Chatting, Bun's Evolution & More!

MMaximilian Schwarzmüller
컴퓨터/소프트웨어구직/면접자격증/평생교육AI/미래기술

Transcript

00:00:00So, let's see, let's see, let's see.
00:00:24Foster encoding preset, I have that.
00:00:41So, let's see, I'm still getting that warning in OBS though.
00:01:01Encoding overloaded, but now I switched it down to a really fast setting.
00:01:06I wonder if it may just be an OBS issue, because there really isn't anything left running on
00:01:15my system.
00:01:19Okay, well, I guess we'll see.
00:01:28I hope it'll stay lag free.
00:01:34I got no idea why it's set, what the problem is today.
00:01:40Really got no idea.
00:01:49I've been streaming for, with the same setup for, I don't know, 50 times.
00:01:57And now all of a sudden it's all broken.
00:02:00Yeah, this will be, I guess, the last attempt.
00:02:05If it starts lagging again, I'll either turn this into a podcast or we'll have to retry
00:02:14next week, which would suck.
00:02:16Which would suck.
00:02:17So yeah.
00:02:18Hi everybody who's still there.
00:02:20I'm super sorry about all these problems.
00:02:22I really got no idea what's going on here.
00:02:27It's kind of fitting that when I want to talk about BUN, which is all about performance,
00:02:32I have everything but performance.
00:02:38It's completely lag free now.
00:02:40Okay.
00:02:41Let's hope it stays that way.
00:02:43I'm too afraid to touch anything now.
00:02:48So I guess for the rest of the stream, you will see the instructions chat.gpt gave me
00:02:54for fixing the problem.
00:02:56I followed this one here, by the way.
00:02:59Yeah, it's possible that YouTube is behind Twitch.
00:03:03Oh God, I touched something.
00:03:08Let's hope for the best and thank you so much for sticking with me.
00:03:15I don't think I can control the delay of the individual platforms.
00:03:19It's not like I changed something there.
00:03:23Okay, anyways, I'm totally off rails now.
00:03:32BUN!
00:03:36Great performance, weird evolution.
00:03:40Definitely becoming a tool tailored for anthropic purposes.
00:03:49Would love to see an AI model creation course coming from you and more generative AI code.
00:03:56So the interesting thing about AI is, so for one, an AI model creation course, I have absolutely
00:04:04no expertise in that area.
00:04:06So there won't be any course on that.
00:04:08A course about using AI definitely planned, just difficult because there's so much stuff
00:04:14changing related to that so that I don't have that.
00:04:18What I do have is, of course, courses on some tools like Cloth Code.
00:04:22But yeah, I don't want to show how I use AI because I'm very likely using it in a different
00:04:28way in a month from now or so.
00:04:31So yeah, thank you all so much for the very nice feedbacks.
00:04:34Thank you Vaishnaf Dattir, but yeah, probably no such course in the very near future.
00:04:42And also the difficult thing about AI, of course, is that some people like it.
00:04:47A lot of people hate it.
00:04:48And I personally have like a mixed relation.
00:04:52I wouldn't be sad if we wouldn't have AI, but we do have it.
00:04:56And so I try to make the best out of it and I use it.
00:05:00But it's hard at the moment to create courses and content that appeals to everybody.
00:05:07But I try my best.
00:05:08I basically try to cover what I'm working with and what I learn, what I enjoy working with
00:05:14to some degree, and that's what I try to share.
00:05:18I've been reading a Bundocks for last week and it's so sexy, I really want to use its
00:05:23own routing.
00:05:24But what is missing?
00:05:25The middleware structure of something like Hono.
00:05:26Yeah.
00:05:27So as I said before, I'm also a big fan of Bun.
00:05:32By the way, speaking of fan, it's probably a good sign that the fans of my MacBook are
00:05:39getting silent again.
00:05:41So hopefully that entire encoding issue is solved.
00:05:45I don't know what the problem was, but I'm not going to touch anything here.
00:05:49Anyway, so yeah, I also like Bun.
00:05:53And I think also, by the way, that the documentation is really, really well done.
00:05:57No matter if you're a human or an agent, agents really have an easy time getting information
00:06:03out of it.
00:06:04You just throw the links at them, copy the page content, view it as markdown and copy
00:06:11that link.
00:06:12So that's really all well done.
00:06:14And since you mentioned Hono, I also will say Hono is a really good framework and I typically
00:06:23these days when I'm building a web application, let's say a REST API or something like this,
00:06:28my default stack is typically Bun and Hono.
00:06:32And I like that a lot.
00:06:33I think Hono is a really elegant, lean framework that's definitely worth a closer look.
00:06:42Can be deployed on a broad variety of hosts and of course also on any VPS.
00:06:48And they have stuff like obviously middleware, they got a router and stuff.
00:06:52But what's also nice since you mentioned it, Bun also has a lot of the built-in, not the
00:06:58middleware part.
00:06:59You would have to build it on your own, but it has routing built-in.
00:07:05Actually also has a file system router if you want that.
00:07:08So without installing anything else, you can use your file system for defining the routes,
00:07:13but it also has its own routing API so that you can spin up a Bun server like this, register
00:07:21your routes like that and also be a bit more detailed there and for example, define different
00:07:28route methods for different paths.
00:07:32Where is it?
00:07:33Post.
00:07:34Is it not on this page?
00:07:36You can define different route methods.
00:07:43Why is it not on this page?
00:07:50Where is it?
00:07:51Is it on the server?
00:07:52Yeah.
00:07:53So you can have a path and then register different functions for different methods.
00:07:58And that's all very elegant and nice, I will say.
00:08:04So yeah, Bun and Hono, really great.
00:08:07It will just be interesting to see how Bun will develop in the future and if it will basically
00:08:14continue evolving as a runtime that can be used for web servers, if they will prioritize
00:08:20performance and new features related to that, which it seems to be the case.
00:08:26I want to be very clear.
00:08:28When you take a look at the BunX account, they share a lot of updates that clearly help the
00:08:39runtime part and so on.
00:08:41So yeah, I don't think Bun will move away from being a great runtime, being a great solution
00:08:50for building web servers with or without Hono, but it will be interesting to see if it stays
00:08:55like this or if it becomes that anthropic tool, which primarily focuses on being a tool that's
00:09:01great for building AI agents, which also would be interesting, but also kind of sad because
00:09:06it is a great runtime, very fast, very feature rich.
00:09:10And I think it will stay like that.
00:09:13Since vector databases are key for AI apps, any good resources to learn them?
00:09:17And would you consider a course covering this?
00:09:19Yeah, I did consider a course on vector databases.
00:09:23And I think quadrant, for example, is a very interesting one, which you can also self host.
00:09:29And I did indeed create a stream, oops, that's, I don't know my own tag.
00:09:40I did do a stream last year on rag and vector databases.
00:09:47It's on a different channel, MaxMunSchwarz were extended.
00:09:49I used to upload my streams to that channel.
00:09:52I don't do that anymore.
00:09:54I'll just keep them online here on YouTube.
00:09:56Anyways, I did a stream here on building a rag.
00:10:00Oops, I'm in front of that.
00:10:02I did a stream here on building a rag system with quadrant.
00:10:07So that may be interesting to you.
00:10:09It's a year old, but it's still all relevant because yeah, I found it interesting.
00:10:13Nowadays I will say, I'm not sure if vector databases will stay as important for AI applications
00:10:22as we thought they were, because nowadays it looks more like the future is agentic search.
00:10:28So you have an AI agent running on a VPS, running on Cloudflare, running on your Mac
00:10:34book or your PC, doesn't matter.
00:10:36And you give it a file system and in that file system, you've got a bunch of documents.
00:10:41Obviously depends on the use case.
00:10:42If we're talking about hundreds of thousands of documents, that may still be doable, but
00:10:49it'll be a bit more tricky.
00:10:51But if we're talking about a use case where there maybe are, let's say a hundred documents
00:10:56or something like this, of which the agent should be aware or which should be available
00:11:01to the agent.
00:11:03Nowadays it seems as if it's more efficient to just give the agent the file system and
00:11:08let it do its thing than connecting it to a vector database.
00:11:13Of course, as always, it's a bit more nuanced.
00:11:15It will always depend on what exactly you're building and semantic search and vector databases.
00:11:20They have their use cases, they are very useful for stuff.
00:11:24But agentic search and just giving your agent a file system might be interesting too.
00:11:29Would I consider creating a course on this?
00:11:31Yes, absolutely.
00:11:32Both on creating AI agents, which have access to the file system and can do all kinds of
00:11:36stuff as well as the vector database part.
00:11:40And indeed, I already do also cover that in one of my courses.
00:11:45In my general CHET GPT generative AI course, I released a big update for that course last
00:11:51year and I added a brand new section on RAG and also I cover the basics there.
00:12:00A standalone course would be interesting and I would love to do that.
00:12:04There are just so many ideas that I can't promise it, but it is something I would be interested
00:12:07in, yeah.
00:12:12No worries, I've learned a lot from your Flutter React Node.js courses, I started my programming
00:12:16journey with you five years ago, thank you, wish you a lot of success, keep growing nonstop.
00:12:20Thank you so much.
00:12:21And by the way, thank you all for staying with me here during these technical problems
00:12:25which hopefully are solved now.
00:12:27And yeah, let's focus on teaching.
00:12:29And thank you all for the very kind words.
00:12:31I read them all and it means a lot to me.
00:12:34So thank you so, so much for the awesome feedback.
00:12:39There are so many coding agents, but it's all about managing context which is sent to
00:12:42the LLM model.
00:12:43So we could mainly use coding agents from big companies.
00:12:47Yeah, I mean, right now we're in the Wild West phase, right?
00:12:52There's so much evolution going on.
00:12:57Every day we got a new agent or a new agent framework or a new service that allows us to
00:13:04run our agents.
00:13:06There's so much going on.
00:13:07We'll have to wait until the dust settles a bit, maybe in a year or so to have a better
00:13:12picture of what's worth using or how we're going to build certain things.
00:13:20Yeah.
00:13:23But speaking of building stuff, the original idea was to explore BUN and I think we did
00:13:29that.
00:13:30So at least like what I wanted to share about its development.
00:13:34And I think if you're building anything JavaScript web dev related these days, BUN is a great
00:13:42choice.
00:13:43And yeah, that's important.
00:13:44It's also a good choice even if you don't want to use the runtime.
00:13:48If you don't like that, you can't use it, you don't want to use it.
00:13:51But even if you don't do that, using its package manager may be interesting because BUN actually
00:13:59is a combination of things.
00:14:02It's a runtime, so that is what you would use for building your web server or for building
00:14:07your CLI.
00:14:09But it's also a package manager.
00:14:12So you can use it instead of NPM to install, guess what, packages.
00:14:17And what's really great and very useful here is that it has this bunfig file.
00:14:32It's a configuration file for BUN.
00:14:34And it also plays an important role for the package manager because in there you can configure
00:14:38a lot of stuff obviously.
00:14:40And one thing you can install, couple of things you can configure for the package manager.
00:14:46One thing you can do is this minimum release age thing here.
00:14:51So that is something you can add to this bunfig.toml file.
00:14:55And this will ensure that whenever you run bun install or bun update and so on, it will
00:15:02only pull and install package versions that are at least whatever you specified here, old.
00:15:10So that is in seconds.
00:15:12So that value here in the example would be three days.
00:15:16And why is this useful?
00:15:18Because we're living in at that time where we got supply chain attacks almost every day.
00:15:24At least it feels like that to me.
00:15:27And at least thus far, most of these supply chain attacks have been detected and mitigated
00:15:34relatively quickly within a few hours typically.
00:15:39But of course, if you install the package in that timeframe where a package was compromised,
00:15:46your machine would be compromised.
00:15:48That's obviously bad.
00:15:49Now, if you use this setting, you vastly reduce the risk of being affected because you only
00:15:57download package versions that have been online for a couple of days.
00:16:00And again, typically the supply chain attacks have been mitigated by then.
00:16:04Now bun is not the only package manager that does that.
00:16:07Pnpm, for example, also has that.
00:16:11So you can of course also use that.
00:16:13I just would recommend using something like this.
00:16:16And NPM also, I forgot it already, but they also have a relatively new feature which you
00:16:23can add where you can also configure something like this.
00:16:35I don't know exactly what it was called, min release age.
00:16:39I think it's min release age.
00:16:42It's also something you can do when using NPM.
00:16:48But you should use something like this.
00:16:49Now using bun for package management does not just have the advantage that you could use
00:16:53the setting, but also that it's really, really fast.
00:16:56So I like it for that too.
00:16:58And then you can also use bun for bundling.
00:17:01So instead of Vite, for example, or Vite is technically not just a bundler, but it has
00:17:07a bundler.
00:17:08It comes with a bundler in the projects it typically scaffolds for you.
00:17:13So you can use it instead of Vite for watching stuff, for running dev servers, for building
00:17:18your projects.
00:17:20And it also comes with a test runner, so you can use bun instead of Jest or Vite.
00:17:26There still are arguments for using Vite or Vite and so on over bun.
00:17:31They come with a bit more features, features you may be interested in.
00:17:36Of course, bundling is just one of the many things it does.
00:17:40For Vite, it's a core thing it does or it cares about.
00:17:46And the team behind Vite also built their own bundler, Rolldown, which is used by Vite.
00:17:51So I'm not sure, I typically don't use the bundler.
00:17:55I am using the test runner actually for my tests, but mostly I'm using the runtime
00:18:06and package manager.
00:18:07But yeah, bun, I really like it.
00:18:11The direction it's taking is the only thing that I think won't be a problem, but could
00:18:16be one, yeah.
00:18:19And greetings back to Berkey and thank you to Selmetri for the shout out.
00:18:26Big fan of your tutorials.
00:18:27Thank you so, so much.
00:18:28How's it going?
00:18:29It's going good now.
00:18:30Had some technical problems about which I'm very sorry, don't know why, but now everything
00:18:35seems to be stable again, thankfully.
00:18:38Am I going to do a video on Remix 3?
00:18:41I noticed they are now in beta.
00:18:42Yeah, I noticed it too.
00:18:44I hadn't had a look at it yet, didn't really have the time for it yet.
00:18:51But I do plan to dive into Remix 3.
00:18:57I think they're saying that it's very much not finished or am I wrong?
00:19:06This is still a pre-release, it is not production ready yet and there is still a lot to do.
00:19:11Yeah, I think I read that and then I decided to kind of postpone this because I got so much
00:19:16to do right now that I didn't want to dive into it if it's very rough and totally unfinished
00:19:22right now.
00:19:26But I will say I watched their talk last year, I think, at where was it?
00:19:32Some remix con.
00:19:34It was at their own conference, I think.
00:19:37And it'll be interesting to see.
00:19:40I think from what I understand, they're trying to build a framework.
00:19:45Yeah, so what is Remix?
00:19:47Remix used to be a React framework like Next.js.
00:19:50That is the vision with which they started in 2020, 2021, not sure.
00:19:57And I really liked Remix.
00:19:59Matter of fact, Akatamite.com, my website, was built with Remix in the past.
00:20:04Now it's not anymore, now we're using Tanstack, but it was built with Remix.
00:20:10But then they decided to switch away from React because they weren't happy with the vision
00:20:15of React, and they basically decided to build a framework, Remix 3, which is optimized for
00:20:22AI agents.
00:20:23So which is built in a way that it's super easy to understand and pick up by agents, even
00:20:29though obviously the remix code will not have been part of the training data of the agents
00:20:34because the training data is all that old data for the most part from the before 2000, 2022
00:20:40and so on.
00:20:41And obviously Remix 3 didn't exist back then.
00:20:43It still doesn't really exist, or the beta does now.
00:20:47But they wanted to build something which is easy to pick up and understand by agents.
00:20:50And yeah, I'm super interested in really giving that a try once it's a bit more done.
00:20:59And I'm interested in seeing how well that works.
00:21:02I mean, overall, I think it will be very interesting to see how this entire web development landscape
00:21:09will evolve, because we have all these frameworks, we used to have all these framework wars in
00:21:17the past, like in 2019, a new framework was released almost every day.
00:21:23Oh, November 22, 2021, it was released, thank you so much.
00:21:30So yeah, we used to have all these JavaScript frameworks in the past, and nowadays that has
00:21:34all calmed down, right?
00:21:36Otherwise it's all about AI and agents, obviously.
00:21:39And it's the same for me.
00:21:40I mean, for me, it's all about AI, I spent my days playing around with different AI models
00:21:45and agents.
00:21:47Because clearly that is where we're moving towards to.
00:21:51Doesn't mean that it doesn't matter to know the fundamentals, it matters very much.
00:21:55But still it matters to, for me at least, it matters to understand how I use these AI tools,
00:22:01how I can get something useful out of them, because I'm a developer, I want to stay a developer
00:22:06and that means I have to develop with AI.
00:22:10But what I wanted to say is that releasing a new framework like a remix free right now
00:22:16feels very anachronistic, which doesn't mean that it's wrong, it just means it'll be hard
00:22:25to get a significant amount of traction, because all these AI models and tools by default suggest
00:22:32React applications.
00:22:34They suggest React and Next.js and so on applications typically.
00:22:38So it'll require a developer to explicitly tell the AI to use remix free, just as it requires
00:22:47a developer to tell the AI to use Angular or Svelte right now, because again, the default
00:22:54is very often React.
00:22:57So in order for remix free to get enough traction that enough people start using it, that'll
00:23:04be interesting to see.
00:23:05I definitely think it is possible, because of course in the past before AI you also had
00:23:12to convince developers, but it was probably a bit easier, because they weren't running
00:23:17their stuff on autopilot.
00:23:19But it can definitely be done.
00:23:24It'll just be interesting to see if they succeed in that.
00:23:27And for that, I think the framework will have to offer some kind of significant advantage,
00:23:34because in the past you could say, hey, if a framework offers a great developer experience
00:23:40and then delivers a fast performant application, that's a huge plus.
00:23:44And that's kind of what was a problem, by the way, for Angular when they made the transition
00:23:50from AngularJS, Angular 1, to Angular 2.
00:23:53They broke so much, and it looked so complicated, that they lost a lot of developers there,
00:24:00I think.
00:24:01That opened up a big window for React, which is why React became so popular.
00:24:06Maybe it would have become that popular anyways, but the transition from Angular 1 to Angular
00:24:112 definitely didn't help.
00:24:13So in the past you had to convince developers with developer experience and so on, and the
00:24:18syntax - for me, for example, I always wanted to work with a syntax I find beautiful or I
00:24:25like working with, which is why initially I didn't like JSX and React.
00:24:35Anyways, nowadays that is less important, because nowadays we still should look at the code.
00:24:43Not all developers do it, and I don't know how things will change, of course.
00:24:48Code reviews may become less important in the future, and the important part is that you
00:24:54design the overall architecture of an application and make certain technological choices, but
00:25:00therefore the syntax is less important right now and very likely also in the future.
00:25:06So it will be interesting to see how a framework can win over developers at a time where the
00:25:14old ways of winning developers are less important.
00:25:19Lots of talking about remakes, but these are my thoughts.
00:25:21But I'm super excited to try it out once it's a bit more stable.
00:25:26I had the same problem finding the npm config key here.
00:25:31The npm documentation is not great, I will say.
00:25:36Thank you for teaching me Angular 9 years ago and still inspiring me today in the AI era.
00:25:41Your content truly shaped my developer journey.
00:25:43Wow, thank you so so much, MS Athianarayanan means a lot to me.
00:25:52Thank you very very much.
00:25:53Super happy to hear you're on board for all these years.
00:25:55Nine years ago is really really long.
00:25:57So much has changed there.
00:25:59I mean, nine years ago is almost when I started.
00:26:01I started 11 years ago, pretty much, more or less.
00:26:05So wow, yeah, thank you so so much.
00:26:09Great to still have you on board.
00:26:10And I'll try my best to keep on sharing hopefully useful knowledge, even though a lot has changed.
00:26:16I mean, so much has changed.
00:26:19The way I produce videos, the way you have to get attention on YouTube and all.
00:26:26Don't get me started.
00:26:27So much has changed.
00:26:28I'm getting old.
00:26:29If AI agents need to execute tools and code, do you think BUN could evolve into an AI agent
00:26:34runtime or would it need a completely different architecture?
00:26:38That is a great question.
00:26:40And I think BUN could absolutely become more of an AI agent runtime.
00:26:46And I'm not sure if at the same time it can stay a general runtime and they just add those
00:26:53agentic features on top of it or if they kind of split it or whatever.
00:26:57But I mean, Anthropic acquired BUN, it would make a lot of sense to build something that
00:27:04is a really strong agent runtime.
00:27:08Would make a lot of sense for Anthropic.
00:27:11Wouldn't win a lot of hearts in the web developer community, probably.
00:27:14But hey, I'm not sure if Anthropic is winning a lot of hearts there anyway.
00:27:18So yeah, I could definitely see a future where BUN is getting more and more features that
00:27:24make it a great agent runtime.
00:27:27Stuff like sandboxing, proxying network requests, tool management, permission management, stuff
00:27:33like that.
00:27:34I could definitely see more stuff like that landing in BUN.
00:27:38And to be very honest, I'm a bit surprised that we don't have more of that stuff already.
00:27:51The direction they're going on is really interesting, but would love to see your video and thank
00:27:54you for answering the question.
00:27:56Yeah, sure.
00:27:57Sure, sure.
00:27:58As a frontend developer, should I learn Docker and Kubernetes?
00:28:02A couple of thoughts there.
00:28:05For one, I think you should in general consider not being just a frontend developer because
00:28:14even though that definitely was a thing in the past, my feeling is that with AI the requirement
00:28:21of companies will be to have generalist developers.
00:28:27At least right now, I don't know how things will change in five years or so.
00:28:33It could be that we want more specialists then again.
00:28:36But right now I think generalist is the way to go.
00:28:39Generalist with dedicated areas where you then are a specialist.
00:28:43You don't have to be a specialist in everything.
00:28:45You can't be.
00:28:46But I think you want to be a generalist.
00:28:49So that's one thing.
00:28:50And therefore, to answer the second part, yeah, I think knowing the basics about Docker is
00:28:55definitely something that's useful for most developers.
00:28:59You don't have to be an expert.
00:29:00You don't need to be able to build super advanced Docker images or know the ins and
00:29:06outs of managing Docker swarm clusters or complex multi-container Compose setups.
00:29:14But especially now with AI, it's easier than ever to work on Docker Compose configuration
00:29:20files or Docker files and understand those Docker commands.
00:29:24And if you then in addition learn the behind the scenes or understand how Docker works and
00:29:31what its core concepts are, I think that makes a lot of sense.
00:29:35For Kubernetes, I wouldn't say that's important.
00:29:38Learn it when you need it, but it's not something you need to know upfront.
00:29:42But Docker, yeah, the basics you should understand and learn, I would say.
00:29:49The same thing happened when Next.js switched from pages to the app router.
00:29:52So with Angular and Angular 2, yeah, absolutely, that is a good example.
00:30:01It's always like big foundational changes like this.
00:30:07They can break a lot of stuff.
00:30:09And I mean, with the app router, the idea was actually they tried to give the community what
00:30:16they wanted, but they still did it in such a way.
00:30:21I'm not even sure why exactly, to be honest, but kind of split the community.
00:30:26And then I think there were a couple of things.
00:30:28Like one problem, of course, was with the app router and Next.js that initially it had that
00:30:33super aggressive caching, which led to situations where the behavior on your dev server was not
00:30:42the same as it was after you built the app, which is, of course, horrible, because you
00:30:46tested everything, everything worked, you built it, you deployed it maybe, and yes, you should
00:30:51test that before you deploy, but still, maybe you deploy it right away.
00:30:55And suddenly the app didn't work the way you saw it.
00:30:57So that was horrible.
00:30:59And then the dev server was also super slow.
00:31:01And I think all these things led to a lot of problems and churn and why people then switched
00:31:06to remix like I did.
00:31:08I switched to remix because of some problems with that dev server and the development of
00:31:13working with the app router.
00:31:15And other people switched to other frameworks.
00:31:17Nowadays, I will say, in my experience, the Next.js app router works really well.
00:31:23And yeah, most of these problems are gone, but it wasn't a smooth transition.
00:31:31Well, we'll get back to the other questions in a second.
00:31:34But do I have a Docker course?
00:31:36Yes, I do.
00:31:39On Akatamite.com.
00:31:41It's right here.
00:31:42Got a Docker course.
00:31:44And it's a big course.
00:31:45I don't have to go through it entirely.
00:31:47Because I started with the basics, of course.
00:31:49And if you just complete like the first few sections, you should have a good overview of
00:31:53how Docker works.
00:31:56What do you think about going back to the basics and leave all the packages dependencies behind?
00:32:01Yeah, also a good question.
00:32:03It's something I've been playing around a lot over the last days.
00:32:07I'm exploring how can I build applications that use almost no packages.
00:32:15And I try to reduce the amount of packages I'm using.
00:32:18Now with AI, it's easier than ever to build lots of stuff on your own.
00:32:22And that has some problems, because of course, using a third party library means that you
00:32:30don't have to maintain that code.
00:32:33It means that someone else is responsible.
00:32:36Not really if it's open source, but you get what I mean.
00:32:40Someone else is maintaining it for you.
00:32:42And AI code also is far from perfect.
00:32:46So you can't generate everything with AI, even though some people may want to believe you
00:32:52that it's that easy.
00:32:54But there are trade offs.
00:32:55But it's definitely easier than it was before to replace certain packages.
00:33:00And that has been something I've been playing around with.
00:33:04And for example, my personal website, maximum Schwartzmuller.com, which is of course a super
00:33:10simple website.
00:33:11It's really just a blog, right?
00:33:14But I used to use Astro.js, which is a great framework for static sites, I will say that.
00:33:22But then I decided, you know what, I don't need it.
00:33:25It's really just a bunch of HTML files.
00:33:28So with help of AI, I migrated that Astro project to a vanilla HTML, vanilla CSS, vanilla JavaScript
00:33:36app, and my blog articles, I just write the text now.
00:33:44And then I let AI convert it to HTML.
00:33:47But I will soon actually build my own very basic markdown parser and use that.
00:33:52So I think you can reduce the number of packages you're using.
00:33:58And now with all these supply chain attacks, that may be something you want to do.
00:34:04That's something I want to do for sure.
00:34:05I want to reduce the chances of me being hit by a supply chain attack.
00:34:10So that's part of the reason why I try to reduce the amount of packages I'm using.
00:34:19Thanks for pronouncing my name.
00:34:20I hope it wasn't totally wrong.
00:34:23Sorry.
00:34:25I also used to watch Angular 2 tutorials from you.
00:34:28See this?
00:34:29We're getting old.
00:34:30Oh yeah, we are.
00:34:31We are.
00:34:32Oh yeah.
00:34:33Those Angular 2 tutorials.
00:34:34That was such a great time, but it's over.
00:34:38But yeah.
00:34:40I mean, back then when I was getting into Angular 2, my big advantage was that I never worked
00:34:48a lot with Angular 1.
00:34:49So I didn't have that rough transition.
00:34:51And I liked Angular 2 because I had a C# background back then.
00:34:55I worked a lot with C# before that.
00:34:58So kind of didn't feel too bad to me.
00:35:03So I liked Angular 2.
00:35:04I was one of the few people to like it, I guess.
00:35:07Max, my man, I want to say thank you for partially kickstarting my career in web.dev in 2019.
00:35:12Thank you so much, Uncle Sam.
00:35:14I'm super happy that I could be part of that career and that I could help there and help
00:35:20kickstart it back then in the golden age of JavaScript framework wars, at least.
00:35:27We'll see how it is for web development.
00:35:30But yeah, thank you so, so much and thanks for being part of the stream.
00:35:34I missed the era where we used to learn new languages frameworks from your courses tutorials.
00:35:39Yeah, I will say I missed that too.
00:35:41I mean, I have that video on my channel, which some of you maybe already saw this NoJoy video
00:35:55where I talk about the fact that for me it was more fun before AI.
00:35:59But the coding part was more fun.
00:36:01Now, of course, you can build more and that also is fun.
00:36:04And I do genuinely find those AI tools and models.
00:36:08I find that all interesting.
00:36:10But I definitely miss the flow state you could enter when you were coding.
00:36:16And I also miss learning new libraries because you can still learn new stuff and I'm learning
00:36:21new stuff all the time.
00:36:23But you learn so differently nowadays, of course, and I'm still trying to figure out how to teach
00:36:28a new framework or library these days because I still do believe there is new stuff to learn.
00:36:35AI is not the only solution there.
00:36:37It's a great help, but it's not the only thing there.
00:36:41And I definitely don't want to work with AI with a technology I don't understand.
00:36:46So I want to understand the technology and I'm still trying to figure out how do I best
00:36:51learn a technology and how can I then teach that in a meaningful way?
00:36:55Because teaching you the syntax of a library like for 40 hours as I used to do it in the
00:37:00past, that doesn't seem to be the future for me.
00:37:04It doesn't feel right because I know that you won't be writing all that code.
00:37:10You have to understand it.
00:37:11But in order to understand it, you don't need the same teaching as you need for writing it
00:37:15because writing is harder and that part is going away.
00:37:19The new part is all about understanding it, knowing about patterns and best practices and
00:37:24about architecture and architecting a program.
00:37:27That is what I want to teach in the future.
00:37:29It will not all be AI stuff.
00:37:30I'm not interested in that, but I'm still trying to figure out what and how to teach
00:37:35that.
00:37:38And that's especially hard since it's all changing so quickly these days.
00:37:42But that's kind of my plan there.
00:37:43But yeah, I do miss these days.
00:37:46I enjoyed learning Vue or Vue Free and I enjoyed creating a course on that.
00:37:54That was great.
00:37:59One more thing.
00:38:00I wish to see Manuel with you on livestreams.
00:38:02Maybe I'll bring him on.
00:38:03Good idea.
00:38:04Why not?
00:38:05I think he's gone now.
00:38:06He sat in the burrow next to me half an hour ago.
00:38:12So I'll bring him on.
00:38:15Oh, another shameless plug.
00:38:17Sorry.
00:38:18Manuel actually launched a new course on Claude Cowork.
00:38:22So that's not about coding, of course.
00:38:24That is a general AI tool which you can use for stuff like data analysis and so on.
00:38:29So if that's interesting and you want to see Manuel, definitely recommend it.
00:38:32It's a great course launched today.
00:38:40I think that is a perfect summary.
00:38:54I don't have a lot to add there.
00:38:58Not using packages has advantages, but the big disadvantage is that you have to maintain
00:39:02it all.
00:39:03You might be overlooking a lot of edge cases.
00:39:06You may have gapping security issues that have been fixed for the popular libraries.
00:39:11It's a trade-off, as always.
00:39:13But for certain use cases, reducing the number of packages is definitely worth a thought.
00:39:21But I wouldn't build a complex application with just vanilla HTML, CSS and JavaScript
00:39:26most likely.
00:39:28It might be a fun experiment, though, but I don't think it would be a good idea.
00:39:37Speaking of frameworks and libraries, something else I mentioned before is Adonis, Adonis.js.
00:39:46So I mentioned it before, but the stream was all laggy and broken before, so I'll repeat
00:39:50it here.
00:39:51I used to work a lot with Laravel, the PHP framework, 10 years ago or so.
00:40:00And I then at some point, I don't know when, maybe six years ago, I don't know, I basically
00:40:08searched if there is an equivalent, a framework like Laravel for JavaScript.
00:40:14And I found Adonis.
00:40:16And I'm not sure if they're still doing it, but I think in the past they marketed
00:40:21themselves as the Laravel for JavaScript.
00:40:26Not sure if they're still doing it.
00:40:28And the idea is that it really comes with all the stuff or most of the stuff that you typically
00:40:34need for building a web application built in.
00:40:37So something like, that's the wrong page, something like authentication, where are the API docs
00:40:45here?
00:40:46Okay.
00:40:47Something like authentication, they would have it built in.
00:40:52So everything from managing sessions, cookies, stuff like that.
00:40:57They would ship a SQL ORM, just like Laravel has Eloquent.
00:41:03They would ship their own front-end solutions.
00:41:06They have their own templating engine, EdgeJS.
00:41:09So that's kind of the philosophy.
00:41:11And I'm not saying that because I'm Adonis expert.
00:41:15I'm not.
00:41:16I haven't used Adonis at all.
00:41:19I played around with it like six years ago or whatever.
00:41:23But I think what made me happy is that over the past weeks in my small bubble on X, which
00:41:31is mostly about AI, unfortunately nowadays, I saw that Adonis seemed to be getting a bit
00:41:37more traction.
00:41:40And I think if we look at the download numbers, internal server error.
00:41:46Okay, no problem.
00:41:59And Tanstech have, I thought Tanstech also had a trends site.
00:42:17Does it not?
00:42:24Here we go.
00:42:28So obviously we want to remove some stuff here.
00:42:35Yeah, that you, I'm not sure if this is just the right package.
00:42:50Probably not.
00:42:52Yeah, that looks better.
00:42:58Adonis.js Core.
00:43:01This is what I was looking for.
00:43:02Okay.
00:43:03So what you can see is that they had a little bit of a upward trend.
00:43:08Obviously not super strong.
00:43:10Obviously this is still a small framework.
00:43:14But I found it quite interesting that they had an upward trend because I like the idea
00:43:21and I always wondered why exactly it didn't take off.
00:43:27And the reason why I didn't dive deeper six years ago or whenever is really just an unlucky
00:43:35coincidence that I played around with it.
00:43:38And then I only remember I basically ran out of time to dive in deeper.
00:43:43I had some course updates or some other big course which I wanted to do.
00:43:48And then one thing came to the other, you do that, you have another project and you forget
00:43:53about it.
00:43:55Sometimes I always thought this looks perfect.
00:43:58And the interesting thing is that in 2020 or so, all of a sudden all these full stack frameworks
00:44:08became a thing.
00:44:09All of a sudden we were not talking about single page applications connected to REST APIs anymore.
00:44:19Obviously that's still a big thing.
00:44:21But we all of a sudden started talking about building full stack applications.
00:44:26With Next.js, with Remix, Angular started getting its own full stack framework, Analog.js.
00:44:38For Vue we have Nuxt, and Nuxt existed in 2018 or 17 already, but that all started to get
00:44:46more traction.
00:44:48So did Next.js by the way.
00:44:50That all started to get a bit more traction, I would say, in 2020 or so.
00:44:57And I always wondered why did we always want to have our React frameworks become full stack
00:45:08frameworks?
00:45:09Why was it all centered around React?
00:45:12Why did we not use something like Adonis, which was a full stack framework like Laravel right
00:45:18from the start?
00:45:19And I think the reason is that back then in 2018, 2019 and before, React and Angular were
00:45:28all the rage, also for me, where it was all the rage where you built single page applications.
00:45:41It felt right.
00:45:42It was so nice to build applications like this with React.
00:45:46If you were coming out of the earlier days of web development, like I do, you still remembered
00:45:52how difficult it was to build reactive interactive user interfaces back then.
00:45:57So we all liked solutions like React, like Angular.
00:46:01And I think that's the reason why this kind of kept being the thing.
00:46:06And we then felt like it was a good thing to have full stack solutions for React.
00:46:10And I still think that is good.
00:46:12I like 10 stack start, next JS and so on, but I think that's the reason maybe why Adonis
00:46:19never became a big thing.
00:46:24And nowadays, unfortunately, I don't have that much more time because I would love to play
00:46:28around with it because I think it would be a great solution for many web applications
00:46:34you might want to build because it comes with so much stuff built in.
00:46:37Yeah, but I think it's just unfortunate for Adonis that React and those frameworks became
00:46:49so popular, probably due to the history of the web, that it didn't have a chance.
00:46:55And for Laravel and PHP, that was probably different because with PHP, that always was
00:47:00on the backend.
00:47:01So it didn't feel like, "Oh, building the frontend is so nice.
00:47:05We want to have a framework that now extends this to the backend."
00:47:08No, with PHP, you always rendered your HTML pages as PHP pages.
00:47:13So having a framework for that all felt natural.
00:47:16That is my explanation here.
00:47:20It would be cool to have an Adonis JS course.
00:47:22Yeah, I would love to do one.
00:47:24But as I said, unfortunately, I'm not sure if I'll find more time for learning it.
00:47:32Getting into it in a stream could be interesting.
00:47:37And maybe I will find the time, but I would have to learn way more about it in order to
00:47:42create a course.
00:47:43But yeah, it would be interesting.
00:47:45The Philippines Adonis is a gay bar, probably one or two places on earth where that is the
00:47:51case due to the name.
00:47:56Did you see the video from web dev simplified on content creators, including you Theo are
00:48:02shifting more and more towards AI?
00:48:04What is your opinion on that?
00:48:05Yeah, I did see it and I will say I was not a big fan of it because I think in the video,
00:48:14to me at least, it sounded a bit like that was a decision made by Theo or me because that
00:48:29is where you get more clicks.
00:48:32Now I will say you do get a lot of views and clicks there and obviously those matter.
00:48:39But I always try to create content on the stuff I work with.
00:48:46That is, it's as simple as that.
00:48:50I used to create coding tutorials when I wrote all that code with those technologies I covered.
00:49:00And now I'm talking more about AI because I'm working every day with codex, with clotcode,
00:49:07with pie.
00:49:09And I'm building stuff with it.
00:49:11Some stuff that was released like this max draw app, which I'm using or my business headshots
00:49:15app or build my graphic and more to come.
00:49:19So I do build stuff and lots of internal stuff too, of course.
00:49:24And AI is simply the thing that is changing how we build software.
00:49:32And therefore that's what I talk about.
00:49:34So it kind of goes hand in hand.
00:49:37Yes, it gives you more clicks, but it gives you more clicks, more views because a lot of
00:49:41people feel that way and want to stay updated with what's happening there.
00:49:47So I think that is why content creators are shifting.
00:49:50And to me it sounded in that video as if that were the bad content creators that just hop
00:49:56onto the next thing that gives them clicks.
00:49:59But it's kind of the opposite.
00:50:01You create content on what people are interested in and what you're using.
00:50:06And that kind of naturally connects there.
00:50:09It's not like that big agenda where you suddenly decide, oh, that is popular, then I'm going
00:50:14to cover that.
00:50:15That is popular because it's affecting everybody, including me.
00:50:19And I work with that stuff every day.
00:50:21So yeah, I found that video a bit weird, but maybe I also just misunderstood it, to be honest.
00:50:30But to give you an example, four years ago or so, the big hype was all about Web3 and
00:50:37NFTs and crypto.
00:50:41And I don't know about Theo, but I don't think he was deep into Web3 back then.
00:50:46And I definitely wasn't.
00:50:48So it's not about covering what's currently hot.
00:50:50It's really about covering what you work with.
00:50:54We should investigate cloning, Max, so we can have all the courses.
00:51:01That is a project for the future.
00:51:07Right now, I mean, that's the thing with AI.
00:51:10Everybody is telling you that there is so much more you can work, that it can do so much work
00:51:18for you, but we all end up just doing more.
00:51:21I'm not sure if that's such a great deal, especially since AI is taking some of the work I really
00:51:25enjoyed and is giving me some work I don't enjoy that much, like reviewing code.
00:51:30But yeah.
00:51:32What do you do for a living now, might I ask?
00:51:33Anything besides educational courses?
00:51:35Well, it is basically the same I did for the last 10 years.
00:51:39It's a lot about courses, content creation.
00:51:43It used to be more client projects and stuff like that.
00:51:47And I have scaled that down as courses have become more and more successful.
00:51:52And then I have my side projects like the sites I just shared, earning some money with that.
00:51:58It's really the combination of all these things.
00:52:01But educational content is definitely the most important part and has been for the last couple
00:52:07of years.
00:52:10But yeah, I try to keep that mixture because I don't just want to create courses.
00:52:14I want to work with the stuff actually.
00:52:16And I want to build useful stuff with it.
00:52:19I think people just tend to pick the most used technology by default instead of the better
00:52:23one causing a herd effect.
00:52:25Sometimes people just lazily don't want to analyze better alternatives.
00:52:28Yeah.
00:52:29And I mean, a lot of people don't get to make choices anyways, because if you're just working
00:52:34at a company or for a client that has certain preferences or rules in case of a company,
00:52:42then you don't make the choices.
00:52:46You can't say, "Hey, we're going to use Adonis or whatever."
00:52:49No, you're going to use Next.js or whatever it is because that is what the company used
00:52:54for the last five years.
00:52:56So yeah, there's that.
00:52:59I mean, I'm actually privileged because I can choose the technology I want to work with and
00:53:03I want to build my projects with, but still obviously learning takes time.
00:53:10Mastering something takes even more time.
00:53:13And right now where there is so much stuff to learn and handle with all the AI stuff and
00:53:19so on, it's really hard to find the time for me.
00:53:23Not sure about how it's about you all, but for me it's really hard to find the time.
00:53:28Is it really difficult?
00:53:32It is really difficult to get a job back nowadays after layoff.
00:53:37Any suggestions?
00:53:38Yeah, I know that it's very difficult.
00:53:42I read that a lot.
00:53:44So for one, I think it will get better, but that doesn't help you right now obviously.
00:53:51And right now it's really rough.
00:53:54Everybody wants to hire senior developers and not a lot of juniors and eventually companies
00:53:58will find out that they need the juniors too.
00:54:07For now it's difficult and when they find out that they need them, they won't be there.
00:54:11So difficult times.
00:54:14What can you do right now?
00:54:16What I've said in the past and what I still think is a good idea is since there are more
00:54:21applicants than jobs, you need a way of standing out.
00:54:26Now how can you stand out of the crowd?
00:54:28A good insurer, a good portfolio, a good CV.
00:54:31That all matters and you would be surprised how many people send applications without putting
00:54:37any effort into it.
00:54:39So if you put some effort into it, if it's a bit fine-tuned for the company you're applying
00:54:44to, which nowadays is easier than ever with AI, but please also read over it, fine-tune
00:54:49it, give it that personal touch.
00:54:51That alone probably sets you apart, but besides that, I think if you find any way of generating
00:54:57some audience, some reach, if you find a way of being that a YouTube channel, be that some
00:55:07Instagram or LinkedIn presence, and with that I don't mean that you need to become a big
00:55:11influencer or a content creator or anything like that, but if there is something, if a
00:55:18company searches for your name and sees that you have some following on YouTube or on X
00:55:26or whatever, no matter how big or small it is, that also can set you apart if the content
00:55:32you're producing there is useful.
00:55:33So I think that can be a viable strategy and obviously I'm aware of the fact that you want
00:55:38to find a job and not become an influencer, but that is one thing you could try to stand
00:55:42out.
00:55:44Because yeah, right now it is unfortunately tough.
00:55:48In the past, since I read it here, you need to contribute and get projects, I think one
00:55:53problem is that before AI and before the rise of AI agents, which we had over the last six
00:56:00or seven months, before that it was definitely a viable strategy to contribute to open source,
00:56:11for example, to create pull requests and so on.
00:56:16And nowadays this has kind of flipped.
00:56:18Because a lot of popular open source projects and even the smaller ones are getting flooded
00:56:24with AI generated issues and AI generated pull requests, making it very, very hard for
00:56:30maintainers to maintain these projects.
00:56:34So nowadays I think contributing on GitHub is difficult because there is a decent chunk
00:56:42of maintainers that don't want you and standing out from the crowd is even harder because everybody
00:56:52is telling their agents to just push a lot of AI slop towards GitHub.
00:56:58And for a recruiter, it will be difficult to tell apart if you have the high quality contributions
00:57:03or if you don't have them.
00:57:05So I think that still matters.
00:57:08It's still a good signal if you're doing something there, but it's way less important than it
00:57:14used to be.
00:57:16Other ways of standing out, ways that show that you're a human and that you care about
00:57:21the job and so on, I think are more promising.
00:57:25That's just my two cents.
00:57:27That all of course is coming from someone who's in the lucky position that I'm not looking
00:57:33for a job, but that I have my own business, that I have my own projects and my courses
00:57:39and the YouTube channel and stuff like that.
00:57:42And that I have a client base, which I could activate to work with and so on.
00:57:49So I'm definitely in a privileged position there, but these are the things that come to
00:57:54my mind when you think about applying or looking for jobs.
00:58:00But coming back to Adonis, I think it might be interesting to dive into that in some future
00:58:13stream.
00:58:14Not today, because I'll have to leave in roughly half an hour, but maybe for next week, or maybe
00:58:25I can come up with some YouTube video format that works, but that's really, really hard
00:58:32because YouTube works so, so differently than it used to work.
00:58:38Just creating normal tutorials doesn't work at all anymore.
00:58:42And of course, views matter.
00:58:43If barely anybody watches a video, it's not useful, obviously.
00:58:52It's not earning you any money, not giving you any visibility, and it even hurts your
00:58:56channel because if you release a bunch of videos that get no views, your future videos will
00:59:01also get less views.
00:59:02So you're really not incentivized to create videos that suddenly perform worse than your
00:59:11other videos and you know it in advance.
00:59:15But maybe I can figure out something there.
00:59:18What are your suggestions on DSA, so that is data structures and algorithms?
00:59:24Obviously that was super important 40 years ago, nowadays I'm not so sure.
00:59:29I think there is a decent amount of companies that still use it in their recruitment process.
00:59:35I've never been a fan of it and I think nowadays it's even less useful because if I were hiring,
00:59:42I wanted to see if you understand fundamentals of the technologies you will use at your job.
00:59:50I wanted to see if you can do research with AI, I wanted to see if you can use AI assistance
00:59:55and not if you can come up with some algorithm because sure, that also proves some fundamentals,
01:00:01but not the fundamentals I'm looking for necessarily.
01:00:06I think that will become less important, but I'm not sure if we're fully there yet.
01:00:12And I will also admit, I have never been a DSA expert because I never fully – yeah,
01:00:27I always hated the idea of a company basing its recruitment decision to a huge extent on
01:00:36the question if somebody can write algorithm X, invert a binary tree, whatever.
01:00:45And you can have different opinions on that for sure, but I wouldn't hate it if that
01:00:51went the way, I'll put it like this.
01:00:54Not because it affects me, it's just something I don't see a lot of value in and I think
01:00:58now in the age of AI it's even less important and valuable, so I would imagine that more
01:01:04and more companies will not use that anymore, but I may be wrong of course.
01:01:08I mean a different problem is that a lot of companies have faced the issue that applicants
01:01:14simply cheated if it were video calls, so that also is a factor why they may be using less
01:01:22of that.
01:01:24Maybe if we show we can use the technology to its max by extraordinary projects, right
01:01:29now I'm developing educational solutions, I'm a kid and teen's educator, yep, absolutely,
01:01:34something like that sounds like a great idea.
01:01:37Are you able to predict that where this AI trend is heading, do you think we'll be hitting
01:01:42a stagnating point?
01:01:43I have no idea.
01:01:47What I do see or what it feels like to me is that these tools are still getting better,
01:01:53so now we have models that have been fine-tuned so that they are really good at following instructions
01:02:01and using tools and we got the tools around them, the harnesses as we call it nowadays,
01:02:08that leverage these models really efficiently and we may be able to squeeze way more out
01:02:15of these tools and models so that they become even better and follow instructions even better,
01:02:22so it's hard to predict for me and there are so many factors at work.
01:02:27I mean there is all the compute capacity problems, so that alone could stop AI development or
01:02:35not stop it but slow it, but of course a lot of new capacity is coming online.
01:02:41Then we have stuff like this weird sub-Q thing here, the sub-quadratic sparse attention model
01:02:49about which I also created a video which is all about a model that is way faster, way cheaper
01:02:55and has way more context than the existing models.
01:02:58If that is a thing, it may be a game changer.
01:03:02If it's just hype, which is probably more likely than not, but I don't know, then nothing will
01:03:09change.
01:03:10So there are many factors that could slow down development and I have no idea if there is
01:03:17some endpoint where we see that models just aren't getting smarter and for every new capability
01:03:25they get they're losing another one.
01:03:28That could happen but I don't know about that.
01:03:30But right now I definitely see things improving, especially the tools, the harnesses that we
01:03:37use and the stuff that's being built around them like the sandboxes which we can use now
01:03:43and the different solutions we find for running these agents securely.
01:03:49A lot of development is happening there still, so I still feel like it's very dynamic and
01:03:54a lot is changing.
01:03:55And especially in the last six months, the pace has been so fast that I don't dare predicting
01:04:02what will be the case or what will be the current state in six months or so.
01:04:10What is the difference between GitLab and GitHub with their CI/CD?
01:04:14I can't judge the CI/CD because I never used GitLab.
01:04:19In general, of course, GitLab is an alternative to GitHub, but I've only worked with GitHub
01:04:26actions.
01:04:27I've never used GitLab, so I can't judge what the difference is or how good or bad GitLab
01:04:32is when it comes to that.
01:04:34What project would you suggest to build for a portfolio?
01:04:37Maybe something to stand out of a crowd.
01:04:40In the past, I always recommended to build a clone of Amazon.
01:04:44Of course, a simple clone, but something like this.
01:04:48But nowadays, I'm not so sure anymore.
01:04:51With AI being a thing, of course, you want to convince the recruiters, right?
01:04:58So being something where you don't just use AI for building, but where you also use AI
01:05:06in the product you're building may be worth simply because it looks more interesting to
01:05:12recruiters.
01:05:15For example, building your own agent harness, your own alternative to Codex or Cloth Code
01:05:22or Pi.
01:05:24That could be interesting.
01:05:25And I mean, you can get inspiration by taking a look at something like Pi, which is a great
01:05:32open source AI agent, which you can use for coding and all kinds of stuff.
01:05:37And I mean, taking a look, not copying it, but getting some inspiration from there.
01:05:42Building something like this could be an interesting project for you personally, and also to stand
01:05:47out of the crowd.
01:05:48And I know a lot of people are building agents, but it is something recruiters could maybe
01:05:53be a bit more interested in.
01:05:54So that would be something.
01:05:55Some tool that leverages AI or wraps AI in an interesting way, I think that could be
01:06:01an interesting project.
01:06:04I think the way to stand out in this AI era is the same as before, demonstrating you can
01:06:09understand systems and tools.
01:06:10People nowadays just want to use trending tools.
01:06:12They don't understand how they work.
01:06:14And I think that's a very good point.
01:06:16And also why I'm, for example, working on that systems design and architecting course.
01:06:24Because I think nowadays it is more important than ever to have a strong foundation, strong
01:06:31fundamentals and to be able to architect software and entire systems.
01:06:36And that's not the same.
01:06:38Software is not the same as systems necessarily, but what I mean is being able to build a complex
01:06:46system which involves things like planning where do I need a queue, where am I using a
01:06:54fan-out pattern or Pub/Sub and stuff like that, which kind of storage do I use for my data,
01:07:04for the data my application generates, object store, file store, which kind of database do
01:07:11I use, what am I optimizing for, for cost, for reliability.
01:07:21And these are all questions that historically have been linked very much to cloud computing
01:07:26to some degree, I'd say.
01:07:27And they still are, but I think stuff like that will become more important because as
01:07:32I said before, I think the expectation will be and the opportunity will be that you can
01:07:39and should be a bit more generalist with some areas of expertise of course then.
01:07:45But being able to design systems, but then of course also to architect software will
01:07:52be more important than ever.
01:07:55Which is why I said I think just being a front-end developer may not be enough anymore.
01:08:00You can be an expert there, but you probably should be a generalist and you should have
01:08:05a general understanding of how to build an overall web application, for example, including
01:08:11the backend.
01:08:13And then you have those, then that gets into systems design at some point.
01:08:18And I think that will be very important.
01:08:22Is GitHub still on the same place it was before?
01:08:25I think people have forgotten about its issues.
01:08:28Well, I think the GitHub issues are very present to many people, it feels like that to me.
01:08:38GitHub has had a lot of issues and I don't know if you saw it, but they shared this blog
01:08:46post with these charts, which are a bit difficult because there is no y-axis and we only see
01:08:51this number, but we don't see where we're coming from.
01:08:54But I will say there are a lot of bad things to be said about GitHub and Microsoft, for
01:09:03example, that they don't have a CEO, that they were moved into the core AI division,
01:09:09and that it doesn't really seem like Microsoft cared too much about GitHub.
01:09:14But the one thing where I have a lot of sympathy for them and which won't hold against them
01:09:21is if you look at that growth, that would have been very difficult for any company.
01:09:26And sure, they are part of Microsoft, they have cloud resources and stuff, and they're
01:09:32working on it.
01:09:33But that is a growth, that is just insane.
01:09:35And of course you could have seen it coming, but I mean, look at this, that here is fast.
01:09:42This is insane.
01:09:43This is just insane what happened over the last few months.
01:09:47So from that perspective, I understand that they're facing a lot of issues.
01:09:51And from what I understand and what they wrote in this post, they are heavily working at fixing
01:09:57the situation, getting way more capacity out there and preparing for that future where we're
01:10:04all getting flooded with AI generating code, which is already here and which will probably
01:10:10keep on accelerating as hard as it may be to imagine that.
01:10:14I mean, we have that one constraint though, we have to compute capacities.
01:10:19So maybe that will slow it down a bit, but I understand that they're facing these issues.
01:10:26Off topic, we heard AI used for war game simulation.
01:10:29Those were where AI always chose nuclear strike.
01:10:33But have you heard of any simulation research of bright and healthy society models, AI applied
01:10:39to you humans?
01:10:44So no, I haven't.
01:10:47I would imagine that AI will definitely be used in lots of foundational research for
01:10:58pharmaceuticals and so on.
01:11:01So I think there will be good stuff coming out of that.
01:11:06But it's obvious that AI has a lot of dangerous and bad potential.
01:11:12I mean, even if you just look at stuff like deepfakes, AI videos, images, and it will have
01:11:20such a big impact on society due to how it transforms the job market.
01:11:26And I don't believe that it will just kill jobs.
01:11:29I'm a strong believer that new jobs will be generated and jobs will simply evolve and change
01:11:35like my job is changing.
01:11:36I'm not creating these traditional coding tutorials anymore, which I would love to, but it's just
01:11:43evolving.
01:11:45But yeah, it does have that big, big impact.
01:11:53And especially here in Germany, where I live, it's even worse, because I don't have the feeling
01:11:59that we're using AI for anything useful here.
01:12:02It feels very much like we're all sleeping in this country on AI, like we always do when
01:12:09there is some new technology around.
01:12:12So yeah, right now, I think there's a lot of bad stuff.
01:12:18You can see and maybe not as much good stuff yet.
01:12:22Not sure.
01:12:24Sad we don't learn that at the university to industry level.
01:12:29Yeah, I mean, that's another interesting topic.
01:12:32I feel like university has always been a bit decoupled from what the industry needs.
01:12:39And of course, that is on purpose, to some degree, because university is about academia
01:12:44and so on.
01:12:45But in the past, I find it was always important to gather practical experience, be that by
01:12:56building your own side projects, by doing internships, and so on and so on, or of course, also by
01:13:03taking courses like mine, or reading docs, learning it on your own, whatever.
01:13:07But I always found it important to gain extra skills in addition to what university gives
01:13:13you to be really well prepared for the job market and so on.
01:13:18And I feel like now with AI, that's accelerating even more, or that gap is getting bigger.
01:13:24Because of course, there are universities that are keeping up.
01:13:30I'm sure there are.
01:13:31But I'm also sure there are many that don't.
01:13:34There are probably many universities that just try to forbid that students use chatgpt and
01:13:42so on.
01:13:43And obviously, they can't control it.
01:13:44But that's probably the only thing they do with AI.
01:13:49And I think that that will be a big problem.
01:13:53And that's why it's even more important nowadays to gather and get extra knowledge outside of
01:14:00university.
01:14:02Not for every university.
01:14:03And I'm not at university anymore.
01:14:05I may be totally wrong.
01:14:06That is just from what I know back from my days.
01:14:09And I can't imagine that it got so much better now.
01:14:13And with AI, everything is moving so much faster.
01:14:15I think it's super important to educate yourself in addition to what you learn there.
01:14:21But at the same time, and I've said that in other streams and videos too, I think it's
01:14:24also important to not follow that narrative of you being left behind if you don't do something.
01:14:35It's not like that.
01:14:37It's moving all so fast.
01:14:41It's basically impossible to be left behind.
01:14:45Because if something is a great solution today or important today, it might not matter or
01:14:53might not work anymore in a few months.
01:14:56So if you poured all your energy into mastering this today, it might be totally irrelevant
01:15:02in a couple of months.
01:15:04So of course, you should stay on board, stay informed, and play around with these tools,
01:15:09for example, if you're a developer, work with cloud code and so on.
01:15:13But don't feel that pressure that you have to master and understand it all right now.
01:15:19That's impossible.
01:15:20And that's also pointless, I would say.
01:15:22"Hello Max, not a question, but just wanted to say thank you for always sharing your honest
01:15:27thoughts and not buying into whatever the trend of the moment might be."
01:15:31Thank you so much.
01:15:35That means a lot to me because that's really important to me.
01:15:39I'm very well aware of the fact, for example, that on YouTube, most of my content is about
01:15:47AI right now, not everything.
01:15:50But whenever I do something that's not about AI, it tends to perform worse.
01:15:54I still do it.
01:15:56But I try to simply share the things I'm interested in that kind of matter to me, that I read about,
01:16:05that have an impact, or where I think that they may have an impact on developers, because
01:16:13I will keep on trying to share good knowledge and help other people grow as developers and
01:16:21master this transition, just as I'm trying to master it myself.
01:16:27And that's the only thing I try.
01:16:28So yeah, thank you so, so much.
01:16:29That means a lot to me because I'm not doing YouTube, for example, for the money.
01:16:39I don't have sponsorships.
01:16:40I advertise my courses, yes, but I keep that super brief.
01:16:44And in some videos, I just forget it.
01:16:48But of course, I do care about views.
01:16:49I want to reach people.
01:16:51It doesn't help me and anybody if I create videos that don't reach anybody.
01:16:56That's not what I want to do.
01:16:57I want to try to spread useful knowledge with as many people as possible, but I'm not doing
01:17:02that for the money.
01:17:05And I haven't ever done a paid placement in any video, and I don't plan on doing that.
01:17:11But yeah, so I try to find the right balance there.
01:17:13And therefore, it means a lot to me if I read that I seem to succeed at that, at least for
01:17:18some people.
01:17:19Obviously, you can't make everybody happy.
01:17:21But yeah, I also would like to thank you a lot, Max.
01:17:24You've helped me a lot with your educational content.
01:17:27Learned a lot from you.
01:17:28It doesn't matter the format of the content.
01:17:29I always will be able to hear from you.
01:17:33Thank you so much, too.
01:17:34And thank you all.
01:17:35Thank you all for joining this stream today.
01:17:38Had a lot of initial problems about which I'm very sorry, but at least now it's stable.
01:17:43And I'll investigate to hopefully make sure it won't be as glitchy next week.
01:17:49But yeah, thank you all so much for being here and for the very nice comments.
01:17:54And I will keep on trying my best to create valuable content, valuable courses, valuable
01:18:01videos here, share my progress.
01:18:04And I got many, many ideas.
01:18:05Time is the only issue like it is for all of us, I'm sure.
01:18:10But yeah, I'll keep on trying to share useful stuff here.
01:18:14Would you be interested to make some boot camp where you have some amount of students and
01:18:17you teach them how to work in teams, etc.?
01:18:19I've considered it, I've considered it, and in general, I would be interested.
01:18:25If I do it, though, I want it to be really, really great.
01:18:29And right now I'm struggling a lot with time.
01:18:34So I've played around with it.
01:18:37I created some concepts here and there, but it's not something I'm fully happy with yet.
01:18:44So I don't have any boot camps planned right now.
01:18:48And also not in the near future, but in general, I like doing these live streams.
01:18:52So doing some educational content that also involves some live element, not like a live
01:18:58class, but maybe a course with live check ins or something like this would be interesting.
01:19:05But yeah, nothing concrete to announce right now.
01:19:08But with that, I also got to leave.
01:19:10So yeah, thank you all for being part, for sticking with me through all the technical
01:19:14problems, the recording of the stream or of the streams, since I restarted it two times,
01:19:21will stay online and I will, will I be back next week?
01:19:25Let me check.
01:19:26No, next week on Thursday, I will not be here.
01:19:31Maybe on Wednesday, I'll see.
01:19:34But if not, then the week thereafter.
01:19:37On May the 21st, I'll be back and maybe on May the 13th.
01:19:44I guess you'll see when I'm online, it'll always be the same time.
01:19:48And typically, it will be on Thursdays and the plan is to stream every week.
01:19:55Yeah.
01:19:58So lots of nice comments.
01:20:02Thank you so much.
01:20:03Also Yaser, Hasan and Liskasmael and yeah, I see what you're saying here.
01:20:16But yeah, thank you all.
01:20:17Thank you all and have a great evening, day, morning, whatever it is for you and thanks
01:20:22for being part and hopefully see you in future streams.
01:20:25Bye bye.

Key Takeaway

Bun is evolving from a high-performance JavaScript runtime into a specialized tool for AI agent development while maintaining its core utility as a fast, consolidated package manager and test runner.

Highlights

  • Bun functions as a comprehensive toolkit combining a runtime, a package manager, a bundler, and a test runner in a single high-performance binary.

  • The bunfig.toml file enables a 'minimum release age' setting for package installations to mitigate supply chain attacks by delaying updates for a specified period, such as three days.

  • Modern web application development increasingly shifts toward a stack combining Bun and the Hono framework for lean, elegant REST API construction.

  • AI agents demonstrate higher efficiency when given direct file system access for document retrieval rather than relying solely on traditional vector database connections.

  • Next.js App Router initially faced adoption hurdles due to aggressive caching that caused production behavior to differ from development environments.

  • Transitioning from complex frameworks like Astro to vanilla HTML, CSS, and JavaScript for simple sites reduces dependency risks and maintenance overhead.

Timeline

Stream Troubleshooting and Technical Challenges

  • OBS encoding overloads often persist even after switching to faster preset settings.
  • System resources may appear clear while background streaming processes continue to lag.
  • Technical stability in live streaming requires consistent hardware configurations that can still fail unexpectedly after dozens of successful sessions.

The troubleshooting process involved adjusting OBS presets to 'really fast' to combat encoding overloads despite minimal system activity. ChatGPT provided specific instructions to fix the lag, which eventually stabilized the stream. This technical friction served as a contrast to the primary topic of software performance.

Bun and Hono for Modern Web Development

  • Bun and Hono constitute a default stack for building elegant and lean REST APIs.
  • Bun includes built-in file system routing and a dedicated routing API without requiring external dependencies.
  • Documentation optimized for Markdown facilitates easy information extraction by both human developers and AI agents.

Bun provides internal routing capabilities that allow developers to define different functions for specific HTTP methods and paths. While Hono offers a superior middleware structure, Bun’s built-in APIs remain highly elegant for server setup. The runtime's evolution may prioritize features tailored for Anthropic and AI agent construction.

Vector Databases vs. Agentic Search

  • Agentic search allows AI models to browse local file systems directly instead of querying vector databases.
  • Vector databases remain necessary for managing datasets exceeding hundreds of thousands of documents.
  • Current AI development is in a 'Wild West' phase where agent frameworks and services change daily.

Traditional RAG systems using vector databases like Qdrant are becoming less critical for small-to-medium datasets where agents can simply navigate a file system. While semantic search has its place, the efficiency of giving an agent direct document access is a growing trend. Managing the context sent to LLMs remains the primary challenge for current coding agents.

Supply Chain Security in Package Management

  • The 'minimum release age' setting in Bun prevents the installation of packages that have not been public for a set number of seconds.
  • Supply chain attacks are typically detected and mitigated within a few hours or days of a compromised release.
  • Bun serves as a faster alternative to NPM for package management even if the Bun runtime is not used.

Configuring the bunfig.toml file to require a three-day release age significantly reduces the risk of installing malicious code from recent supply chain attacks. This security feature is also present in pnpm and recently added to NPM. Bun further integrates a test runner and bundler, though dedicated tools like Vite still offer more specialized feature sets.

Remix 3 and the AI-Optimized Framework Shift

  • Remix 3 aims to be a framework optimized specifically for AI agent comprehension and pick-up.
  • React and Next.js remain the default suggestions for most AI coding models due to training data dominance.
  • A framework must offer significant architectural advantages to win over developers who now use AI on autopilot.

Remix is pivoting away from its React-centric roots to become a framework that is easier for AI agents to understand and use. This transition is difficult because AI models naturally default to the most common technologies found in their training data. For a new framework like Remix 3 to gain traction, it must provide a developer experience that clearly surpasses current standards.

The Impact of AI on Coding and Education

  • The role of the developer is shifting from writing syntax to architecting systems and reviewing AI-generated code.
  • Generalist developers with deep expertise in specific niches are becoming more valuable than pure specialists.
  • Reducing package dependencies lowers security risks but increases the long-term maintenance burden for the developer.

AI has removed much of the 'flow state' found in manual coding, replacing it with the task of reviewing and prompting. Effective education must now focus on patterns, best practices, and architecture rather than 40-hour deep dives into syntax. While building with vanilla JS and minimal packages is safer against supply chain attacks, it requires the developer to account for every edge case manually.

AdonisJS and the Full-Stack Perspective

  • AdonisJS provides a batteries-included experience similar to Laravel for the JavaScript ecosystem.
  • The industry's focus on React single-page applications likely stunted the growth of dedicated full-stack frameworks like Adonis.
  • Adonis includes a SQL ORM, authentication, and a templating engine (EdgeJS) out of the box.

AdonisJS follows the philosophy of providing everything needed for a web application in one package, contrasting with the modular but fragmented React ecosystem. The framework is seeing a slight upward trend as interest in full-stack solutions grows. It remains a viable alternative for developers who prefer a cohesive, structured environment over piecing together disparate libraries.

Career Strategy and the Future of AI

  • Data structures and algorithms (DSA) are becoming less relevant in recruitment compared to systems design and AI proficiency.
  • Standing out in a flooded job market requires building unique AI-leveraging products rather than standard CRUD clones.
  • GitHub's recent performance issues stem from unprecedented growth and the influx of AI-generated code and issues.

The current job market is saturated with junior applicants, making personal branding and unique portfolio projects essential for visibility. Developers should focus on learning how to design systems—optimizing for cost, reliability, and scale—rather than just frontend components. As AI accelerates, the gap between academic teaching and industry needs continues to widen, necessitating self-directed learning.

Community Posts

View all posts