
01:36:46Casal Torres.
01:36:47Thank you very much.
01:36:49Hi, I'm coming from X.
01:36:50Welcome.
01:36:52Have you heard that China has banned AI girlfriends today?
01:36:55I have not.
01:36:56And I will say...
01:36:58AI...
01:37:00So, I've said it before.
01:37:02If we wouldn't have AI, so this form of AI, I wouldn't be sad.
01:37:08But it's here, so you have to adapt.
01:37:10That's fine.
01:37:11There are aspects of AI related to video generation, image generation, and other aspects, which I find troublesome.
01:37:21And not just because of the copyright stuff, but I'm not sure if there are good use cases for generating images with AI.
01:37:30The same is true for video, certainly.
01:37:33But stuff like AI girlfriends or...
01:37:37And I know that's not just about video and stuff.
01:37:40That could also just be a text, a chatbot.
01:37:42But I think it's really dangerous what can come out of AI there.
01:37:49I think it's really dangerous for some people to maybe get quite lonely, to enter darker areas of life, if you want to put it like this.
01:38:00We have the entire deep fake problems and so on.
01:38:04So, I think there are some problems where I understand why stuff gets banned.
01:38:10But, and I have some experience with bans and regulation in Germany and in Europe, but I think that will not be a good solution.
01:38:21China maybe has a bit more tools at their disposal to enforce bans.
01:38:31But ultimately, people typically find ways around bans and regulation.
01:38:39So, it can be part of what you do and it may be important.
01:38:44I'm not saying that.
01:38:47But I'm not sure if that will solve all the problems that come from this technology.
01:38:55May solve some, but I'm doubtful that it's the ultimate solution.
01:39:01But unfortunately, I don't have an ultimate solution.
01:39:05So, it may be better than nothing.
01:39:06But yeah, you could have an entire two-hour discussion or longer just about those negative effects.
01:39:14Any good UI library like Mui, but it is powered by Tailwind CSS.
01:39:23So, not one that I would be using.
01:39:29Shared CN, of course, is, depending on how you look at it, maybe not what you might be looking for.
01:39:37But I typically just use, I either use plain vanilla, modern CSS.
01:39:43So, because, again, AI is pretty good at writing that.
01:39:46Or I use Tailwind or I use Tailwind with Shared CN.
01:39:50So, not a lot I've used over the last years other than that.
01:39:54So, can't really recommend anything there.
01:39:56Sorry.
01:39:59Can you give some tips on how to market applications?
01:40:01You mean how to apply successfully?
01:40:03So, basically, what I shared before, I think it's more important than ever to find a way of standing out of the crowd.
01:40:09And one aspect of that or one approach could be to build an audience or build visibility on social media platforms like X-LinkedIn, YouTube.
01:40:23Not on all of them, but on some of them, at least on one of them.
01:40:28And, again, I get that this is not the most satisfying recommendation.
01:40:32But I think it can help with getting more visibility in a very crowded market.
01:40:39And anything that gets you more visibility is probably good.
01:40:49Which C-sharp book I'm reading?
01:40:52Yeah, let me quickly take a look.
01:41:01I think it's really just one of the first.
01:41:08Yeah, so it's really just, give me a second.
01:41:13Just this one.
01:41:15So, nothing fancy.
01:41:16I think it's one of the first books you find when searching for C-sharp on Amazon.
01:41:22Really in-depth.
01:41:24Really comprehensive.
01:41:28But, yeah.
01:41:29I mean, it's not like I'm learning it from scratch.
01:41:32Though you could do that with that book, too.
01:41:34But I'm just getting back into C-sharp.
01:41:37And really good for that.
01:41:38But, again, I'm weird.
01:41:40I like doing stuff like that.
01:41:42Yeah.
01:41:45How do you think WebMCP will evolve in the future?
01:41:48Yeah, good question.
01:41:49I mean, MCP in general is tough.
01:41:54The one big advantage I see in MCP is that adding authentication is easy so that the agent can act in an authenticated way and you have more control over that.
01:42:08Other than that, I feel like CLIs or direct API access where the agent can write its own code to make an API call is maybe more promising.
01:42:22But then again, this is all evolving and we're still early.
01:42:26And WebMCP kind of is a bit related to that.
01:42:30Pretty niche right now.
01:42:32And I don't know if it will ever be or ever become an important technology.
01:42:39We may see that and kind of hurts me to say that, but we may see that many traditional websites disappear instead of something like WebMCP becoming more of a thing.
01:42:55And we instead have more APIs and CLIs where agents can directly communicate.
01:43:00Because, of course, if I were building an agent that should order food, the most efficient way would be for the delivery service to have an API my agent can directly talk to.
01:43:12That's way more efficient than anything else.
01:43:16That could be an MCP, but why not just give me the API?
01:43:20Again, authentication could be a thing, but there also we'll see better solutions as this all evolves.
01:43:32Today, I read about an announcement of 1Password, which makes it easier for you to expose or to manage access to password-protected resources from inside Cloud Code.
01:43:45So that Cloud can request access, you grant it, but the agent, the model never sees the password or anything like that.
01:43:54So that authentication part will probably be figured out.
01:43:57So I'm not sure.
01:43:58Yeah, we may just see more, a trend towards more APIs and so on and less websites.
01:44:05Not this year, not next year, but over time, slowly and steadily.
01:44:16And, of course, not all websites, just to be clear.
01:44:22You could say, just like Shay Cormac in Assassin's Creed Rogue, I make my own skill.
01:44:27I haven't played that one, unfortunately, so I don't know how to say it.
01:44:33I tried Angular and, man, it's amazing.
01:44:35Yeah, I know.
01:44:36Angular, I like Angular a lot.
01:44:39They fumbled a bit, quite a bit, when they released it 10 years ago, Angular 2, the good old days.
01:44:50And React pulled ahead.
01:44:52Angular 2 was too clunky, too cumbersome, too complex, too much boilerplate, I think.
01:44:58So for many reasons, React pulled ahead.
01:45:02Angular 2 could be way more popular if the Angular 2 they released 10 years ago would have been closer to what it is today.
01:45:12But that doesn't help, of course, it is what it is.
01:45:15But, yeah, Angular, the modern Angular is pretty good.
01:45:19For system design, do you have any book or reference material recommendations?
01:45:23No, I had a look myself and I didn't find any great books except for one book.
01:45:35I haven't read it myself yet, the second edition.
01:45:39It's relatively new.
01:45:42Give me a sec.
01:45:50Designing data-intensive applications by Martin Kleppmann.
01:45:54The second edition is pretty new.
01:45:56It was released a couple of months ago, I think.
01:45:59I haven't read the second edition.
01:46:01I have read the first edition.
01:46:03And this touches on system design topics.
01:46:08And I find it really, really useful.
01:46:11So that is one resource I can probably recommend because the first edition was good.
01:46:18And I assume the second one is two.
01:46:21Any thoughts on Golang in 2026?
01:46:28Golang is a great language.
01:46:30And Agents AI is pretty good at it.
01:46:33So, I think, for the last five, six years or longer, I've worked with Go a bit over six years ago.
01:46:51So, it has been kind of popular, but then again, not so.
01:46:59Or maybe the better word is, it has been, or the better way of saying it is, it has been used quite a bit, but it hasn't been loved a lot.
01:47:09And I think that is still true.
01:47:11But Go is a great language.
01:47:15It's really great for building performant applications.
01:47:19I like most parts of its syntax, not all of it.
01:47:24And nowadays, it, of course, doesn't matter as much anymore as it used to in the past.
01:47:29But yeah, Agents being good at it is pretty helpful already.
01:47:32So, yeah, Go is a good language.
01:47:36Do you read your coding books in German or English?
01:47:39English.
01:47:40I read not just coding.
01:47:41I read pretty much all my books in English, but yeah, definitely coding.
01:47:44I've always consumed all coding-related resources in English because English is the main language for coding.
01:47:52I mean, even though we don't write the code by hand anymore, or at least some don't do that anymore, and myself included for the most part,
01:48:03even that's the case, all the words in the syntax are English and so on.
01:48:09So, I always found it weird.
01:48:12And fun fact, I created one German course in my entire career as an instructor.
01:48:18Nine years ago, I created a German version of my Angular course because Udemy asked me to, and it was horrible.
01:48:24I hated it because I constantly wondered, do I now translate this word to German or do I keep the English one?
01:48:36Like, variable is in German, variable.
01:48:39It's pretty close, but it's a different word.
01:48:42So, different pronunciation, obviously, and so on.
01:48:44So, I found it weird to translate it.
01:48:48I found it weird to not translate it.
01:48:51And I also find it weird if I read it in German.
01:48:54So, that's why I read it in English.
01:48:59I have been doing Angular for five years now.
01:49:02I'm thinking of jumping into back-end roles now.
01:49:04What technology should I get into?
01:49:05Also, do you think Angular has a scope in future?
01:49:08So, regarding which technology matters in the future, I think that will become less and less important with AI because it's easier to switch technologies and people just don't care as much about it.
01:49:20Now, historically, Angular is quite popular in enterprises, but React is definitely the more used library out there.
01:49:30So, that's something to keep in mind.
01:49:33But, yeah, Angular has a future.
01:49:37Google is running, I don't know how many services on top of Angular or is using Angular there.
01:49:42So, it has a future.
01:49:43But I think the choice of which technology you use for what, especially on the frontend, becomes less important in the future.
01:49:53But, obviously, still, there will always be popular technologies or ones that are particularly performant or whatever.
01:49:59But, yeah, Angular, I think, is a good choice.
01:50:01But not a lot has changed there.
01:50:04React is definitely used more, but Angular is also popular.
01:50:15Which course was it?
01:50:17It was an Angular course, but it's super outdated now.
01:50:21It's an Angular 2 course.
01:50:24As I said, I released it in 2017, and I haven't updated it.
01:50:30So, it's outdated.
01:50:32But if you want, you can buy it, but don't expect up-to-date Angular knowledge from it.
01:50:37Golang is the language for cloud and infrastructure and platform engineering, is it not?
01:50:41Maybe a bit.
01:50:42I wouldn't say it like this.
01:50:44It is popular there for sure.
01:50:46It can also be great for building CLIs, all kinds of utility tools or scripts you want to run, no matter if it's part of your infrastructure management work or for anything else.
01:50:58But, yeah, it is definitely good for that.
01:51:00And it is popular for that to some degree.
01:51:03That's not wrong.
01:51:06But you can also build web servers with Go, and it's great at that, too.
01:51:11Really high performance.
01:51:12So, it's a versatile language, and that makes it pretty appealing.
01:51:18Given that now even someone with limited technical skills can create impressive projects, how can a senior developer build projects that have clearly shown that you are an expert?
01:51:29So, it's the process where your expertise matters.
01:51:35It's the process, and it's the long game, so to say.
01:51:40It's easy to build a prototype that looks nice.
01:51:44You don't need any programming knowledge for that.
01:51:48It's quite a bit harder to build a product that is really production-ready.
01:51:53It's a lot harder.
01:51:55It's even harder to build such a production-ready application in a short amount of time with AI.
01:52:06So, there are different levels there.
01:52:08Wipe code a prototype?
01:52:09Easy.
01:52:11Wipe code production-ready?
01:52:12Extremely hard.
01:52:13Have some amount of knowledge and build a production-ready application?
01:52:18Doable, but will take a couple of, I don't know, days, whatever.
01:52:22Depends on the project.
01:52:24Being an expert, you can be faster.
01:52:27And then, of course, comes the maintenance part.
01:52:29And the part where you add features, where you fix bugs over time.
01:52:33Again, there you can shine, because with your knowledge, there may be fewer problems to begin with.
01:52:41And then you will probably have an easier time fixing issues or adding features.
01:52:47And I think that's how it shows.
01:52:48Not by how your application looks or anything like that, but it's this process that matters.
01:52:53Do you think future versions of frameworks will be more optimized for AI?
01:52:58Yeah.
01:52:59I'm 100% sure that we'll see future versions be more optimized for that.
01:53:04I think we'll see more programming languages and frameworks being released that are primarily
01:53:09built for agents and AI.
01:53:12And yeah, I 100% see that happening.
01:53:17Max, just wanted to let you know how much your courses have helped.
01:53:20I recommend them to everyone who joins my team.
01:53:22And they have helped me out earlier before I became a team lead.
01:53:25Thank you.
01:53:25Thank you, John Wood.
01:53:27Thank you so much.
01:53:28Means a lot to me.
01:53:29And makes me super happy to read messages like this, to hear that the courses had a positive
01:53:33impact on you and your team.
01:53:35And therefore, thank you so, so much for sharing this amazing feedback.
01:53:40Truly means a lot to me.
01:53:42Thank you very much.
01:53:43And thank you all for being part of the stream.
01:53:45It's actually a good point to end it, since I worked my way through all the questions.
01:53:50Maybe in the next stream, we'll do some actual coding so that I can show how I do that
01:53:55with AI.
01:53:57My current state, at least.
01:53:59But I gotta go.
01:54:00So I'll do that.
01:54:01Quick answer to the last question.
01:54:03I see GCP versus AWS.
01:54:05I prefer AWS because I've always used that.
01:54:08I find the Google Cloud Console horrible.
01:54:11The AWS one is horrible too, but Google is even worse.
01:54:15So yeah.
01:54:18But I just always worked with AWS.
01:54:21So Google Cloud may be amazing, and I just don't know it.
01:54:25So yeah.
01:54:26Thank you all.
01:54:27Thank you also for the very nice comment about the React course.
01:54:31American Dream in Europe means a lot to me.
01:54:37And yeah.
01:54:38Regarding Hetzner, I like Hetzner.
01:54:40I like the value for money you get there for renting VPSs.
01:54:46I think Hetzner is great just to really share that quickly.
01:54:50And with that, I'll wrap up.
01:54:51Thank you all for joining.
01:54:53The recording will stay online in case you want to watch it again or you missed the beginning.
01:54:57And I'll hopefully be back.
01:55:00Yeah.
01:55:00Next Thursday is looking good.
01:55:04So have a great evening, morning, day, whatever, and hopefully see you in the future.
01:55:10Bye-bye.