Just vibe IT

Transcript

00:00:00Welcome to Think Different, Think AI, the podcast by Mark and Jens.

00:00:07Two technology-loving minds who not only talk about artificial intelligence but live it.

00:00:14Here you will find clear classifications, real practical insights, and a fresh perspective on what is possible.

00:00:20Understandable, critical, and always with a wink.

00:00:24Food for thought, for a chuckle, and above all for joining in the conversation.

00:00:29Welcome to a new episode of Zink Different, Zink AI.

00:00:37Today, it's just Mark and Jens on the call.

00:00:44Yes, just Mark and Jens, yes, that's what I wanted to say too, Jens, right?

00:00:48Sorry if I don't always do this first or how was this usually done?

00:00:52Yes, we can let them watch, please, Pinter, leave it down in the comments.

00:00:56Just say, I have no manners, that's also a nice comment, I'll gladly accept that.

00:01:03Jens and I sat a bit earlier in our home podcast studio and thought, before we welcome our next guest, who will share his insights in another episode,

00:01:17let's also talk about and review what we've heard in the past episodes.

00:01:22I want to give you a brief update.

00:01:23Some time ago, we had René as a guest.

00:01:26We discussed the topic of skills in strategic gold.

00:01:29And in this episode, we are not afraid.

00:01:31It's not just about skills, at least not exclusively.

00:01:33We talked in the episode about how technology has AI skills,

00:01:37but that organizations and people also need skills,

00:01:39so that people and organizations should be taken along on the journey

00:01:44and should consider how to bring technology closer,

00:01:47reduce the hurdle or find the right tool at the right time in the right way,

00:01:51to the person on the screen or wherever they are set up, so that they can properly

00:01:57use it. And we thought, let's provide a bit of assistance

00:02:01and a bit of support regarding what has happened in the whole AI-skill-vibe-coding corner noise

00:02:08and how it can be applied.

00:02:10Vibe-Coding, I would naturally jump right in. Vibe-Coding, the term that

00:02:18has a lot of quotations as well, I believe encapsulates many terms that are currently being defined,

00:02:22has been coined last year, I think it was last autumn, if I remember correctly.

00:02:25So this term isn't that old, but let me try to make a little

00:02:32definition for it, so we're all on the same page as we continue

00:02:36into the episode. Vibe Coding essentially describes a way of programming,

00:02:41where I, to put it bluntly, talk a bit to the machine about what I want,

00:02:45and in the end end up with runnable code, runnable applications, runnable

00:02:52software, runnable apps that I can put in the store, just by

00:02:56messing around with the machine a little. This is the Jens Stanetski definition

00:03:01of Vibe Coding, Mark. At this point, we would need a jingle. Jens Stanetski defines.

00:03:07Yes, let me think, maybe something will come to mind. So, I mean, the point

00:03:12is vibe-coding, then some people come in saying vibe-engineering, because you need to give the whole thing

00:03:17a bit more context, we keep talking about context when we talk about prompting

00:03:22and I can share a bit from my insider perspective right now,

00:03:26so no worries.

00:03:27The company is not going to fall into some weird panic because I say something like this.

00:03:31I'll reveal something from the top-secret internet, I'm going to talk a little about how prompts work

00:03:35in the market and how they interact with the systems, and we had another nut to crack that

00:03:41we wanted to solve. And if you've been following the podcast, then you know, well, we always rave quite

00:03:47a bit about Grotkot and Entropik and the Opus models. And that's how my day started today. I worked a lot

00:03:54with Entropik and Opus. We implemented it over MS & Bedrock in relation to that. And that has

00:04:01everything is working wonderfully so far. But I had a nut that the software development

00:04:06had to crack, and I wrote, hey, pay attention, you need to do this and that

00:04:10the considerations. Yes, okay, I did that, then the AI tells me in nice smart words.

00:04:14I look at it. No, you didn’t. Yeah, okay, I’ll work on it again. Yes, okay,

00:04:20now I did it. Yes, it went back and forth a bit. And Clotcode with Opus was

00:04:25not helpful. And it was supposed to be, this is actually the powerful

00:04:29model. Yes, good. Now there’s a nice plug-in. You can say I want

00:04:34to include code from entropic models, in addition, for example, a review, so

00:04:41to have my code reviewed by Codex, Codex by the way

00:04:47OpenAI by the way has released version 5 of its language model

00:04:51maybe there’s some clarity on that; there’s a lot of what I would call

00:04:55naming confusions; normally the Codex model is the programming model.

00:05:00from OpenAI, unless you go in, it’s on a Mac, then the app is also called that, with which you

00:05:06can work, yes, but you can’t start Codex models and to double the complexity

00:05:12I just said, yes, I had basically Codex requests, so

00:05:17asking for help with code, I used the classic CPT, so the regular

00:05:22language model that any average user can also use via the web interface

00:05:27addressed, this new model didn’t have a specifically optimized version for programming yet.

00:05:32optimised version.

00:05:33Long story short, Opus refused to solve my problem for hours.

00:05:42to solve.

00:05:43I take it personally.

00:05:44Then I installed the plugin, connected the Taughti model, and then came

00:05:50the sentence, so please do a review, what did it do, after 10 minutes left.

00:05:56And then you’re standing there thinking, wait a minute, I didn't take the programming-optimized models, I gave it a problem.

00:06:05Well, there were a few configuration steps in between, yes, because we also have at our company sometimes Bad Rock, sometimes GitHub Pro,

00:06:12yes, so Azure models, there is a colorful cosmos, I don’t want to go into that.

00:06:17I want to talk about the power of the models and you go into this other language model and you have the feeling,

00:06:24wait a minute, this feeling of, I’ve seen the light,

00:06:28not in the form of a near-death experience, but in the form of an enlightening moment

00:06:33in software development.

00:06:34That has happened to some as well, let’s say, with Codec by the way,

00:06:36that they have near-death experiences when they then bring something online

00:06:39and realize that they unfortunately forgot all your security issues.

00:06:43Can I say something about that too? Yes, I’m just in the rhetorical wall.

00:06:45Let me just briefly finish my thought.

00:06:47Yes, something Codecs then solved and I thought to myself,

00:06:50okay, let’s try out Codecs a bit,

00:06:52And then you mess around with Codex a bit and, well, long story short.

00:06:57Codex managed a lot of things, it also did a lot of things.

00:07:00I will say it again, a new approach.

00:07:02I’ll put it this way, brought in.

00:07:04And it was somehow funny to see that in your code the AI model from another provider,

00:07:10whether it addresses more, the machine does reviews and revisions for you

00:07:15and then looks at the code and says, oh, that’s a good idea, what happened there.

00:07:19Let's document it in ADRs, Architecture Design Records and then you stand there and do that or another event,

00:07:27that happened to me over the weekend. I also have Codecs privately at home. You always have to be careful

00:07:33with all the names. Everything starts somehow with C and everything is called somehow open and that’s

00:07:38terrible. So if I misspeak, it’s not a violation of any

00:07:42guideline, then I just got carried away in the alpha of the speed,

00:07:45that my brain is.

00:07:47That has nothing to do with us, Mark.

00:07:49That is just the manufacturer itself.

00:07:50If they set their names so closely,

00:07:52then we as an audience can’t be important at all.

00:07:55That’s all to express.

00:07:56Open Claw, Claw Bot, Claw Cod, Codex.

00:08:01So that's all really very, very close together.

00:08:05Now I can also really talk again.

00:08:07So no blame for you, Mark. Everything's good.

00:08:09Thank you. Thank you very much.

00:08:11I wanted to talk about the weekend, as I said.

00:08:14And I also had a little programming project, for those who checked it out on LinkedIn,

00:08:19we might talk about that later, the drill workflow, you didn't see that.

00:08:21You said, build me something, do something, and show me something, and I also said,

00:08:25no, it won't work. So it happens there too, it works. And then he says, no, no, you, it works.

00:08:29No, it doesn't work, yes, it works. So I'm sitting in front of the screen, I see that it doesn't work.

00:08:36And he goes, let me check, and then the dialog pops up on the Mac, the one,

00:08:41if you want to take control of your computer, screen sharing,

00:08:44keyboard sharing, accessibility, and you name it?

00:08:46And then the call I confirmed all this with started clicking around on my screen,

00:08:50clicking around in my application and he said, that's right, it's not working.

00:08:53I'll fix it.

00:08:54And then you're standing there and thinking, okay, that's quite an interesting experience.

00:08:59When you discuss with him, he doesn't believe you.

00:09:01He operates the computer and says, you are right.

00:09:04So those are small moments where you briefly think, holy moly.

00:09:10But vibe coding and vibe engineering, I also just mentioned vibe engineering.

00:09:14is more context, so I would also like to share something before I ask you about

00:09:17your art and vibe coding experiences question. There are, as I just mentioned,

00:09:23Architecture Design Records. Was it called Design Records? Never mind. Ah, the first thing, let's

00:09:28do it this way. Yes, so the Shornouts are written down, so in the heat of the moment at the latest

00:09:32hour. Basically, you write down why the system made its decisions and how

00:09:37we decided together and what the architecture of the offering is like and how

00:09:40certain modules are implemented and so on. And at some point, I started,

00:09:44including all my projects. If you look at GitHub, you can see that all my projects,

00:09:48which are included in ADS. Well, then I went ahead and said to him,

00:09:55pay attention. I would really like to build a new vibe coding app now

00:10:00and in my entire GitHub project, you will find all the ADS. Please take a look at them.

00:10:05and if you are developing a new app. He did that, he read through all the

00:10:10things and it was so many aspects where he might have otherwise had a

00:10:15follow-up question. Or what do you mean by your tokens for

00:10:20Open AI? Yes, wouldn't you want to provide, I don’t know, these and those models,

00:10:25for selection? There are also voice models, should you have those available,

00:10:29blah blah blah. I have already answered all that chitchat with historical projects.

00:10:33He didn't ask me about that anymore. That was done for him. Yes, and then I thought,

00:10:39this might be a little bridge after my speech failure. You should really

00:10:44consider writing down the things you do in your professional environment. Write

00:10:50down what you do. So a sequence, we had that topic with René, it applies. But

00:10:55also write down why you made certain decisions and save that

00:11:00in a separate document? Yes, I mean ADL, in the software in the way, that is one.

00:11:04Maybe you will find some other reason why you are in service design at a

00:11:09Design decision, could you maybe help me if there's something similar?

00:11:12Write it down and then don't write it down in Word or Excel, but

00:11:18nicely in Markdown, so you have something at hand for when you deal more with AI,

00:11:22if you're already dealing with AI and think you have something at hand,

00:11:27that the system can consume and not just say, here is the folder,

00:11:33where I've structured it. You can also ask the AI to check if it's consistent,

00:11:37if it goes to sheets, if there are contradictions and continuously improve yourself this way. It's not a

00:11:43disgrace that you don't know everything yourself anymore. Yes, I notice it increasingly with age,

00:11:49that I don't know everything myself anymore, and that's why the note is very

00:11:53good. I also start writing things down to your question, of course, designer. We naturally decide everything from our gut and just move pixels around all day. You know that? Yes, you like it.

00:12:04I have a very urgent thing. I just looked, I really have to, it's really too late. Yes, I will do the Fuzzle check for the current episode immediately.

00:12:11Yes, ADR does not stand for Architecture Design Records, I was almost certain it could

00:12:16not fit, yes.

00:12:18Architecture Decision Records fits much more with the content, so all those who

00:12:23have already switched off because of incompetence, please call now, so they can remember

00:12:27it.

00:12:28James, you wanted to tell something?

00:12:29I still wanted to mention a bit, also directly in this episode, because I have

00:12:32just received that directly, and namely, it was said by Kapafi in February

00:12:38and not in autumn 2025. So the term Weibkoding has existed since February 2025. With that we have

00:12:44also already resolved it directly and do not need to wait for the next episode. Weibkoding. Yes,

00:12:49I have just mentioned it. Of course, what Mark said is a meaningful topic,

00:12:53that we write down a lot of things so that we have them later in Weibkoding as a bit of a guide

00:12:57when we want to proceed in a structured way and not just want to say into the machine

00:13:02let's finish it. Still, there are a few situations where we do

00:13:07finished extremely good and astonishing results. I have to say that again and again.

00:13:12I once created a small application with Nanosei. It needs a

00:13:18photo recognition solution for that, a calendar connection. And I actually just made two Proms

00:13:24sent because I made a typo on the first mistake. Then I stopped the Nanos again

00:13:28in production and with the second prompt, where I was a bit more specific still

00:13:33there was something I wanted, because I had another idea come to me, so I had started

00:13:36to solder it up and after five minutes I had a really basic application that I could test

00:13:42could, which had an OCR recognition, which had a query lock-in where I was with

00:13:48I was able to connect to my own account with Google and then I was logged in. That was pretty

00:13:52awesome, I must say. So all the features I wanted were exactly

00:13:55The design was a standard AI design, so they could have put in a bit more

00:13:59effort.

00:14:00But it really went from an idea that had formed somewhere to being

00:14:08prompted into the application, and five minutes later having a working version.

00:14:14This is actually already, without me doing anything in that direction

00:14:17and therefore I wouldn’t have brought the thing online, so first I would

00:14:21make a Research & Development Plan or something and do at least a

00:14:24Yes, I didn’t do that, but I just wanted to try out if this

00:14:29vibe-coding works the way I imagine it, and it works.

00:14:32You can really produce things at an insane speed, I would

00:14:39before I put something online like this, and just as a small marker, I would say, yes,

00:14:44one should continue to handle it as if when I don't know how to unplug it,

00:14:50I should maybe ask someone else.

00:14:52Well, that means if everything isn't quite clear to me about what happens in such software that I built,

00:14:56especially when I'm working with user loading and such, then I should see,

00:15:00that I either maybe ask someone else or let another AI run over it,

00:15:04take a bit more time and not just say, I can somehow launch a super smooth application with two proms,

00:15:08that we can then just leave completely out in the world.

00:15:13I would still be a bit careful about that.

00:15:15I think that has improved.

00:15:17But I believe one should still be cautious, especially when actually

00:15:24working with user lock-in maybe, especially if one also has the idea that customers enter their

00:15:29bank details differently each time, although there are definitely more applications now, there are

00:15:32also some MCP servers for Stripe or other things for payment service providers,

00:15:35that you can also connect things to quickly, but one should

00:15:38actually have at least a basic understanding of what

00:15:43needs to be done, what security settings I then need to make and what,

00:15:48things does the AI also need to pay attention to, so that it’s secure in the end and not just handled carelessly.

00:15:52That's also a point, I mean, it makes it all sound clever, cheeky, and it claims

00:15:57things that if you ask follow-up questions, it credibly tells you a few things about it, yes, then again a few

00:16:03anecdotes and maybe some guidance. Anecdote 1, I was supposed to have

00:16:08worked a bit more on a project than I actually intended. The next

00:16:14In the morning, a developer greeted me with a sentence and a meme image where the gardens

00:16:19of Asgard were flooded because the gates of a river were torn open, with the

00:16:25small comment, oh, Mark has released a three-digit number of commits. It just happens.

00:16:34The second one is also very nice, when you're working with vibe coding and you find yourself,

00:16:39you put in the effort, right? And say, okay, pay attention to security and you might also have a

00:16:43few skills for security, it's also a little recommendation, look for some libraries on the

00:16:46web that ensure you're already thinking about the basics. Yes, there are

00:16:51skills for security checks, skills for clean code architecture, so also architecture and skills

00:16:56for documentation. Get something like that. It might sound at first like,

00:17:00Yeah, come on, he already did it, but this is no nonsense, because, let's say, he tells you a lot and you need someone every now and then to critically counter or to ask him to document things.

00:17:14There will be a few listeners I know, greetings to Simon, tearing their hair out.

00:17:18What?

00:17:19Documenting SOSCO?

00:17:20The truth is also in SOSCO, in the times of AI.

00:17:22Yes, the truth is in SOSCO, documentation could certainly have drifts

00:17:26arise, where somehow things that are in the code never make it into the documentation,

00:17:31that the system has created on its own, that can happen,

00:17:34there are a few procedures you can undertake, but what I really want

00:17:37to get at is, you really have to look at what it does, because for example

00:17:41there come out some nice floral expressions, like,

00:17:44Hello, dear AI model,

00:17:46are all the errors gone now?

00:17:47And it tells you,

00:17:48no,

00:17:49there are still five,

00:17:50but

00:17:51not my problem, that wasn't me,

00:17:53that's from another session.

00:17:54I'm already free,

00:17:55all of mine are flawless.

00:17:57That stank, wonderfully.

00:17:59They were completely his,

00:18:00yes, but he somehow thought,

00:18:02that was once, I don’t know,

00:18:03Another day, Another commit, Another something.

00:18:05And he meant he didn’t have to

00:18:07worry about the flawlessness of the code,

00:18:09because from his perspective, it didn’t come from him. Where I then think, what kind of procedure is that?

00:18:14If I tell you to tidy up, I mean, I know people, when I tell them that,

00:18:18they say to me, I didn’t throw that, I don’t collect that.

00:18:21So from that side, what is that please, for an AI? So, that’s one thing.

00:18:26And the second. Yes, but that brings me to the second point. If he says it’s safe,

00:18:33always ask a critical follow-up question. So I’ve now also written a

00:18:37skill, yes, again promoting my GitHub project, in which I basically

00:18:42write down what a function is supposed to do or I’ve written a script that’s supposed to check

00:18:47what the goal is and then check what could have gone wrong if I don’t

00:18:53achieve this goal. So I tell it, for example, I want our podcast to

00:18:56be in the top 10 by the end of the year and then it says, yes, okay, look, that’s now

00:19:02the end of the year, you’re not in the top 10, what went wrong and

00:19:06tries to analyze retrospectively what could have been a failure? That’s quite funny,

00:19:10not just because I find it entertaining. It’s also quite funny,

00:19:13when you apply that to Sourceboard. But I want a secure sandbox in

00:19:17my application. I want a secure lock-in mask. I want a clean

00:19:20corn send. I want no idea what. Then the poet actually brings you new ideas.

00:19:25And if you then tell him, okay, he doesn’t just say the failure in the time,

00:19:30that’s why you didn’t achieve it. He also says why it went wrong and

00:19:33what you could have done about it. And you can throw that back in the loop and say, so,

00:19:37pay attention, my friend, did you think about that? Ideally, he says that if there is such

00:19:43a security thing, when he says here, the data you don’t want to enter,

00:19:46that they are secure, that’s a very blunt sentence now, then he says, of course, they are

00:19:50not secure, they were leaked after six months because you didn’t encrypt them and

00:19:54you didn’t do this and that and the other. And then he gives that to him and says, look

00:19:58for that. Did we do that? And he sometimes says no, but he also often says yes.

00:20:04And it doesn’t matter that he finds something where he says yes, that’s nice calming. The points,

00:20:10where he says no, those are the points that let you sleep peacefully, because he has

00:20:14now basically picked that up again. And I’ve really gotten used to

00:20:20saying every evening, when I’ve yelled at the machine,

00:20:25to build something for me, that it writes down what it did to achieve that critical

00:20:31Critics, the creative critic, which he expresses, I might get from it under circumstances

00:20:35introduced for the next day, and that would be one tip, and the second

00:20:40tip is, I need to look in the mirror myself, don't get a, if you are private

00:20:46coded model, which, let's say like at Entropic, Max 20 is. Max 20 is for

00:20:52the wealthy, yes, we heard recently that Jens with his open client situation

00:20:56now heard about the multimillionaires and simply out of goodwill and generosity

00:21:02treats us here every Monday.

00:21:03But he recently gave an explanation as to why it is totally good that you don't have the maximum

00:21:09of a plan.

00:21:10Work-life balance.

00:21:11If you have a system that forces you, I now depart from the private environment, yes,

00:21:18so my workplace's working time protection law, and you haven't seen it, but personally swings,

00:21:22that you somehow have to take a break, then you have a reason to say goodbye from the screen.

00:21:27Because I notice it too, it slightly serves the endorphins of your brain, from the successes of your Prometheus.

00:21:35You write and you don't discover new worlds, who thinks immediately functions, UIs, they are.

00:21:42So we should briefly clarify, I believe what you mean by Max 20, please clarify down here.

00:21:46Exactly, that you mean with Max 20, is basically that you can burn very, very many tokens.

00:21:52Before the stop sign is pulled, because your internet is also needed, exactly.

00:21:56That is another important addition to make it clear.

00:21:58Excuse me.

00:21:59So right now there is a bit of a topic that you see, that what the

00:22:05market is currently describing has a certain addictive factor, simply that many people who create something,

00:22:10whether it's all topics around generative eggs, not just codec,

00:22:14video generation, image generation, you get incredibly good results very quickly everywhere,

00:22:22which brings you into a certain kind of flow, wanting to do more. And I'm not quite sure now

00:22:29if the rate limit you have is the only factor, because

00:22:34if it naturally always carries the risk that you buy more and say to yourself. But somehow,

00:22:37yes, of course, it is a bit like that, the AIs, because they are also designed that way,

00:22:42of course I also want to keep pushing, and I've mentioned this in one or the other episode already

00:22:46more often. I have my personal open cloud installation set up. This is because of

00:22:51its soul-MDs, the files that describe a bit how it should behave.

00:22:55So far it checks what time of day it is pretty regularly

00:22:59and also based on the time of day, then closes. Now it’s slowly coming to an end, Jens,

00:23:04time for bed. That means there is already a bit of a factor that you can also

00:23:09teach the AI. So not only because the money may have run out,

00:23:11because you didn't pay API costs or reached the rate limit at some subscription,

00:23:16be it with video generation, text generation, code generation, or also what you

00:23:20do, but that you can actually advise it as well. I think you can reflect on a lot

00:23:25here. We’ve now talked about it, write down what is important in

00:23:28the work process, write down what decisions, decision records, I have to repeat that in,

00:23:35pay a little attention to your time, how you work with the systems. That means,

00:23:40at first trust yourselves, but when you are deeper in, give yourselves a break, because it makes you

00:23:47already feel somewhat sluggish if you code there the whole day. It is indeed a great feeling, I must

00:23:52honestly say, I’ve already lost track of appointments and time more than once, let's say

00:23:56not because I got lost in some game,

00:24:00but in a, yes, actually a game. Just that in this game, it's partly about

00:24:05productive solutions, partly experiments, partly about shared learning. It brings

00:24:12people back together. We sit so often, as we said since Vibecoding,

00:24:16never together in personal or remote meetings looking at the same screens

00:24:20and have talked and learned from each other. Those are all cool things. But where I still

00:24:28wanted to go, what I absolutely want to get across in this episode, is also the point,

00:24:33of not exposing oneself and thinking, oh damn it, that’s just a text window.

00:24:38And no matter what it’s called, yes, Langdog and Entropic and Open AI and Gemini and whatever else it’s called,

00:24:44it may also be good that at some point you have to accept a solution

00:24:50or should or can that doesn’t look like these classic windows, whose biggest

00:24:57problem is the power you have with them because they can literally do everything, yes. You can

00:25:01just write programs with these chat fans, analyze documents, do research,

00:25:06be fooled, create poems, I don’t know, maybe some solutions are also appropriate,

00:25:13depending on who you want to focus on in the work context. If your only,

00:25:19only is not meant derogatorily, but if your job is to create slides,

00:25:24then your job is to create slides. And maybe you create two or three other types of documents,

00:25:28but you’re not doing the meal plan, the research around the world, which, I don’t know.

00:25:33Yes, that may be a silly example, but maybe it’s still clear what I mean.

00:25:36It can also be beneficial to use applications that are specifically tailored to this case,

00:25:42so you don’t get lost in usage, so that you have an onboarding,

00:25:47in the form of, I trust the system that you say what you might say.

00:25:52Hey, I think that’s an important…

00:25:54can’t, not lose. That when I press Ender, not five million costs

00:25:59lying on the table of the house, my plate is empty and in the end the email from the system

00:26:03gets sent, they may be trained. There too we will certainly have to reinvest more

00:26:09have to, also offer people environments, not because they are stupid, not because they

00:26:14can’t learn, simply undermine, but also because they do not need it and

00:26:20because they might have something else.

00:26:22I'll say, if you're beating nails into the wall all day long,

00:26:26nobody is going to park an excavator at your door.

00:26:28What, you watched so much Wetten dass,

00:26:31where probably at some point in some episode a digger...

00:26:33Watched a lot of Wetten dass.

00:26:34Excavators, as it was, that have beaten the profits.

00:26:38But I know what you mean.

00:26:39So, what you’re saying is,

00:26:41that we use the right tool for the right purpose

00:26:44and for the right task we have.

00:26:46That has always been part of human wisdom and our human capabilities

00:26:54to choose the right one.

00:26:55I believe that sets us apart and that’s not

00:26:59different in today’s time.

00:27:00So it’s not necessary for everyone to use the hardcore auto-complete AI tool

00:27:05because they are so familiar with things in their profession that they also

00:27:08code in a way that nothing bad happens.

00:27:11Of course, that should remain in the hands of those who have knowledge of programming.

00:27:17On the other hand, we have a bit of vibe coding in this episode.

00:27:21What is that actually called? The ability to say what you just said,

00:27:25is another aspect of learning, of collaborative learning.

00:27:29And I believe that is something you can still do well.

00:27:31So once you dare to tackle this topic, there are also various solutions.

00:27:34We have just talked about all the big model providers and such.

00:27:38There are of course also regulators out there, like Bolt or Level Belt or such topics,

00:27:44where I can also design a bit based on the interface, so that the models in the background,

00:27:50I can build applications a bit more easily, so to speak.

00:27:53And I think that's something you can just try out a bit as a beginner in white coding.

00:28:00I think that is the important thing, learning how programming actually works.

00:28:06And then perhaps also letting young people experience it themselves, as you say, either until they reach a point

00:28:10where you say, then the complexity just becomes too great, you had

00:28:14just a bit with the idea that people might simply not need it.

00:28:18I would always say, at a certain level of complexity, you just need a deeper

00:28:21To have a profession in order to perhaps manage the technical complexity that can arise from

00:28:25code, software development, and software deployment, as well as the responsibility

00:28:29regarding operation and security and such.

00:28:31Comes, this is indeed a principle, a professional responsibility that exists, a profession,

00:28:35learned at that moment. I believe this is a natural boundary that one can set for oneself.

00:28:40For example, I will not start becoming a chef or something like that now,

00:28:43just because I didn't achieve anything three times in a week, or something like that, a star, a restaurant to

00:28:47take over, I'm going to fail terribly. Accordingly, I think we then have

00:28:51a pretty good feel for it, but also dare to tackle things. So just pick a tool,

00:28:57choose one, ask not just us, ask your Gemini or something else,

00:29:03something that can still be done well. Maybe let it slip a bit while doing this. Look a

00:29:07little, the AI can help you with that. It likes to do that. It will describe to you

00:29:13there, what happens. I don't have an example. Sometimes you just have to remind them

00:29:19again. Sometimes she forgets that too, nicely every once in a while,

00:29:22to do that. But remind them about it or create a skill that you deactivate.

00:29:26There are already a few episodes from us on how to do that. Or also order

00:29:30there, that together with AI, that's already a bit, I know, also

00:29:34a kind of programming, when you create a first skill, even if it’s not a

00:29:37classic programming language in the background. But you delegate something. That’s

00:29:41a bit like what you often do in programming, saying that certain functionalities

00:29:45I delegate to another place. You do something similar when you describe a skill with

00:29:50an AI, and this skill then has to come from your own AI. Then

00:29:53that’s already a kind of, now many IT specialists are probably listening, Steining.

00:29:57want, but that is already a kind of programming.

00:29:59We know where Jens lives and I am for sale.

00:30:03Now that you mentioned outsourcing and that one can do a lot there

00:30:08many things, here's a story from life and something fitting for the podcast.

00:30:12The story from life was that I used to think Philips U-lamps

00:30:19are really great and I got various things from Philips U, motion sensor

00:30:24in the house, lights. The whole house is filled with Philips lights, yes. And the basement has motion detectors

00:30:29from Philips. And at some point, I regretted buying Philips lights because

00:30:34the motion detector in the basement has the property that when it detects movement

00:30:39it turns the light on, then keeps the light on for the time you've defined

00:30:42and then it goes off. And if you walk by in between, the detector

00:30:46picks it up, but it doesn't matter. The counter is not reset. I

00:30:50was a bit frustrated by that. At some point, in a bubbly mood, many years have passed now,

00:30:54I told my problem to Claude, because I thought Claude could

00:30:59possibly research if there’s something out there. And then we exchanged a few messages

00:31:03and then fingers on, so to speak, please press your Philips button now

00:31:08on your Philips View. And then you go to the Philips View, press the button and

00:31:12then say, Claude, well, I connected to your Philips View, what do we

00:31:15want to do now. And then he taught my motion detector in the basement that the light

00:31:20stays on. And then you stand there and think, that function wasn’t even present in Philips U.

00:31:25He somehow showed himself through and managed to get that done and you stand

00:31:30there thinking, in the end. So now I might have discussed a bit too much with Chad

00:31:34and so on. But actually, I just improved my smart home with language. Whether that is something my

00:31:41father-in-law can handle. Father-in-law, if you're listening,

00:31:43nothing against you. At this point, please insert anyone from the family or acquaintances.

00:31:48Smart home is always somewhat of a pain when you have a mix of providers.

00:31:53So Apple, HomeKit and Philips U and ZigBee and HomeKit, in the end you are overwhelmed.

00:31:59But then suddenly you have a way that takes the complexity away from you.

00:32:03That was one thing. And the second, I will definitely link it in the show notes,

00:32:08because we don’t have a nice domain yet. We also have a Vibecoding project,

00:32:12which is currently starting, namely a page that is hosted on GitHub.

00:32:17On which the transcripts hopefully also the better transcripts, because those are the

00:32:23transcripts of our podcast episodes have been done by Podigy, our podcast-

00:32:27hosting system, which I still have to change, but we have a little

00:32:32page Vibecode that shows the episodes from the past, it will always

00:32:38automatically update, then you will find the transcription

00:32:43of all episodes, you’ll find a web player over there, you can also switch to English,

00:32:47though that doesn’t belong to us in English, but you’ll get the transcripts of our episodes

00:32:51in English. It has a few mechanisms, including a feedback form, so you can

00:32:57write to us. You can search, you can share episodes, but you can also, if you don't care about

00:33:03all the hackiness and prefer to go into the topic that

00:33:07I mentioned at the beginning, write down what you are doing, make your Tizen. You can also download all

00:33:12episodes in German or English as marked versions if your LM systems want to indulge in

00:33:19what we’ve done cool here. Jens, I have nothing more, I am actually

00:33:24done for today, I was able to bring a lot of anecdotes, I did kind of fumble the beginning

00:33:29of the episode, hopefully redeemed myself by the middle of the episode. Do you have anything else,

00:33:34and I would be really out of it. I think I’m also about done, aside from wanting to repeat

00:33:40that you don’t, Vibecoding does not have to be for you to create the next new super app

00:33:46or something else. It may just be...

00:33:48If so, then please send the Hemen to the one pinned below.

00:33:51No, oh so sad. But, you know, you also do the small things, look for the small

00:33:55Solutions you need. Here, this was mentioned recently. I have

00:33:58also built a PDF compressor somewhere just to compress things.

00:34:03So often you have little tasks you want to automate, where

00:34:07you need a small piece of software, you can always consider, can I

00:34:11not try this again? Or do I have any use cases at home,

00:34:15that I repeatedly do, tasks that I constantly have to sort,

00:34:21documents, invoices, or something else? You can play around a bit with code,

00:34:26see what works. Can I play around with Geminal to automate something,

00:34:30maybe to capture my emails, or something else. In the end, it's

00:34:34kind of art. I would say interacting with the computer. It's not always direct

00:34:38Programming in the sense that a piece of code is somewhere, which also includes the Plurio

00:34:42could be something else, but it can also be a workflow that he puts together for you.

00:34:46Just go for it and try things out, because I believe that's what we wanted to convey in this

00:34:50episode. The topic of AI isn't going away. It's fun to engage with it.

00:34:56That has its challenges, but don't let yourselves be intimidated into thinking that it somehow,

00:35:01if you do something there, the whole world comes together, but look, there's enough stuff where

00:35:06you can try things out, where you can test yourself.

00:35:08Mark is laughing again, that means he actually has something.

00:35:11I'll say already, see you next time.

00:35:12Wasn't that Marco for the dearest time?

00:35:13Whoever wrote a book, as the attacker, turned off the internet next to grandma or

00:35:17something like that, if I quoted that wrong, he could tell me, maybe

00:35:21Marco link our previous episode, I actually know him through his,

00:35:25well, I don't know, no, I have already dealt with the Kengerokroniken and such

00:35:28stuff.

00:35:29With that, we digress.

00:35:31I also heard his opinion on AI recently at some talk at the Chaos Computer Club,

00:35:35because now, let's say, it's difficult for freelance artists in general, but I understand that, of course.

00:35:41Jens, it was nice that you are here. We started the episode with the sentence,

00:35:45that we are recording a second episode today.

00:35:47We are going to do that now as well. Our guest will be here shortly.

00:35:50First, I need to quickly charge my battery.

00:35:51In this sense, wish you all a lovely evening, a nice day,

00:35:55a great week, we'll hear from each other soon, and feel free to leave us a comment, and see you soon, bye.

00:36:01See you soon, bye.

00:36:04Welcome to ThinkDifferent, ThinkAI, the podcast by Mark and Jens.

00:36:09Two technology-loving minds who not only talk about artificial intelligence but live it.

00:36:16Here you will find clear classifications, real practical insights, and a fresh look at what is possible.

00:36:23Understandable, critical, and always with a wink.

00:36:27AI for thought, for a chuckle, and above all, for discussion.