Wie KI unser Arbeitsleben verändert!

Transcript

00:00:00A disclaimer at the beginning. All statements from me and my guests are our personal

00:00:06opinions and do not automatically reflect the views of our employers.

00:00:10Welcome to Think Different, Think AI, the podcast by Mark and Jens. Two technology-loving

00:00:20minds who not only talk about artificial intelligence but live it. Here you'll find clear

00:00:25classifications, real practical insights, and a fresh perspective on what is possible.

00:00:30Understandable, critical, and always with a wink.

00:00:34AI to ponder, to chuckle at, and above all, to discuss.

00:00:44A warm welcome to a new episode of Think Different, Think AI.

00:00:48Today, we're out without Jens.

00:00:51He's celebrating, what? Carnival, Kölralaf, oh, I'm from Hesse, I have no idea.

00:00:57But now I'm sitting here alone, yet I've still brought support,

00:01:04namely not one guest, but two. One of the guests we've actually had here before.

00:01:11Klaus Rodewig is back with us today and he will be joined by Alexander Heusingfeld.

00:01:19And I would simply like to ask you to introduce yourselves briefly, who are you, what do you do, and then

00:01:25let's dive into the topic, everything related to the world of work in the context of AI.

00:01:31Yes, I am Mr. Klaus. I think I've actually been here two or three times and I know the market

00:01:39for a terribly long time. I already said that last time. We have a shared

00:01:43iOS background, we've given talks together and committed terrible apps.

00:01:47Giving talks together meant he was tinkering on stage while I had to come up with the

00:01:52script, but let's move on.

00:01:56That already falls into agentic work, right?

00:01:59It's great.

00:02:00So, in my profession, I deal with security and AI, as both are, in my opinion,

00:02:08inseparably connected, meaning by AI I mean the part AI, formats,

00:02:16AI, Edit Development.

00:02:17I would now only refer to it as Genetic AI or AI-Base-Developer, since one can no longer speak of help

00:02:26but rather, and we will probably talk about this today,

00:02:30the AI will simply take over significant parts of the development process.

00:02:36We also have a second guest in the round.

00:02:39Alex, would you like to say a few words about yourself as well?

00:02:41Yes, then once again thank you for the invitation, dear Mark.

00:02:46My name is Alex Heusingfeld. I am also a podcaster on the Conversations about Software Engineering Podcast.

00:02:53And I am Klaus's colleague at Vorwerk, previously also at the Elektro-Werken.

00:02:59I helped to build the Cookie-Doo, the recipe platform.

00:03:03And before that, I spent a lot of time in software architecture, consulting, software development.

00:03:09And we hope to discuss a bit today about how software development

00:03:14so has changed in the next months, or how it has also changed in the last months.

00:03:18Now it's time for a commercial break, because you mentioned the name of our work,

00:03:24the Vorwerk, which makes this beautiful Thermomix.

00:03:27And I am responsible for ensuring that the Thermomix is safe and that we will use AI with you in the

00:03:32future.

00:03:33And I hope you all have a Thermomix.

00:03:37But Mark, you wanted to say something.

00:03:39Yes, I was just thinking about whether I should, well actually I wanted to add a podcast by Alex to the show notes.

00:03:45Maybe we can also get a link to your shop, so that the corresponding products can be purchased.

00:03:51But let’s see what happens anyway.

00:03:53You already hinted at it, in the sense of how AI changes work life or the lives of those,

00:04:00who deal a lot with IT, looking especially at software developers.

00:04:05In preparation we threw a few nice headlines at each other,

00:04:11which we can take as a red thread through today’s potpourri of conversation.

00:04:17The first one that came up for our introduction.

00:04:20What, AI, is something changing?

00:04:22No, so, is it good, is that to me practically

00:04:27whether in software infection, whether in infection, wherever,

00:04:30wherever text is worked with, whatever.

00:04:33Isn’t this all just science fiction and demo stuff and doesn’t this all really work

00:04:38properly yet?

00:04:39Do you know that?

00:04:40Or is it, of course, in your situation, everything is always different, that's clear too.

00:04:44But now, hand on heart, do you know that or what does it look like?

00:04:48Yes, I have to eat my hat now, because that wasn't that long ago.

00:04:52I had a passionate argument about this particularly with Alex,

00:04:58about how we will work in the future, and the dispute arose over such simple things as which

00:05:04metric we actually want to attach to our software development in the future.

00:05:09This was at a time when I was actively working with GitHub Copilot and was already

00:05:14quite amazed at what was coming out.

00:05:16But I still couldn't imagine that this is basically a turning point that not only

00:05:24turns programming itself upside down, but the entire development cycle as well.

00:05:27And now I've spent the last few weeks very intensively working with Claude Code.

00:05:34By the way, I am totally thrilled with the integration in Xcode 26.3, just a side note.

00:05:42And that was an eye-opener moment, where I realized, okay, in the future we will no longer measure

00:05:47lines of code or as test metrics, but we will simply measure how many

00:05:51apps a person puts out in a week.

00:05:53And that’s, yes, I can understand that well.

00:05:57There are such things. Some things you just have to have seen or experienced to know what they impact.

00:06:05I mean, both of you are kind of, I would say, affected by my weekend escapades, not because I’m somehow partying with you, but because you sometimes have to endure it when I stumble upon something and then we share information in our little chat bubble.

00:06:27With information I mentioned, that wasn’t long ago, or Saturday night, when you both wrote to us,

00:06:35Wow, I’m so angry, I’m about to throw my Mac Mini against the wall.

00:06:39I wanted to switch Open-Claw, back then it was still called that, what was it called again?

00:06:42Open-Claw Bot to the local model and the thing completely broke.

00:06:47I’m going to go watch a movie now and after an hour you then wrote, oh, he just sent me a WhatsApp message,

00:06:53that I have actually called it a day, but he has restructured everything for me, asking where I was.

00:06:59Yes, I actually reported on that in the last episode here on the channel, and it was Telegram and not WhatsApp.

00:07:05Yes, so you are basically involved for better or worse, but it’s also insane how quickly everything is turning right now.

00:07:12So, if I had been sitting with an N8N half a year ago, I honestly wouldn't even try to roll the dice for N8N today.

00:07:21Yes, that's somehow already off the table for me, I don't know.

00:07:25The challenge I see is that I think we don't really bring our colleagues along with it,

00:07:30in between.

00:07:32So, I see it in myself that I'm looking, okay, do I have a test system on which I

00:07:41can test this, what are the use cases I want to test, and while

00:07:46other colleagues are just jumping in and actually they are also

00:07:52involved, then others have already landed with Craft Agents for example and say,

00:07:58but that is, well, that is much more sensible than Openclaw and all the installations,

00:08:04that I had from Openclaw, I don't need them anymore. And if you ask them about N8n,

00:08:07then it was like, wait, that was in 2025, right? So the speed at which we actually

00:08:16have usable tools and at which we replace them overwhelms many, and I'm back to

00:08:23the point Klaus just made. I have been trying for some time to introduce software architecture

00:08:30principles like Build for Replacement into my colleagues, into our

00:08:39development teams, because in classical software education you have things like

00:08:45don't repeat yourself and you have to find abstractions and you have to make things reusable,

00:08:52and getting this classical software education out of people is really hard.

00:09:01So this no, that's not your baby, that's Pets versus Kettle, that's Kettle,

00:09:06it’s not meant to be malicious, but it’s just a lot and it just goes through and

00:09:12I think we need, at times when we should change our diet,

00:09:16maybe a different image, but you know what I mean.

00:09:19So this image replacement.

00:09:21What's wrong with pizza?

00:09:23What?

00:09:24I don't understand that.

00:09:25What's wrong with pizza?

00:09:26I meant Airpads vs. Kettle and I don't know how you got to pizza with both, but

00:09:30it doesn't matter.

00:09:31Good.

00:09:32Pizza.

00:09:33But that's a whole different topic.

00:09:36Yes.

00:09:37I would like to throw in a provocative tea for the rest of the podcast.

00:09:41I predict the end of recyclability, I predict the end of software architecture,

00:09:49not the overarching architecture, but something like app architecture.

00:09:54And I also predict the end of formal languages.

00:09:56I found that quite funny. I actually had the other day with a few colleagues,

00:10:00who looked at me completely incredulously, like,

00:10:03but I'm now essentially an expert in a language, what he means,

00:10:07Let's take PRP, for example. Let's take the language that

00:10:11decides at runtime, am I a number or a text? Excuse me if it was so necessary

00:10:15it was. No, no, it was. And I'm there, yeah? Let's say you're in now, yeah? You know

00:10:20yes, you can recite that, you can read it, you’re the expert here. And then

00:10:26one can ask the question, do you really think that’s going to happen? That has never been the case.

00:10:32And the grace of my early birth came to my aid because back then

00:10:39I actually had something like assembly language at my fingertips and then eventually something

00:10:46weird like PRP and Turbo Pascal came along, as the entire Quad standard was called at the time. So

00:10:52we already had that, just with the difference that, I’d say, it took longer and

00:10:58secondly, the language then became more understandable for humans, so that humans didn’t

00:11:04have to, I mean, writing assembly is funny as a party gag today, but no one really wants to

00:11:10do that today, maybe for the Thermomix, I don’t know.

00:11:13But maybe no one wants to do that for normal software development now.

00:11:16I can see Alex is twitching, he probably wants to say something, but what I’m

00:11:19wanted to get out, is this, this look in the eyes of the people, because they then somehow

00:11:24realize, I don't want to talk about impacts now.

00:11:28towards, we don't have to make everything modular anymore, but also a good portion

00:11:35of disposable software. You built something, it's there, it works, and in half a year

00:11:40it might work differently. That's something where you also have to take the people

00:11:44along, because technology, let's be honest, can handle it.

00:11:47The crucial question for me is or rather, when I talk with

00:11:54colleagues or also with people in the network about AI. It quickly comes up

00:12:00at some point, I ask, well, but what is actually still my job? Why am I

00:12:06even here? And this awareness, and you just brought up the assembler example,

00:12:13I have a different anecdote about that, I found myself, not even last week, actually,

00:12:18where we were discussing it with a colleague.

00:12:22The disputed ones. No, that was something else.

00:12:24No, that was the other one.

00:12:26Where we talked about the fact that we said, well,

00:12:29metrics that we used to have as quality standards,

00:12:34and said, well, the quality of the metrics is important,

00:12:38they are no longer so relevant today.

00:12:41Because why were they important before?

00:12:43Well, because people had to manage to understand these metrics,

00:12:47to understand what it should actually do end to end? What is the part that I have here in mind,

00:12:53that I'm currently processing, keyword Context Window, and what is its contribution to the end

00:13:00to end? And just like it was back in assembler times, I can remember, when I started with high-level languages in the late 90s,

00:13:06I also had two colleagues who programmed in assembler

00:13:12and told me, ah, with high-level languages, that won’t catch on,

00:13:16that won’t prevail. We need efficient programs and nothing is

00:13:22as efficient as assembler. And that kind of efficiency is only relevant today in niches,

00:13:29so there are niches where you need it. I don't want to get into brake technology,

00:13:36but there are just things where you're already in the sub-millisecond range, where you say,

00:13:41no, it needs to be somehow very, very efficient so that the

00:13:47response can happen.

00:13:48And then Java came along.

00:13:50Yeah, well.

00:13:51That’s my point, do you believe, exactly.

00:13:54Exactly.

00:13:55And microservices in the brake pads, exactly, exactly.

00:13:58I understand, of course, if that's one thing, I can now somehow whip up software at home

00:14:04quickly, I sat down this morning

00:14:08to work on a piece of software that helps me, I don't even know why I have such hobbies,

00:14:14but recognizing Pumped Injections in the Guilds and stuff like that. So I put together a little piece

00:14:19of software with an agent, to just try out how well that works and

00:14:25you basically build your little helpers. For a business application, you probably

00:14:30would still have to go through two or three extra loops, that's still the wording, what

00:14:33to find out. I really want to start before I might roll this out to clients or internal processes

00:14:37But nevertheless, it is exactly heading in this direction and bringing people along.

00:14:42on a journey. How do I work with something like that? How do I handle it? Yes, I mean that the

00:14:49some say Vibecoding, others say Agentic Engineering. Depending on how you shape the role again,

00:14:54this is certainly a shift in the world of work. On one hand in the

00:15:00empowerment of people, but also on the other hand in the question, what is my job? What am I allowed

00:15:06to do in the future? When do I start looking again? Do I let go? Do I determine

00:15:12me? No idea, what? There will definitely be a lot of changes, right?

00:15:15Absolutely. I see nodding. The nice thing about an audio podcast, you have to confirm the nodding

00:15:20also audibly. I see nodding. And I think it's also hard for people

00:15:26really said to accompany this way, because everything is so fast-paced that

00:15:29one honestly has no idea. What will happen in three months again?

00:15:33the next thing that comes up that turns everything inside out, because I don't know how it is for you.

00:15:39With this Craft agent, the last time I dealt with it wasn't that long ago.

00:15:45Before that, there was Open-Claw, and before that, Cloudcode, and before that, I don't know what.

00:15:52And sometime before that was Gemini. It's all happening really fast.

00:15:57Yeah, you see, you see a problem that we've had since IT exists,

00:16:02that one always tends to rush in, I mean companies

00:16:09tend to rush to throw solutions at problems without having thought it through first.

00:16:14And I believe the question, the answer to the question, what role do I play as

00:16:21I can of course answer the individual ones in this process.

00:16:24But I think, to be a company, you would need to take a step back now

00:16:30and consider what agentic working means for us, no matter if I do this

00:16:35with NLN-orchestrate, with Open Claw, or with Tralala, it really doesn’t matter.

00:16:40The worst that can happen is that you just take old working methods

00:16:45and have incorporated AI, which we have actually always seen in the history of IT

00:16:50.

00:16:51Bad processes have just been digitized, and it was still just as bad afterward, and now it has

00:16:57just become digital and it's more expensive. They might run a bit faster,

00:17:01but it's not a solution. And yes, I think what you're touching on with the question,

00:17:05what is my personal task? You mentioned this earlier, I believe, Alex,

00:17:12that the ability to unlearn and to learn is becoming increasingly important. I still remember,

00:17:19when I graduated from high school, which is already a year or two ago. It was always said, yes, there was

00:17:26even the theme park Germany, which the then Chancellor called it, and it was said,

00:17:30we need to prepare for lifelong learning. Yes, and now it's also added to lifelong

00:17:36unlearning, which I believe is an even greater challenge, because something has

00:17:41you said earlier, Alex, that getting things out of our heads, which has many,

00:17:46psychological facets. So, you want to become personally obsolete. Perhaps you've built up knowledge over decades.

00:17:54And now you realize, oh damn, everything I’ve built up over decades, I mean, you have, we both have published a lot.

00:18:03I've written many books on Epic Infection. Now I look at code, and I don’t need to read a single book.

00:18:09to read, and a finished solution comes out.

00:18:11And what skills do I actually need for that?

00:18:13I don't think they are the kinds of skills that are currently being postulated,

00:18:18like, oh, someone has to read this source code and check everything,

00:18:23if it’s correct.

00:18:24That's a typical human self-aggrandizement.

00:18:28If the LLM, the code from the out-of-LLM field, is good, and if

00:18:33I then add another layer on top and have it checked, then I know that

00:18:36it's good, and I don’t need to look at it as a person afterwards.

00:18:39Especially since many of the tasks that you can do, like for example, in

00:18:43using a linter, can also be done by Cloudcode, from my perspective it’s relatively

00:18:50important that we look at what are today’s things that

00:18:55people still value, and many value, for example, their role designation, that

00:19:01they say, I am a developer and proudly so, exactly, but I find that,

00:19:06I find that difficult because it’s not my role designation that defines me, but

00:19:12the things I do and for which I feel a passion for you, so in that sense

00:19:18a role is always a temporary position I take, a responsibility that I

00:19:22assume and that can change tomorrow and yes, what’s important to me,

00:19:28is a, what do I actually contribute to the end result, what is the end result, can I

00:19:32I can, so I would like to have contributed, I want to contribute and on the

00:19:39other hand, who am I working with on this, does it bring me joy, does it fulfill me. That sounds

00:19:46maybe a bit sentimental, but that is immensely important to me. I know,

00:19:50that you mean that I get up in the morning and know I'm working with people who

00:19:54who A are just as eager for it and B are also competent in contributing to it

00:20:02parking.

00:20:03But I do believe that it often sounds like one knows better, one is

00:20:08part of this machinery oneself and also affected, involved, yes.

00:20:16Nevertheless, if I look back just briefly, when I started at MBW

00:20:21I couldn't form a coherent sentence in front of more than three people. I was more

00:20:24So, the basement kid developer. That eventually changed. Fortunately, I no longer develop

00:20:29not because AI is now at the door, but because I found something else with the

00:20:33professor. But still, there are people who, I

00:20:39would say the topic of software development, software architecture, the beauty, understandability

00:20:43of code, the cracking of nuts, which previously might have been approached with

00:20:48an eternity of studying Stack Overflow and nights full of beer and rum, coming up with some stuff

00:20:54that were totally cool. And now suddenly, I would say the

00:20:59colleague, the student, whatever, comes around the corner, has somehow written three cool POMs

00:21:05and maybe achieves a better result in perception or in the speed at which

00:21:11a goal is reached. It might not be the cleanest architecture behind it, but

00:21:15no one is really interested anymore. Yeah, right. I wouldn't be a programmer today either

00:21:20necessarily. Although the ability to see things in a larger context,

00:21:26remains unchanged. So, what I just said about the end of software architecture. You

00:21:35don't need to discuss whether you build an iOS app with Viper or MVV, Pralala, like

00:21:41you know all these, we used to debate them to death back then,

00:21:47that's all irrelevant. But a complete architecture, if I can also test it that way.

00:21:51But in the space, we're still discussing it, right? Yes, exactly. A complete architecture of

00:21:56embedded controllers, from the motor, from the heating in the Thermomix to the cloud via

00:22:02mobile app, you'll still need a bit of experience for that, although I

00:22:09would say, with the speed at which this AI is learning, we shouldn't rate ourselves too highly now.

00:22:14That too will architecturally be able to do something well at least in the not too distant

00:22:21future.

00:22:24But the point I wanted to make, maybe I got a bit sidetracked.

00:22:29in my language overflow. I remember when we recently discussed the topic of Native versus

00:22:36Cross-platform app development. What should I aim for? There were certainly, I

00:22:43would say, camps, which probably still exist, that favor one or the other

00:22:47and continue to educate themselves in those camps, whether it's Kotlin, whether it's

00:22:52Wilf, or I don’t know, achieving very good software results, and that is

00:22:57becoming increasingly irrelevant.

00:22:58Do you know those lost souls who think web views in native apps are a good idea

00:23:05?

00:23:06Eyeframes were also already a thing. Yes, okay.

00:23:10Yes, yes, but these are things that change, yes. So from that side

00:23:14someone was standing by me the other day and said, that's similar to the introduction of the steam engine, where I said, are you JAK?

00:23:19The steam engine took a long time back then before it was practically everywhere in mass.

00:23:25You have to set up the building, you have to set up the machines, the people go to the machines, yes.

00:23:29Then you have the whole issue of how to operate it, okay, electricity came into play at some point, but what I’m getting at is that it took time.

00:23:35And today, here we are, greetings to Peter, yes, who somehow threw over the Open Clore there,

00:23:41hits deploy, and how many weeks has it been now, four

00:23:45or five or six?

00:23:46No idea.

00:23:47Now he is basically Adelaide of Sam Altman at Open AI and just like that, yes, that

00:23:53would, that all goes so fast, it can be overwhelming.

00:23:58But coming back down to earth, this example with those

00:24:01webviews in the native apps.

00:24:05I think that’s a nice example of how AI answers such architectural

00:24:11questions.

00:24:12Previously or up until now, the argument has been, let’s go into the webview, there we are

00:24:17faster.

00:24:18We just build this website in, we don’t have to adjust anything native in the app,

00:24:21especially since the website changes.

00:24:23That doesn’t matter anymore today.

00:24:24If that thing changes, then I throw the new JSON structure in there or

00:24:29through Figma or whatever, and then it is simply implemented natively and

00:24:32within minutes, not within days.

00:24:34Depends on your process, Klaus. So you have implicitly, let’s put it this way, from

00:24:40I don’t know, I don’t believe it, I can’t imagine, into concrete examples.

00:24:43come, because you made it concrete. You specifically set up a

00:24:47process. And I think it is relatively important that we talk about such

00:24:52practical examples, because otherwise,

00:24:58it remains at the level of 'I don't know, I don't believe it,' but not for us.

00:25:02So, very concretely, if you have an app deployment process that has so many approval stages

00:25:11and manual steps that you need like three days before you submit something to

00:25:17Apple, then I don’t understand.

00:25:19How do I understand you?

00:25:21What kind of moderation is this?

00:25:25What don't you understand?

00:25:28What does my colleague mean by that?

00:25:30Who, I want to say, your hypothesis is, well, the AI can quickly build a new version of the app and submit it for review, and since that's a small change, this review will probably go through quickly.

00:25:44No, that is again a much too large specification of a special case.

00:25:50It's about, you have a moldy form that you have on the website or

00:25:54an online shop, where you then previously said, we'll just put it in the app

00:25:59because we don't want to rebuild it natively.

00:26:03That's work, whether it's in Zwift, Kotlin, or whatever, building it natively

00:26:07is just work.

00:26:08We'll let that happen now.

00:26:09It's not about whether you can change it every day, that's certainly

00:26:12also an aspect.

00:26:13But first of all to say, we're just taking the API instead of the website and

00:26:18letting the UI from there, we give them a screenshot of the website and say,

00:26:22let's rebuild that in Zwift, that's a no-brainer, works out of the box.

00:26:26Yes, but also the change, that's the point.

00:26:30Until now it was, oh no, we'll leave that to the backend and the backend delivers

00:26:34the website because separation of concerns and that is a concern for the backend team

00:26:39and the backend team and the data structure.

00:26:42And now you say, we would have an explicit tag, let's leave out the dubious expressions.

00:26:48No, you have an agent team and that is responsible end-to-end.

00:26:52Yes, okay.

00:26:53That is responsible end-to-end and it now simultaneously changes the code of the iOS app

00:26:58and in the backend and ensures that the stuff is deployed at the same time.

00:27:02And that changes thinking and working methods.

00:27:04Yes, so that works.

00:27:08If you think so, that's my point.

00:27:10Point. But if you mess up in your structures, saying, well, I have a backend team

00:27:15and if they're not solely responsible anymore, then they first have to coordinate with

00:27:18the frontend team and then they both have to separately set up their agents.

00:27:23No, no, no. That's the issue with when you have a not-so-good process,

00:27:28then a not-so-good result also comes out.

00:27:30But from the beginning you have vibe coding, vibe coding becoming popular, somehow you saw

00:27:35a nice quote where someone said, as long as AI can't replace the discussion

00:27:39with the product manager, that doesn't help me at all. I think that's going

00:27:45already in that direction. The pure coding at the end is just a small part of the entire

00:27:50Processes and you have to look at the whole process and put it to the test and not

00:27:55say, we are now using AI as qualified autocompletion. By the way, with Vibecoding, I still have

00:28:03to think about that 'never with toilet paper' idea, like, well, it’s just been thrown there,

00:28:09but then we would eventually want to get rid of it again because it starts to smell.

00:28:13I would like to put this main- so from my professional experience, what I would claim is

00:28:17that it applies to 80 percent of the software that has been written on

00:28:20this planet so far, invisible.

00:28:23I say again, picture before replace, if it starts to smell, maybe you should warm it up.

00:28:30I am considering how big Windows should be for us to be able to do that? But then again, the nerd in me came out. So, I would like to come back to the topic of end to end. So, what is super clear to me is that if you, if you stay only in your, in your IDE reality and say, well, the iOS development

00:29:00They are just on Xcode and then there are some who use IntelliJ

00:29:05and others are still on Emacs.

00:29:07Have you searched for your Emacs?

00:29:09You can do it that way, then you shouldn't be surprised that the AI somehow,

00:29:19well, also doesn't have that end to end understanding.

00:29:23But if you say, wait a second, agentic engineering is more than just the coding

00:29:29of my backend or my app, but it’s about end to end understanding. What do we actually want

00:29:36to achieve? What is the value or the problem of the customer? What is the problem

00:29:41of the customer that we want to address? What is the added value that we want to offer

00:29:44to the customer? And if you think through that together in a small team of subject matter experts,

00:29:50then you initially have the real full potential of today's AI agents on

00:29:56your side. And that is a topic that from my perspective has not yet reached everyone's minds.

00:30:02The one thing is, what can it already do in existing processes?

00:30:06The other thing is, if I'm ready to take a step back for a moment and say,

00:30:11wait a minute, what did we actually want to achieve with this team structure, with this organizational setup?

00:30:15What was our goal?

00:30:16Why did we choose it this way?

00:30:17I mean, what you mentioned earlier, like the modular development,

00:30:20and all the specifications were meant to enable distributed development across the

00:30:25You can best handle Klobos in large teams somehow, and something with backend and no idea what.

00:30:32Then you have imposed an organizational structure over it so that all the people have a home in personal development, in organizational collaboration.

00:30:40How do we work in a team?

00:30:43Then the whole Agile stuff came around, trying to shape and lead the teams.

00:30:48And now suddenly an agent comes around the corner and says, you might want to take a look, we might have a tool that is not only.

00:30:55I would say rapidly changing in the technology it provides and what it can do.

00:30:59It’s not just about the individual, perhaps needing to be a bit more of a generalist,

00:31:04and braver and looking at the big picture.

00:31:08But then there’s also the topic not only of the individual but also of the organization.

00:31:13Even within which one performs their work, no matter how big or small the companies are,

00:31:19it raises the question of how one needs to organize themselves to

00:31:24get the best possible outcome, so that Team A doesn't say, I have a really great agent,

00:31:29which is running on Copilot, and you haven't seen it, and it prints for you, you know, I print

00:31:33a Word document so that this goes here, you know, I print

00:31:38you an editable form again. That this doesn't happen. As I said,

00:31:45there’s the Fax-GPT, and I’m still mentally struggling with that, but…

00:31:49Are you there?

00:31:52No.

00:31:53Yes, I know.

00:31:54The experience of getting hands-on with the process sounds a bit sidelined right now.

00:31:59But then there are still one, three really big things that I’m hearing again now

00:32:04that sound strange, that we should really be hands-on with because in all three areas, it's a continuation.

00:32:10just like now with vinegar.

00:32:12Short disclaimer, the examples Mark brings are not intended for implementation.

00:32:19I will remind you at the next weekend chat, yes, but...

00:32:24But that’s bringing...

00:32:30No, but jokes aside. If we say it like this, in terms of, yes, you also have to...

00:32:36You have to consider that you might think outside the box and say, okay,

00:32:41what could we perhaps change in the organization?

00:32:44This is not a, this is not a call for,

00:32:49oh yes, I can't change anything in the organization.

00:32:51Then, then I am completely powerless,

00:32:53but then I can't decide anything anyway,

00:32:54then that doesn't help me, right?

00:32:56Rather, you go in that direction.

00:32:57I briefly thought you were heading in the direction,

00:32:59that colleagues of ours are listening and thinking,

00:33:02oh my God, what’s coming around the corner.

00:33:04I thought you were bringing the disclaimer.

00:33:06Yes, but of course, that’s...

00:33:09No, they know that already.

00:33:11So Klaus and I, the colleagues already know that.

00:33:15No, so what I was getting at is, when we say, agentic engineering is actually more

00:33:21than just programming software.

00:33:23Until two years ago, I was still managing operations, and just thinking about

00:33:29the possibilities that exist to replace existing thinking or existing processes

00:33:38in operations. What opportunities I have suddenly to support something like classical anomaly detections

00:33:46through an LLM or generally through agents that bring different aspects,

00:33:56different personas, and with these personas, for example, look at log files

00:34:01and perform log file analysis, it’s fantastic that suddenly you're pointed out to things

00:34:06and I will not tire of telling colleagues this, when you manage to

00:34:13write specifications for your software, to say what do I actually expect of the

00:34:18software. Just imagine, because then the agents in operations

00:34:23can say, hey, you’ve written down that the software should do this and that and that

00:34:27but it shouldn’t do this strange stuff here. Is that expected or is

00:34:32what I told you yesterday, you can't know, Alex Weiß was saying, because I told you,

00:34:38rambling about the ISMS, the Information Security Management System, according to ISO 7291, that we

00:34:47would like to implement to achieve compliance with the EU Cyber Resilience Act, so that we can also continue to sell

00:34:53nice thermobics in the future. I hope, assuming, that this ISMS,

00:34:59which is actually a very cumbersome, rigid, organizational, procedural framework,

00:35:05must function for us as agentic implementation, embedded in this, so overarching,

00:35:16the software development process embedded in this ISMS, but things like,

00:35:21I haven't thought this through completely, but if we imagine now,

00:35:24this ICMS consists essentially of a regulation and a KVP, so a control loop.

00:35:30And if I no longer sit people down and say, look, now check

00:35:36once every quarter if we have improved, but rather let agents do this continuously

00:35:41and with all these opportunities they have from the development cycle, from

00:35:44the operations, as you just mentioned, Alex, from architecture, what I always,

00:35:50you get such a comprehensive feedback loop for your regulation that you can't get

00:35:57by traditional means and from things like the requirements that this regulation imposes

00:36:03on development, so CQ-Coding, system maintenance, patch management, you can also

00:36:09implement all of that agentically and let it flow into your code, into your operations.

00:36:14This is our big project that is pending, to think through all this and

00:36:21implement it in a way that is also contemporary.

00:36:24I see a challenge with the topic of verification.

00:36:31Some also call this guardrails, that you say, if I use agents somewhere,

00:36:37then I want to check that they are also working in the direction we set

00:36:44and that they might not be, well, not like the good arms dealer, supplying both sides.

00:36:51Checking with you and handing something to the attacker. Yes, yes, sure.

00:36:56More or less. Yes, but I find the example with OpenClaw quite interesting

00:37:06in terms of architecture, you can, for example, deny it write access to its core

00:37:12files, so Soul, MD, Identity, you can prevent it from editing them.

00:37:19So filesystem access. And such hard limits are where you say, and you cannot

00:37:28manipulate that yourself. You must not rewrite your configuration and at the latest

00:37:33at the highest level, you want to, you want to somehow, I’ll say,

00:37:38have meta-review processes, quality assurance processes that continuously monitor your agents

00:37:44and see if they are adhering to the framework conditions we have set for them,

00:37:49have they manipulated the framework conditions somewhere, are they still contributing to the corporate goal,

00:37:55are they actually working right now or are they in such a deadlock where they keep confirming each other

00:37:59all the time and are actually just burning tokens. This meta-level, some call it

00:38:06the agent orchestration platform, some call it the developer management platform, whatever you prefer.

00:38:12I think this meta-level, where you say, you have a way to manage the agents,

00:38:20to check what is running with you. And not at all in the sense of, there is one instance

00:38:27to rule them all, but rather that you create, ultimately, an infrastructure component,

00:38:33where you can check that, from my perspective, is the challenge we are currently facing.

00:38:38And no, no one has coined the word entrapreneurial.

00:38:40Yes, exactly.

00:38:41If that were the only challenge we faced.

00:38:45What would you advise people on how to prepare for this, let's say,

00:38:50new future or how to embark on this journey?

00:38:55So I think maybe you should find something with which you want to start, I

00:39:00I would just say, well, if he has said something, the topic is basically already taken, and then you look forward to other information from the guests.

00:39:09Therefore, I propose that I want to basically build on what Klaus has said about iOS development and publishing documents, so with books and articles.

00:39:20I believe he thinks that his expertise makes him irreplaceable.

00:39:24makes, he has not understood the dynamics. The AI that is now putting us before. This means,

00:39:31I am an expert in a topic. Here please register the expert status. Is nothing that is sustainable

00:39:39as a statement for the masses. There will certainly be exceptions where people somehow due to

00:39:46stand out in a certain area of expertise. But I would rather emphasize that

00:39:52you try to use the stuff, either to get even better or to work together

00:40:00with AI to achieve multiplication effects. What else do you perhaps have that you

00:40:07can contribute? For me, I’m speaking now as a developer, for me as

00:40:14a senior developer, where I particularly mean the number of my years of life, of course also the

00:40:20number of years during which I could gather experience. For me, AI is a super boost. So it’s

00:40:25not like what I’m doing now… I think it was last weekend, when

00:40:30I had four cloud sessions open and was building four iOS apps in parallel. You can do that

00:40:37even without any technical knowledge, only at the first point where it doesn’t work, and that

00:40:42is inevitably Co-Signing with iOS, you hit a wall. Then you start, then that’s the

00:40:47project done. And with the experience I have in that area, I can actually go from

00:40:53specification to having the thing up on AppStoreConnect for beta testing,

00:41:00on TestFlight, without interruption. So I can utilize my expertise to leverage the advantages

00:41:07of AI, this enormous processing speed, almost like an extended production line.

00:41:14And I still indulge in the illusion that this will also be the case in the future. You don’t have to

00:41:22compete with AI. I don’t have to program the same thing anymore, but I understand

00:41:27the basic structure of how an iOS app is built. I fundamentally understand when to use Core

00:41:34Data or when to pack something into the SQLite database or the

00:41:39text data or however. Yes, AI is getting better at that, but some things, especially when you

00:41:46are building software for humans, are always hard to answer digitally. What it looks like in

00:41:55two years, I have no idea. At the moment, I still indulge in the illusion that I am quasi

00:42:00the orchestrator of the many agents running around here, that I have well under control. But I

00:42:07I also don't want to rule out that this knowledge could be completely replaced in two years.

00:42:13One shouldn't have to worry about it anymore.

00:42:15The question is, what will happen to the beautiful WWDC if we're only watching promotional videos,

00:42:23from Mr. Force One and the rest, letting the agents do the work.

00:42:27There are so many insiders involved today.

00:42:30It's unbelievable.

00:42:32now already again on the comments where it says either it was very nerdy or a lot

00:42:37of technical terms or no idea what it would actually be her force one you can google that it's

00:42:44this old let me google it for you unfortunately no longer exists but no idea chatty chatty

00:42:50chatty we call it some kind of chatty chatty is yes klippi was that yes the following

00:42:58Yes, from the co-pilot and with that I'm also done with my personal boot filling out.

00:43:02Alex, I'm trying to salvage this.

00:43:05You need to come up with something intelligent now.

00:43:08You know, that's not just your personal assessment, let's say quickly.

00:43:18I believe this is a generally known, how should I say, assessment of the value contribution

00:43:25of Microsoft's own AI.

00:43:28No idea what the colleagues did, it's a shame.

00:43:31Exactly, I have three points that I would like to share with the people.

00:43:36And the first one is something that I personally had to get face-to-face from dear Klaus.

00:43:41had thrown right in my face first.

00:43:44Don't assume, dude.

00:43:45So don't make any assumptions for the output.

00:43:47Roll up your sleeves and try out specific use cases.

00:43:50Don't be afraid, instead I said in the end, OK,

00:43:54then I'll take a separate computer now,

00:43:56because I'm scared my data will be stolen. So I'll take a

00:43:59separate one and think about, OK, what are the things that annoy me in everyday life?

00:44:06Oh yes, extracting these PDFs from some emails and then preparing

00:44:12for the tax declaration. So everyone has some use cases where you say,

00:44:16I always wanted to automate this. Roll up your sleeves,

00:44:21do it. Because nothing, nothing gives as much as doing that. The second is, when you do this,

00:44:29penetrate the patterns. Things like skills, plugins, validation loops are MCP. You can

00:44:38tell yourself, yes that's all from Anthropic, but yes and at the same time what we saw in

00:44:45OpenClaw, in terms of patterns, I would now say was not so

00:44:51well-known before, by now there are plenty of frameworks that build on it.

00:44:55Those are things, if you understand them, then you understand better what such a

00:45:01technology is actually useful for and what it is not for, by the way exactly the same for

00:45:05LLMs, whoever still thinks we need to train LLMs so that they are 100%

00:45:10deterministic, has not understood what kind of technology this is.

00:45:13And the third is without and control the data flow. So know who you give which data to when and where

00:45:23and where it's going. There are no excuses anymore to say, yes, but I thought,

00:45:30the data stays on my computer because I have the app installed on my computer.

00:45:33Sorry, no, that's ignorance and ignorance should not be paid for.

00:45:40but it becomes difficult when you say, do it and now I also think Tanja, you all have,

00:45:45I think you've talked to me about this before, that I asked how would you do that

00:45:48what to do?

00:45:49Once a month I have to pull my PDF invoices out of the emails and have to

00:45:55somehow collect my invoice from Microsoft and from the other providers

00:45:59compile them, compare them with the Contour statements, and send them to the consulting.

00:46:04Would you now also tell the Open Cloud, feel free to fetch my Contour statements

00:46:08?

00:46:09Yes, right. Are they present in my banking folders?

00:46:12No, I’m not crazy, but I can give the year a part CSV export from my banking app so that it can process it.

00:46:23But less. Yes, but maybe...

00:46:27No, it's not about black and white, it's about what... So just the fact that I...

00:46:35Each of us probably has some Apple subscriptions.

00:46:39We also have Android listeners.

00:46:41It might be that someone has Spotify on Android or something, right?

00:46:44So, I would do the podcast like this, Mark.

00:46:49That may be.

00:46:51What I want to say is, I always get these beautiful emails from Apple and I,

00:46:55well, I can no longer count them on two hands,

00:46:58how many discussions I had with tax office employees.

00:47:01Yes, then you should call Apple.

00:47:03No valid tax invoice.

00:47:05You could make an Openclaw, so you know about it, your learning network.

00:47:08Yes, but you could go there and say, okay, you might forward it to a second email inbox, then you can use Openclaw or...

00:47:15Yes, I just told this Craft-Agent, you know, I have here on the Mac, I have Mail, I have Calendar, I have Reminders.

00:47:23And then the thing started and built me an MCP server that runs on my computer, and

00:47:28I can now chat with an LLM of my choice, using my data. Is that a good idea?

00:47:34Yes, I at least think it was a good attempt. But as you said earlier,

00:47:40the data flowing in, that's of course a lot harder in times of AI,

00:47:46because you don't just have the data as dumb information, but

00:47:53such an AI system has. Its memory of you, not from you, but from the work,

00:47:58it might have done with you, can recognize connections much faster,

00:48:02than you could possibly do. From that side, yes, it's also more exciting, but try it out,

00:48:08try it out, try it out. So, when you described your case there with here,

00:48:11once a month I thought well, but that could be solved wonderfully with it.

00:48:16Yes, but you gave me a totally outdated recommendation. Use N8N,

00:48:21that’s also not right. Wait, I didn't say N8N today, my old self said that,

00:48:27my current self says, use Craft-Agent, so dude, have you ever seen how much

00:48:32Yes, I think that we actually, I really think, it always sounds so pathetically

00:48:40somehow.

00:48:41When you think you've seen and climbed a mountain because there's a new technology,

00:48:46a new kind of solution in this AI sector, and then you think, yes, you didn't

00:48:51climb it.

00:48:52Now it's continuing, now it's heading towards sunshine, I don't know, ski slope, something

00:48:57pretty and then suddenly a new mountain rises, because, oh, now there's a new

00:49:02technology and damn it, what I've occupied myself with all this time,

00:49:05that's also something we, I believe, have to learn and approach.

00:49:10They always say in IT that everything goes so fast, but at this

00:49:15speed, that's new for me.

00:49:17So that's new to IT folks, right?

00:49:19Yes, that's what I mean.

00:49:21Oh, Alex doesn't see it.

00:49:23Yes, yes, and at the same time, at the same time, no, honestly not, because I can’t tell you,

00:49:34when the last time was, when I last read a ten-page tutorial

00:49:39or even more.

00:49:40So, what I find fascinating, now for about two years, so basically

00:49:47since I've had, well, it seems longer. So since I've had the LGBT account, you

00:49:55can check the application possibilities with the AI. And that's

00:50:00often, the thing messes up once more with a dracen, but there are references,

00:50:05where you can look it up. And I find that an invaluable advantage, that you can

00:50:11troubleshoot with the AI. Yes, nothing is so, still the speed

00:50:16nonetheless always gets faster. And we didn't have that rhythm. There was once

00:50:21oh, now we invented Op-Pack and red programs. Now we invented the WBW.

00:50:26Now JavaScript is coming. Then blockchain came, but that was all so. And now it's a

00:50:35weekly pace or, let's say, a monthly pace. I find that quite crazy. So the

00:50:40demands on understanding are already enormous.

00:50:43They fit.

00:50:44Do you all remember this show from the 80s?

00:50:50Once Upon a Time... Man with that trailer.

00:50:53What is time?

00:50:54I just looked with tools, I'm tired of people.

00:50:57What is time?

00:50:58I could build a car from a can of matches, you can’t.

00:51:02Huh?

00:51:03That was the late 80s.

00:51:04I mean, the early 80s.

00:51:05Tools could threaten everyone, then there was an office item, from the

00:51:08side.

00:51:09I’m just going to be cheeky looking at the timer.

00:51:12I’m doing a closing statement because Klaus sent me such a nice picture in preparation.

00:51:18Klaus is already in Continent, where do you want the closing statement?

00:51:21Well, Klaus is sitting there; I can't see that in the picture.

00:51:25Yes, I have to make that explicit, personally I made it explicit.

00:51:30He sent a nice picture, along the lines of, who needs a script.

00:51:34Yes, we somewhat adhered to that.

00:51:38I thank you for being here.

00:51:40Alex, you still have to send me the link to your podcast,

00:51:43so I can add it in there.

00:51:44I would love to do that for you.

00:51:46I thank everyone who listened.

00:51:47Thank you for being here.

00:51:49If you liked it, give us a like.

00:51:51As I said, if a bot should have been listening,

00:51:53feel free to subscribe to the channel and spam all channels,

00:51:56so as many people as possible listen to this podcast.

00:52:00I thank you all.

00:52:01And if Jens is listening, everyone can...

00:52:04Wait, wait, wait.

00:52:05This isn’t how it’s going to go down.

00:52:07Exactly.

00:52:08No.

00:52:09It’s very philosophically interesting now because you’re saying for everyone who has

00:52:13listened to this, but they haven’t listened yet.

00:52:15I know.

00:52:16Yes, but they will have listened, isn’t that future perfect?

00:52:21That sounds just like at my home, where the kids aren’t having fun,

00:52:25but school camp for the final exam of the school year.

00:52:28I don’t want to go into that, yes.

00:52:30School education is another topic.

00:52:32I wish you a good time.

00:52:35Thank you for being here.

00:52:37Ciao. Bye.

00:52:40Thank you all. Ciao.

00:52:43Welcome to Think Different, Think AI,

00:52:46the podcast by Mark and Jens.

00:52:49Two technology-loving minds,

00:52:51who not only talk about artificial intelligence but live it.

00:52:55Here you will find clear classifications, real practical insights

00:52:59and a fresh perspective on what is possible.

00:53:02Understandable, critical, and always with a compelling eye.

00:53:06A.I. to ponder, to chuckle at, and above all, to discuss.