Krieg der Agenten

Transcript

00:00:00Welcome to ThinkDifferent, ThinkAI, the podcast from Mark and Jens.

00:00:07Two tech-loving minds who don't just talk about artificial intelligence, but live it.

00:00:14Here you'll find clear insights, real practical experiences, and a fresh look at what's possible.

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

00:00:24Food for thought, for a smile, and above all, for conversation.

00:00:29A warm welcome to Think The Front Think AI. Today again with Jens and Mark.

00:00:40Last time, Jens also mentioned our full names; if interested, you can

00:00:45listen to that again in the last episode. Today we have an old, new, always

00:00:52relevant topic. Recently, a colleague asked me in

00:00:57the office: Could you give a presentation about the crab with me?

00:01:02Apparently, it's a common topic both professionally and personally. And personally and

00:01:07crab, that's my segue. Jens, I heard you’ve been taking a crab for walks lately.

00:01:13How do you feel about it?

00:01:14Yes, it’s still not alive. I'm still at the very careful, secure setup,

00:01:23that this crab, we're talking about OpenClaw, also here...

00:01:26Yes, exactly, no animals have been harmed.

00:01:28Exactly, no animals have been harmed, but that this

00:01:30autonomous, living, in quotation marks, AI agent can work here

00:01:35in a safeguarding and ruling capacity.

00:01:37I got myself one of those Kailenecks.

00:01:38I'm currently setting it up with the help of Claude.

00:01:41I'm going through each individual step here of

00:01:44which things are allowed to be synced.

00:01:46with my Apple account and stuff, because I obviously don’t want to,

00:01:49have that little crab go wild in some way

00:01:51and infects a lot of things with my data. That’s why I’m currently setting up a very

00:01:56secure setup. Just a little sneak peek, maybe we can talk about it in

00:02:00other episodes. For example, I will also have some kind of memory then

00:02:05have installed locally, which saves the things I do, so that it is in

00:02:10principle perhaps no longer located with the large AI models on some

00:02:14servers in America or anywhere else in the world, but rather locally with me

00:02:17being processed. And then, in principle, only when I interact with other AIs,

00:02:22the small part that is needed is transmitted. But I no longer want perhaps that

00:02:27everything is swirling out there in the big net, especially when it comes to such a principle as my Memory-File,

00:02:34neuronal cells is something that I am gradually releasing to a kind of AI. But I'm not

00:02:40really discussing this installation much today,

00:02:45what I'm doing, we can certainly do that next time, Mark. But not so

00:02:49despite that, our topic today is the war of agents and Open Claw, it is right in the middle, and I would

00:02:57say has really brought the whole topic to 2026, that's quite something

00:03:02to take to the streets, honestly, but to bring it to the street and your endeavor reminds

00:03:06me of something that I think you even sent me, a report,

00:03:11in China, that there is a situation where people are basically lining up with their, probably not Macmeas,

00:03:19I believe those were MacBooks, standing in line and engineers setting up booths to help the

00:03:23people do local Open Claw installations. And if you look at the

00:03:28photo of what is happening on the street, after the

00:03:34Motto, people come here, bring their, their things

00:03:38past by. There are really a lot. It's quite like a happening, like a

00:03:43event. It's somehow astonishing. We only knew that from America before,

00:03:48in front of the iPhone stores, when people were camping out there. But now there's a principle,

00:03:52that happens in America and China. It's a bit crazy. We have it here again

00:03:58for the listener. I was just talking two minutes ago about wanting to be particularly

00:04:03safe doing my Orton Crawl season, and now in China, they're massively

00:04:09installing developers with older people like the crazy Orton Crawl right now,

00:04:17because awareness is growing that it's no longer just a model story,

00:04:23but the person who has an AI running 24 hours, seven days a week for them

00:04:30might be the one who will have the advantage, because their model

00:04:35can simply do things for them and not just be like you said in such a

00:04:39Prompt world or I need some N8N workflow, but I have something that

00:04:44can independently handle things super well for me. And actually

00:04:48it also makes real money sometimes. So making money is a good keyword.

00:04:52I just quickly checked again. There were over 1000 people

00:04:56there on March 6, kind of from 10 in the morning, and they were dealing with the topic of

00:05:03Open Cloud installation, and there are freelancers, uh freelancers, nice

00:05:08word.

00:05:09There are people making some extra money, let's say, for 500

00:05:13years, so roughly 70 US dollars, to carry out Open Cloud installations.

00:05:19The Chinese are indeed crazy, if you consider that after it

00:05:23first saw the light of day and then eventually blew up

00:05:28because it gained some notoriety, there are now over 140,000

00:05:32publicly visible Open Cloud agents in the world, and half of them

00:05:36are in China. So they are really going for it, they employ it in business processes,

00:05:42they integrate it, while here, we might also touch on this,

00:05:45you hinted at it in a previous episode, we need to see,

00:05:48When we consider that we first want to ban it, secure it, and integrate it,

00:05:56they are integrating it into customer support, hotlines, and care for the elderly, in that sense

00:06:01like, ask questions, you can get answers, that's quite a different approach.

00:06:05Definitely.

00:06:06And it's not just because we briefly have the topic of money, I also had

00:06:10an article in preparation this week, of course also this

00:06:13Open-Claw installation, there are also a ton of stories swirling around outside, which show

00:06:19of course how one can use an Open-Claw installation, a Taigiri agent,

00:06:29simply to really make money. So there are people who

00:06:32already now say, after a month I've made money because I automatically create social

00:06:37media posts, the product is automatically further developed, and it's monitored how

00:06:41customers respond, my customer service runs through this, my crab-installed

00:06:47or produced F is also monitored, how it can be further developed, how

00:06:51customer service runs, where people have to be careful, a lot of it is also

00:06:55sometimes just fake stories to go viral, but they’re already producing

00:06:59$7000 a month just by adding these things in the background

00:07:04that stop.

00:07:05So it's not just making money with the installation service,

00:07:08constantly interesting things outside, but actually also really making money with

00:07:14an agent that I somehow let loose into the free world and give a certain task

00:07:19and it takes off and generates real cash for you. It’s basically crazy,

00:07:25we had hinted at it last time, maybe we’ll touch on it today as well, about the whole

00:07:28skill topic, so how do I teach machines to do things, and the funny thing is

00:07:31it’s not just him who does it, the other big players, OpenAI and Claude and whatever they’re called,

00:07:36are doing it too, they can operate browsers

00:07:43or now the new models from OpenAI also tackle computer use.

00:07:49I operate keyboard and mouse on a computer, always check with screenshots where the keyboard

00:07:55mouse is currently positioned and then continue operating.

00:07:58This will change the way we work.

00:08:02Why am I telling this?

00:08:03We just talked about China, that was also funny.

00:08:05Over the weekend, I followed an influencer, I'm trying to follow one or two Chinese

00:08:09influencers, not because I understand the language, but I always check it through

00:08:12translation services, so I can follow it a bit, and he had something like a

00:08:16talked a bit in that direction, that the government there is currently also

00:08:20dealing with the question of how do we actually handle it when AI increasingly

00:08:24takes over jobs, yes, Robleck has recently mentioned it in some way.

00:08:27Study released, where can AI perform a job today, I would say,

00:08:32see the potential that an AI can perform a job. And how the governments

00:08:36are essentially dealing today with the question of what are we actually doing,

00:08:39when AI brings about a change for humans in their jobs, let's say,

00:08:46up to the question so that there are no unrests. There is

00:08:50a program in the Chinese government that deals with this stuff,

00:08:53where you think, okay, while you, let's say, are walking through the

00:08:57pedestrian zone and pushing people, they say, what does AI pass by,

00:09:00again? Health? Is the government concerned with the question, damn doctor, what are we

00:09:05actually doing to ensure that the sense for humans and their income and all that stuff,

00:09:10is not threatened if AI potentially takes away jobs? New things will certainly arise

00:09:17too. But I also found it an interesting story, again about the Chinese. It's worth

00:09:22taking another look over there. Exactly, but things are really happening. So even the

00:09:30big ones, whether it's Alibaba or Tencent or anything else, they are couching services

00:09:35to offer, so that you can install a Clouder on it. I saw pictures from the

00:09:43school, which I just wanted to mention briefly, where essentially a

00:09:49secure network is also offered so that everyone at the school can set up an Open Cloud installation

00:09:53in the school network. So that's how it is here.

00:09:58So we remember what happened during COVID and the people

00:10:06among our listeners who may have children of school age, how long

00:10:10it takes for such an emergency measure to really function or something else. And then you see

00:10:14one of the largest countries in the world taking up this

00:10:20topic at an incredible speed, and then you look at our German bubble and okay, nice bubble,

00:10:27so you rethink it a bit. Is this just normal hype and are they so smart that we want to miss this

00:10:35hype on purpose, or is this maybe again a bit too much?

00:10:40We've been talking about this topic for a long time now and we're also happy that we have German

00:10:47European security standards in many areas, that we are looking at what happens very,

00:10:51very closely, but what I'm missing a bit is this momentum.

00:10:57If we take a look at China right now, things are definitely moving

00:11:03faster there.

00:11:04Is that a competition?

00:11:05That's it.

00:11:06I'm just thinking, why not that?

00:11:07is this competitive pressure in China again, I don't know Chinese culture well enough.

00:11:12So it's not like I could look back on Chinese culture,

00:11:17let alone that I have ever been there.

00:11:19At this point, I can only call on anyone who wants to correct us, feel free.

00:11:22email in the comments.

00:11:24What I once picked up somewhere was that in China, for example, there are

00:11:30cities where, let's say, everyone works for the same store,

00:11:34so that they can optimize transport routes and paths for people.

00:11:40So they seem to be a bit different from what we might be

00:11:45used to here in our highly modern, highly technological German land.

00:11:50From that perspective, I can imagine it well, and I believe it promotes even more

00:11:54the topic of competition and you have to learn to be at the forefront, and it is

00:12:00supported when you are at the forefront and not just starting out.

00:12:04I once heard the saying, build a nest. If you do the same thing for too long,

00:12:09you start building a nest, and you have to change jobs accordingly.

00:12:12or the activity or something. At this point, please, as I said, feel free to leave comments

00:12:16here. We both know a person from the company who has been active in the Chinese market

00:12:20. When I meet them again, I can ask how it was? I believe

00:12:24that people there are simply socialized differently and are set up differently, and that perhaps they

00:12:29also seize opportunities more easily, because you

00:12:33must not forget one thing, regardless of whether an autonomous robot, for example

00:12:38can go upstairs or up a ladder or not, and that it now

00:12:42shows Kung Fu tricks at some annual festivals is one thing,

00:12:47but robotics is much more present in everyday life there than it is here, and I also once had an

00:12:54interview in some German documentary where someone roughly said

00:12:59in free citation, if the machine does it, then we'll just do something else.

00:13:04That's just the way things go, and if we were just a little bit, I want to say, that's always mean,

00:13:10when you generalize everyone, it sounds ignorant or something like that,

00:13:14but if we were to deal with the question here, it should all be okay and then think that

00:13:20no one has to be left behind at all. I can't help it,

00:13:27that AI is coming and I can't help it that robots are coming, and I'm not saying that you

00:13:31have to implement everything blindly, but this sense of exploration, this joy in something new.

00:13:39When I think about what you can actually do with the system,

00:13:46that you couldn't do before? Yes, because you, what is it I? I've always

00:13:51tried to write a book. That never worked. The publisher blew up

00:13:54because I somehow forgot my context window after page 120

00:13:58and then worked slightly differently in wording, and I always

00:14:02had stress, never to be annoyed at the work with editors when it was more than

00:14:07just a scientific article, like a book. In this case it's simple. Then you say,

00:14:10let's make sure, it's always in the informal you-form, finish, just run the text through

00:14:14in the you-form, then you can at least assume

00:14:18that you at least have these major mistakes off the table, without having, yes, since

00:14:22page four you introduced a term that you didn't explain, on page five

00:14:26you might have repeated the term and I don't know what else, so just that you

00:14:31help not to write that, but just to ensure that you submit a quality

00:14:34that is worthy of being published, not because you were too dumb earlier

00:14:39content-wise, but simply had too little time and leisure and I don't know what,

00:14:44to find the right words.

00:14:45And if you've ever tasted that sweet honey of AI and tried things out

00:14:51you have. I’m very curious, when your crab learns to walk, then

00:14:56to be honest, it’s like a good drug. I’ve never done drugs

00:15:00in my life. So, okay, correction, I do enjoy the occasional beer

00:15:03if you want to call that a drug. Or I like eating pretzel sticks, that’s an insider

00:15:08reference to the preparation of this episode. You don’t want to let go of that anymore. And before I

00:15:13give you the floor again, maybe one funny thing, when I

00:15:16set up my crab, I had already tried a few things, because

00:15:19honestly, I also hadn’t managed to get the Telekram board to work properly with my main.

00:15:23You go to the board fiber, register a new agent and that didn’t work out.

00:15:27Then I had another agent and yeah, so anyway my agent was named Gemini.

00:15:32In the end, that one worked and I connected it to my crab.

00:15:36And my crab then said, nice, I'm here, give me a name or are you trying to say,

00:15:41that you would rather go to the competition because he has interjected the name of the board.

00:15:47And then you first stand there and think, okay, this thing can’t really think in that sense and it doesn’t matter because it has no emotions, but you get so flabbergasted, which also presents a problem when you are chatting, I don’t know how it is for you, at this point a little peek, go to the dark side of my soul, when you're chatting and the other person doesn't understand you and asks something stupid back and then you write again and they ask something stupid back again, you write and they ask something stupid back again, eventually your letters turn into capital letters.

00:16:16Letters. When will you perhaps also get angry? And then it comes back, yes, yes, I understood,

00:16:22no problem at all and um um um. But you have to understand, I still can’t do that,

00:16:27wear it about 10 minutes later, it works immediately. Yes, so everything is a bit, it's about so much

00:16:32humanly, the unbelievability. It is human and it's wild and I think we can

00:16:38talk a bit deeper about it another time, also

00:16:40when we talk about the topic of perception, that is, how something was, like a perception

00:16:45actually of things. It is often reported that we, just as you describe, based on this

00:16:52almost exact

00:16:54Copy this

00:16:57Being like a human in the responses, it naturally becomes difficult for us in our perception

00:17:03to not personify AI in some way. But that's okay, I mean we live in a normal cultural environment, simply because we

00:17:11perceive things, have a certain perception of these things, and when we perceive it this way, then that's also okay, it's really not that bad.

00:17:18Whether it's always good that they do it this way, is one question, whether they can do anything about it, is another, because they are of course also trained on human data and thus mimic behaviors.

00:17:29So that's perhaps a special topic. Let's go on for a bit, because we want to talk a little about the war of the agents today, what's happening around it.

00:17:39We've now talked about Open Claw as a thing, mentioned it, and discussed it a bit.

00:17:45Let's touch on this topic of China and Europe, to summarize in my words, I would say

00:17:50in China we're currently seeing a festival around the topic of Open Claw and we are rather

00:17:55happy that there's somewhere a hackathon in a co-working space on this topic.

00:18:01So I think that's the difference that we can critically note for ourselves.

00:18:07They might have just missed the train overall there a bit.

00:18:14I think we need to pick up speed. What might be fun here

00:18:19is a recommendation for you if you want to surprise someone.

00:18:23You will likely deploy your crab in operation, you will probably use it with telecrap

00:18:27when it comes out of a trip and it sends you an error message.

00:18:32Save that and if someday someone writes to you via iMessage,

00:18:35WhatsApp or something, just take that error message and send it back as a response.

00:18:40Okay.

00:18:41If they then ask what’s going on, hopefully you have another error message

00:18:46from your crab that you can send afterwards.

00:18:47So I managed to get up to the third error message with this, sending big regards to Alex,

00:18:52yeah, so in the sense of, what’s going on, is it already, again this one

00:18:56AI in the background, are you actually that in the chat, because you can tell that too.

00:18:59The more human this chat feels. The more you might run the risk that if you think you're talking with your best buddy, there might not actually be a best buddy behind it. But well, we had mentioned that topic.

00:19:11I’m still glad that we both are still real. We also decided early on to keep our video running when we record this podcast, even if we’re not on-site.

00:19:21That could of course be changed, we’ve talked about that in subsequent episodes,

00:19:25but we still trust each other that we’re both recording this podcast in real life,

00:19:30our real thoughts and not AI-generated voices coming here, and this discussion

00:19:37is still being filled by real people made of flesh and blood.

00:19:40But say, let’s now, yes, let me tell you more.

00:19:43I was just about to say, I’m wondering how I can get back to the moderative role

00:19:46back to our next protagonists on the battlefield, so to speak?

00:19:51when we talk about the war of the agents. Yes, we can do that right away, because I still have some,

00:19:57I have this year also there, let it be read and let me have the summary given. The one or

00:20:04the other scientific paper also around the topic of AI. And there are also two things,

00:20:09that I wanted to briefly mention, which were interesting and that I remembered from last week.

00:20:13Ah, there was once a scientific study by MIT, I believe, as is often the case

00:20:23also lately.

00:20:24It's about the fact that when AI agents communicate with each other, it eventually leads to

00:20:35the bad actors among them, or rather those who perhaps want something negative,

00:20:42the AI agents, gradually contaminating the whole network of AI agents.

00:20:47spoil it.

00:20:48Yes, that's another topic, which means, how great, you know, last episode,

00:20:54Pentagon, yes, blah, blah, blah, and now you're bringing up The Dark Side.

00:21:00It is like this, what should I do when scientific research says the firm

00:21:04I can only repeat it here. And it follows right after the study that says,

00:21:09that military conflicts are rather suggested to be solved with nuclear by AI. It's funny too, isn't it?

00:21:14Yes, that's the joke. So we can talk about one and the other of course now but

00:21:20we can also land at the Journal Days, but also about the fact that the problem is,

00:21:24as soon as you have a bad actor in such a AI agent network, then the agents

00:21:30will eventually have a hard time making a real decision among themselves.

00:21:33Then they get confused by this one agent, when they

00:21:37discuss, so they can no longer agree because then

00:21:40they will no longer be able to start a consensus. So these are two studies that have me thinking this week

00:21:44whether they would no longer trust each other?

00:21:46Exactly, a little bit in that way. And that is of course, when we get into that,

00:21:50it is massive what research still needs to be done there.

00:21:54Because, as I said, we've already mentioned this a thousand times here,

00:21:57it is not deterministic. We are not talking about a rule-based zero and one system,

00:22:02that always reacts like it is, but we are talking about neural networks,

00:22:06that try as best as possible to copy our human brain cells and our thought processes

00:22:13to be able to respond to all possible answers in this world. And that

00:22:19they do not really work in a completely normal way, that is normal and that one

00:22:23and therefore also approaches it with purely typical, almost psychological research methods

00:22:31to better understand how such agentic networks function.

00:22:35And at the moment, we are talking about a few dozen, 15 agents who interact with each other,

00:22:41who then exhibit behaviors and become strange.

00:22:44Fast forward to what's currently happening in China, where millions, billions are behind it.

00:22:50with AI bots interacting with each other. I'm curious how much that can handle. So much

00:22:58for the transition. Politically, not quite correct. Agentic, sweet, sour, please. Okay, there

00:23:05we are back. What is that? Excuse me. Yes, it's not correct to say something like that.

00:23:10So, the topic of agentic, as you just mentioned, is of course

00:23:16also quite interesting at a certain point because you mentioned the crab and you said

00:23:22earlier that if a bad actor is involved, not because of a bad actor, but OpenClaw has

00:23:27now also added support for ACP, which is the Agenteecline Protocol, allowing the thing to basically

00:23:34be able to address coding agents like Cloudcode or Codex directly and not,

00:23:41because they're doing any wild hacks. Like, give me your ear-outtalks,

00:23:45so I can somehow use this thing, but they're actually talking with the application,

00:23:49so you can let such things be controlled.

00:23:52So you could ask your crab to work with Open Cloud, so Open Cloud is the crab, excuse me,

00:24:00with Cloud Code or Codex to build applications for you that can then be deployed right away and

00:24:05send you a link like, here, I'm done, you

00:24:09can download your dashboard, your website, your Electron app here as a zip file or

00:24:14as a website or whatever. I have made it available and that's

00:24:20already, yes, yes, yes, yes, yes, yes. So this is not a bad actor, as I said, but that's

00:24:24really intense, I think. How seamlessly everything integrates with each other and when

00:24:33you watch them as they, I will say, write together, yes. I

00:24:37found the funniest message recently in software development, where it wrote

00:24:41back to me, oh, that's a tough task. That requires precise manual work.

00:24:47I'll start now and you think, didn't you just write precise manual work?

00:24:51Which hands? Which hands are supposed to be those? But that's, yes, that's interesting.

00:24:57Yes, but they are not the only ones involved. That means, a topic with

00:25:03a few themes. I think, as you just briefly summarized, usability,

00:25:08user perspective again, that will also take a while. So even if you now

00:25:11install a cloud, you'll have this completely normal chat window that you know,

00:25:19which I believe has already arrived with what you also know from

00:25:23OpenAI, then you actually have, then you have Cloud Code and Cloud Code.

00:25:31It's just that first you have Clotcode and then Clotco-Effect.

00:25:36Yes, yes.

00:25:37And it’s like, you already have three tabs you can choose between, where I say,

00:25:42well, yes, my dear gentlemen’s choir, you have to deal with it first

00:25:46as a user, what is the most important for the respective situation.

00:25:51It is relatively well struck, but I would say, that is now also

00:25:56yes, not mass market compatible, what is happening right now.

00:25:59But it's also additionally complicated, yes? So I think, we have to add,

00:26:05I come from a generation that has seen the topic of terminal, yes, small

00:26:10black screens with a blinking cursor and you write your commands and the computer

00:26:14executes. Cloud Code used to be in the terminal. And it‘s still there and there

00:26:20it is also, I’d say, the most performant and best maintained, and you can

00:26:24work with slash commands and plugins and no idea what, and then they somehow started,

00:26:30packing that thing into the web UI or into their app UI, not web UI, into their app UI. That's then

00:26:36also Cloud Code, but somehow slower, because with the web interface in between it becomes somehow

00:26:42still difficult. Then everything was somehow a bit complicated for the average user.

00:26:47Motto: Cloud code is for developers, we still need something for the

00:26:50knowledge work and they somehow launched Co-Work. Then you see,

00:26:54ah, what skills I am using, what boards I am using, which files I am working with,

00:26:58where I have an interface open, you get a bit more displayed, you also get

00:27:02a bit more displayed, what is my plan and who am I currently working on. But then

00:27:07you stand there and train a skill. We had that in one of the last episodes,

00:27:11what is a skill, right? Motto: it’s a work instruction, how should an

00:27:15agent work in Markdown, that is a file format and now you have defined that, you have

00:27:20perhaps defined it with the chat, what it looks like, what you want it to be able to do. Okay, you save that

00:27:25and it is then also available to you in Co-Work. Great, you go to the code in the terminal,

00:27:30it's gone. Yeah, then, ah damn it, then I have to, ah in the terminal, I need to put that in the

00:27:36folder so that it can be found again. I think that's

00:27:41already confusing. When you're trudging through it, especially I showed someone something

00:27:46recently, then you find the skills under the menu item 'Skills', where you think, okay, I

00:27:52can maybe translate that somewhat, but then the settings disappear and are

00:27:55always somewhere else. So about my main point, I always say, static surfaces like

00:28:01we used to know, we won't have anymore. It's just a constant change. If

00:28:04the vibecoding or whatever decides that something looks nicer somewhere else,

00:28:09that's where it will go. But now, if someone is looking for an entry point and sees all the YouTube videos,

00:28:15that are out there on how to do it, one can definitely get a lot of joy

00:28:20and mood from it, because that's not at the Washington Post, where they somehow show

00:28:24have, following the motto, what are some trendy ideas and they are then programmed one after the

00:28:27other so that you can see how powerful it is. You have to

00:28:31experience this sweet honey at least once. But you have to

00:28:36engage with it a bit and maybe stick with it for about three minutes and also

00:28:40be prepared for the fact that it doesn't look great right away in terms of design, it's not totally pretty. You are

00:28:47in a very text-heavy world, which also relates to our graphic UI episode a bit

00:28:51here it goes. Definitely. Exactly. So I think there's a little bit, I believe that's the important part,

00:28:57that we, I mean when one wants to engage in this field back then.

00:29:04I would always try to see whether it's the Washington Post and whether one can

00:29:08program something after it.

00:29:09I would always look at myself, what are the things I do daily?

00:29:14So if you really want to engage with AI agents, I think it's,

00:29:20a smart idea to write down one, two, or three things a day that you

00:29:24keep doing repeatedly.

00:29:26Check the emails, check the calendar entries, check calories while eating or something

00:29:33Otherwise, each individual at home always knows in a rehmesen kind of way

00:29:37does.

00:29:38And to take that when he notices something, to take that and maybe just simply

00:29:45to discuss with an AI agent first before implementing it into an app ID or something else and then saying how we could now

00:29:50in an app ID or something else, and then to ask, how could we now

00:29:54for example, automate better?

00:29:56Because the A.I.s are actually quite good at that, they can help you,

00:30:00They can even help you with such questions...

00:30:02They find everything great.

00:30:03They first find everything great.

00:30:04But they can also help with the crucial questions in that direction, and where do you do

00:30:09that now the best?

00:30:10These things are really damn smart.

00:30:13We must never forget, they can partly solve scientific problems, mathematical tasks

00:30:19and blue ones, where people have already failed.

00:30:22That means, if I have something like that and we are talking about it, this is of course

00:30:27still not perfect in terms of usability in many places.

00:30:29But if you consider going into this AI-agent world and want to implement use cases for yourselves

00:30:34start small, start with topics that really help you, before

00:30:40you think, oh, I need to create the next big SaaS solution to replace Salesforce

00:30:44or when I become a multi-millionaire or something, but rather to say,

00:30:48hey, I really have everyday problems, what annoys me all the time, I always want to

00:30:52have, whether that's in sports, during training, or whatever you do, and build

00:30:57something there together with your AI and see what works and also ask,

00:31:02do I need this better in the cloud?

00:31:04Is it maybe a cloud desktop or a Plexi-computer application that I could use

00:31:10?

00:31:11What I always do, just as a little tip from me, is I also take another

00:31:14AI and ask her about the other one.

00:31:16You know, then I might take a January and ask her, or I sometimes ask myself

00:31:19a cloud or then again a chatchity and say, well, couldn't I actually

00:31:22implement that well with the cloud?

00:31:23And surprisingly, they are not framed like the AIs.

00:31:28The models are not such that they are already branded and somehow would not recommend

00:31:32other competitors.

00:31:34They are totally open and happy to make recommendations that

00:31:38it works well, or also to say, if it works maybe

00:31:42go if not.

00:31:43So just from today's example, with my Open-Claw-Fast installation,

00:31:47what I've done so far is that I have already

00:31:51installed my local Volama model with 1.4 gigabytes. That's running locally on my

00:31:57computer, so 1.4 gigabytes of parameters, but not such a huge model as you always

00:32:01find out there. But it is a small model that can run quickly. Claude

00:32:05recommended that to me. So it's a competing model. That's somehow a nice thing.

00:32:10So AIs are not somehow already spoiled, like maybe some four on websites

00:32:15are; they just do it. I say to myself then, when Siri at some point

00:32:21gives me tips about Android. Android is not the word that Apple would often put in

00:32:27their mouth at that point, at least in the good old days. But what? Anyway, the good old days are to be set

00:32:32aside. What comes up again and again at this point, as you said a bit

00:32:37earlier, is where the term Saas-pocalypse comes from, that now due to

00:32:43the fact that with skills and cloud code, you can very quickly build yourself

00:32:49an agent team. I can define skills in cloud code and also in co-work

00:32:54Skills is indeed the mentioned topic with a

00:32:58Work instruction and you can tell me work instruction is front-end designer,

00:33:02Backend, developer, front-end, developer, blah blah blah blah. And all the

00:33:06you can then combine and say, so take all these skills

00:33:09and build me, I don't know, some application that you know on the

00:33:12internet, yeah? So my wife is self-employed, she

00:33:16wanted a new invoicing system because the manufacturer she had all the time,

00:33:20has now somehow broken down.

00:33:22So it didn't break down now, it broke down a while ago and doesn't

00:33:25provide updates anymore.

00:33:26That's also completely terrible.

00:33:27And then we were faced with the issue, damn it, all the data is in this file format

00:33:31from this manufacturer and overall, we went ahead and

00:33:35engaged with Craft Agents, which is an interface that can utilize various models in the background.

00:33:41can be used.

00:33:42I used Opus from Autopik again.

00:33:43In that thing, I said, keep an eye out.

00:33:45Here are some examples, a few screenshots of the application, here are some files,

00:33:52that the application generates, and you can find this and that in the files, and then it started

00:33:56and built the application in seven iterations.

00:33:59And we went through two or three more rounds, because then he couldn’t import some files

00:34:03and the exports looked weird, but after the

00:34:07second evening, we had an application that my wife then ran on an updated

00:34:13I could also use a Mac, because, as I said, the original manufacturer went bankrupt, so they didn't update the

00:34:17thing, and that was sort of the last passion that ensured that

00:34:22you couldn't update the Mac.

00:34:24And that is such a, what should I say, small helper tool, it was software that was once

00:34:30sold on the market years ago.

00:34:31Whether it's still worth anything today would be another question.

00:34:33It used to exist, and now you're standing there with all these questions that companies

00:34:37sometimes ask themselves, Mac or Buy, yeah, so do I do it myself or do I buy it, those are

00:34:42going to shine in a whole new light because Make, it's not bad software that

00:34:49comes from all these coding agents, what they're putting together, yeah, it's, it's

00:34:54fascinating stuff and some things, I mean, there are things you wouldn't even think of,

00:34:59that you could ask the AI, even I, yeah, I stand here, I sit

00:35:03here with you very regularly in this podcast, dealing in some way with this stuff day in and day out,

00:35:07then I still found myself standing there one day when my wife

00:35:10said yeah, that's the problem and all, and can you help me download

00:35:14and separate it, split the audio and video, you name it, and then eventually it repeated

00:35:20itself, and I said at some point, so sweetheart, I must confess, we can automate this.

00:35:25Yeah, she takes part in an online training, and she'd love to listen to it in the car

00:35:31and listening in the car is always tricky with videos, because the car doesn't play

00:35:36videos, so she needs the audio track, and then she wants to have just the part from the audio track

00:35:39without the whole introduction, because in the beginning, there's always a lot of

00:35:44oh, it's nice that you're here and all that.

00:35:45And then we said, okay, that's no problem.

00:35:48Craft Agent, the tool of my choice at home right now, thrown in, said, you pass

00:35:53just check if that needs to be transcribed, at the beginning I’m not interested in all that,

00:35:57the rest that’s left, please make an audio track out of it and what did that

00:36:00thing do?

00:36:01It installed FFM-Pack, it installed Whisper from OpenAI, broke it down

00:36:06and then keeps the timecodes, splits the file into audio, converts it and sends

00:36:11it right away so that you have it in the application that syncs with the car

00:36:16and you stand there thinking, okay, well, hmm, that went fast. Yeah, for that,

00:36:22you then have an application with a surface, yes, and then the surface was ugly,

00:36:26just briefly, because I found that very funny. The surface

00:36:28was ugly and then I said to him, you just wait, I like Apple

00:36:32and Apple has the Human Interface Guidelines and that’s a website.

00:36:35I said, you here on the website there’s everything you need to know.

00:36:38Just look at the pages underneath.

00:36:39I don’t want it to be that ugly anymore.

00:36:41And yeah, then it was in essence that.

00:36:44Whether Apple would build it exactly like that is another question.

00:36:46But it was in accordance with the guidelines for good user-facing design from Apple.

00:36:52Again, tastes are different.

00:36:53But it looked a thousand times better just by saying,

00:36:57you have something there, just do it and leave me alone and I’ll come back.

00:37:01And that’s the awesome part, that’s this thing, right? I think that’s why we also have this, so this war of agents is maybe a bit too crazy expressed,

00:37:12But it is, of course, really the case that we say we are now in 2026 in the age of agents, and there's no need to discuss that anymore.

00:37:21And since we are in this age of agents, which means that we are not talking about a model, but about applications that can execute tools independently, that can independently acquire websites, that can look up information, as you just described, and then bring something to execution.

00:37:42What we could have imagined differently back in the day, we couldn't have imagined that it would run so independently.

00:37:47Back then, people were already used to switching prompts back and forth; there were maybe some who had multiple terminal windows open,

00:37:54running several RIs in parallel to review code and other things.

00:37:58And now we are talking about everything almost exploding independently, up to a perfection, even optical perfection, as you just described.

00:38:06What is also a bit, and this is actually the point I wanted to make, what still works on such a human abstraction level, you know, because there are still the software levels that we have, programming languages, just like we actually created them at the moment, he also talked about this in a recent episode, that the whole topic of architecture and development is completely being discussed right now, who knows if there are any programming languages at all.

00:38:36at some point in the sense that we know them today, actually still useful, because

00:38:40in fact, they only exist to better understand how people and humanity are.

00:38:46That humanity is better understood. So that's just the way it is, and I don't even want to

00:38:50think further about what else might happen in the future, because this level not at all

00:38:55more is needed for the AI. You know, when you say that there are emerging dynamics from the network

00:38:59principles of architecture, programming principles, development principles of their own,

00:39:07which people don't even need, you know? By the way, I want to mention one thing at this point

00:39:12just to get things off my chest. So, I am also aware that the whole topic of AI and

00:39:19vibecoding, however that is all called, is all very IT-heavy because the nerds

00:39:25have let the cat out of the bag. And we have also said that programming languages

00:39:29don't matter to the machine, yes, it can handle SAP abab just like Kotlin and Zwift on iOS and

00:39:34Android, just like Cshab and all the other programming languages out there. But for the

00:39:39AI, there is also no distinction between a programming language or any other

00:39:45things that are in text form. So, it doesn't matter whether it has a process description

00:39:50in hand or a database schema. That's irrelevant for it. And if you tell it,

00:39:57rewrite that. In IT, it's called refactoring, and I have no idea what it's called in other

00:40:03disciplines, as they are called. At the end of the day, the AI processes text and whether it writes a document as mentioned,

00:40:10or whether it writes a program. That is

00:40:15AI nonsense, you often mix it up, saying, well, look, this is IT. The

00:40:21IT can now deal with vibecoding, why is this not finished yet? That would be too short-sighted.

00:40:26When you were talking about all the models earlier, I wanted to say, we have

00:40:30currently on our show platform of the agents, still a relatively, well, an old player,

00:40:35that has been around for a long time, but with a new function not yet mentioned, Perplexity. Perplexity

00:40:40has now released perplexing the computer and perplexing the computer

00:40:44basically aims to break down the main task into various subtasks and then respectively

00:40:50start subagents that take care of these tasks, somehow deploying

00:40:5419 different AI models and the funny thing is they use models and

00:41:00certainly all of their competitors, yes, from Claude and Ob Mei and Gemini and

00:41:05Croc can be found in there somehow, and they use it for

00:41:10something where they say, okay, this model helps the most. So Opus46 from Claude is

00:41:16such that the central reasoning engine Gemini does deep research. If they want images,

00:41:21they use Nano-Banana. If they want video, they use VO3.1. If they want fast,

00:41:25balanced tasks, they take Croc. If they have tasks that

00:41:29require a very extra long context, they take something like GPT52. And that thing

00:41:33can run indefinitely until it gives you an answer. But forever is also relative, forever has

00:41:42a price, it only works if you have secured the most expensive contract with

00:41:46Peplexen. Yes, exactly. I did that, yes, from the side of the

00:41:54motto that there is something new, click on it and see if it works. It's cool, it is

00:41:58really cool, but compared to other models like for example

00:42:02Openclaw, I mean models and runtimes, it still has some teething problems.

00:42:08So you can also start tasks, like, check LinkedIn every morning

00:42:13to see what happened and give me a summary, and when you check the next morning, it has not

00:42:16done anything.

00:42:17Why?

00:42:18I closed the chat window when I defined that, and when you close,

00:42:21it shuts that down.

00:42:22And forgets it.

00:42:23You can tell it a topic to remember, it didn’t do that,

00:42:26it just said, thanks for the honor and goodbye.

00:42:28This will probably get better as well, but you can tell from that,

00:42:32Ockenclaw, or whatever it was called in the story, has hit an interesting point. People

00:42:38are increasingly entering a system where you're not saying, oh, do I have the best model available

00:42:45to me, but more leaning towards the question, I have a problem, will you solve it for me, you are

00:42:53the AI in the room, don’t ask me, just do it! Yes, yes, and I think that’s where the interesting topics

00:42:58come into play from a user experience perspective, that I say,

00:43:05I don’t even want to deal with which model it is.

00:43:11And then I also want to deal with, actually in the future,

00:43:16whether it’s cloud code doing it, whether it’s cloud desktop doing it, or the normal chatbots

00:43:22doing it at that moment. Yes, so that's still essentially a current

00:43:26abstraction layer that is being demanded of us, that we have to think ourselves into this complexity,

00:43:31to choose the right action. To be honest, I don’t know if that, like I said, so,

00:43:37Now the Chinese show that one can still engage with the topic.

00:43:42But from a visibility perspective for the large mainstream, of course, it's nonsense.

00:43:49I mean, Google also grew so big back then because it simplified the search so much,

00:43:54that it was just one field and not in lists where I previously had to weave through the

00:43:57categories, because of course, at the time Google was released,

00:44:02there were much better, let's say, phone books or categorization engines back then on the web,

00:44:09that actually delivered better results to the users. But what Google won back then,

00:44:14the truth simply was this topic. You don't have to think in drawers beforehand,

00:44:18you don't have to think in the category of sports to then get the current news,

00:44:23but simply just enter the thing into one search window,

00:44:27and you'll get the best results for it. And I believe this moment must be

00:44:31that can also happen. So that's where we are right now, I think it's also in its wave movement

00:44:35regarding the topic of AI, I think that's okay too because it has to do a bit with costs

00:44:40and resource conservation, the topic that I say doesn't have to be

00:44:43always the big model, it doesn't always have to be the programming model,

00:44:47to be the fastest model, that you consider a bit across different

00:44:50Entry points that already guide the user, I believe it will still be necessary in such Argentine

00:44:55systems for a while that I say, you absolutely have to

00:44:59be the really big model to answer simple questions, and it can't be

00:45:02a small model that is still somewhat human-controlled at the moment

00:45:07but honestly, even there, right? So quickly as the development is

00:45:10going, we as humans don't need to worry about it anymore, that

00:45:13we can already fix the AI, that's best avoided.

00:45:17I mean, what struck me the last time, I actually cleaned up digitally again

00:45:22at home and watch out, I'm pulling out an old piece of clothing, N after N, back then, workflow unit

00:45:29and now there was an email saying, oh, basically the payment didn't

00:45:35work, because I mean, not paying didn’t work, but here, I’m gone

00:45:39and whether I wanted to come back again, so to say and then you sign in there,

00:45:44Because, yeah come on, let me save the workflow.

00:45:47Who knows if you might want to use the workflows sometime, and I don't know,

00:45:49maybe he was still locked inside.

00:45:51And then you look at these workflows.

00:45:52And then it strikes you that between you stopped,

00:45:56having a paid subscription until returning back to basic and then you go back in

00:46:00and you notice, oh, in the notes, I mean, in the little boxes,

00:46:04in the workflow, where you basically define the sequence of work

00:46:08and where you say, at this point please calculate something, at this point please query something,

00:46:12at this point please use an AI model. And there was then basically, it would

00:46:17have run into errors because there were models that OpenAI has since canceled.

00:46:21And you think, what a pain it was to change a model in that system,

00:46:30because the sentence, I'll just go for the next bigger one, it might cost more, is true,

00:46:35but on the other hand, all models behave differently. That's

00:46:40the same with Autopik. If I say something like, do this and that and if I say Opus, do the

00:46:46same, then even with normal text tasks, one suddenly spits out texts where

00:46:51it apparently stands only AE and UE and then you say, what's going on? I have to tell it separately,

00:46:58no, remember, let the special characters be as they are. Opus, yes, show the sun, sure,

00:47:02like the sun, show me what do you want, why do you have to do that again? What

00:47:06is going on? Yeah, so that's pretty astonishing, how you still fall for that,

00:47:10But when I had this workflow open, I thought, those were times when I had

00:47:15fears, as if May Eye and Co. had changed the models and turned off old ones

00:47:21and I actually no longer have that fear, I took the workflows

00:47:25back then and threw them into my Craft Agent, enough Craft Agent is an open-source

00:47:30tool, can be learned for free, said this is a workflow, do

00:47:34something with it, please create an automation, I want to see this no more, yeah and

00:47:40then it goes through this N8n workflow and builds an application for it.

00:47:44How crazy is that? But okay.

00:47:47Yeah, that's a bit, I think, that's a bit, what is it called, the Kapafi,

00:47:53one of the big development geniuses of this world, also presented this week

00:47:59not so much an outdoor look again, because for the past week or two, the

00:48:03topic has come up that AI can also be wonderfully used to generate very small models.

00:48:07So building other AI models for specific use cases, which can simply be achieved with

00:48:15such a prompt idea in the direction of saying, build an AI model with as few parameters as possible,

00:48:21that solves a certain task very, very well and then you just let that thing run,

00:48:24and from the idea, it actually only does that you tweak a bit

00:48:29on the prompts for the agents, so they get better and then let

00:48:33everything test in the loop, see if the model works, check the weights,

00:48:38how big, how many parameters it has to make it better. And that thing explodes

00:48:42it just works. And that shows where it's going. You know, we're saying this looping.

00:48:46The human is almost no longer in Beloop, to be honest, because it

00:48:51doesn't even make sense. Because sometimes it's just too complex for us to understand.

00:48:56I mean, at this point, it's worth noting this is our personal opinion.

00:49:01Yes, of course, it's definitely not the employer's opinion, but that's what I'm saying.

00:49:06Yes, I also don't know how a plane actually flies, that was a fun side note, but as humans we also need to

00:49:12understand each individual system 100 percent in order to be able to use it, and I think that actually is the case now

00:49:19possibly someday at the nice field of software this moment is reached.

00:49:25Where we've been talking about, and we talked a lot about WordPro Tools last year, well, N8N,

00:49:33we both recently discussed the topic of orchestration layer as a visual layer over gaming interfaces,

00:49:41so that maybe you feel like you can still control these agents a bit,

00:49:46if we're honest, it's just more like, you know, grasping at straws.

00:49:54Oh, so that we somehow have it under control, you know, and this gives us the feeling

00:49:59that we can be in Beloop.

00:50:01I thought that was nice, I had...

00:50:02And not following anymore, it was, you know?

00:50:04So these abstractions are exciting, when we don't say, well, there is

00:50:07this little economic flow that somehow works, even on my computer there are

00:50:11some chips that somehow toggle back and forth, and on these chips,

00:50:15which toggle back and forth, we've built software on top for years

00:50:18and we haven't thought about how these chips are actually

00:50:21controlled, and these abstraction levels keep rising and we understand and

00:50:27we don't even need to fully understand what is happening on the atomic level, what on the cellular level,

00:50:32what actually happens, so that in principle something larger comes out of it. And I believe,

00:50:36that's what we will see this year. So if we now talk about the war of AI agents,

00:50:42it's actually similar to the war of models last year,

00:50:46where everyone still said that the best model is the decisive factor. Now it’s about,

00:50:50which agent is the best orchestrator. And I believe that is the path we will continue to see

00:50:56and also into the next year, that it actually takes us even further away

00:51:02from what is actually happening in the background and more towards the topic that we just need to

00:51:09think about problems and potentially shove these problems into this loop,

00:51:16which then says, okay, let me just solve this problem. And I will test it for as long as needed and try

00:51:21for as long as needed to find a solution for this problem, until I eventually come back with a satisfactory

00:51:26solution. So also a point where the circle closes again, if

00:51:31someone chooses something, how can AI help me, whether I build something or a

00:51:36skill helps me. For me, it has really proven effective, so that you can use a

00:51:43AI that you can utilize, the topic of shadow altei or something like that, we have also discussed that

00:51:47it is a bad idea, but that you build skills with which you can also discuss

00:51:52topics, things, and challenges you face that can help you move forward.

00:51:56Because you mentioned the orchestrators earlier, it's always so shocking.

00:52:00As I said, N8N has become outdated due to agents, and it feels like N8N, I say

00:52:06let's say the ... I mean, whether N8N continues to exist or not is a completely

00:52:10different question.

00:52:11Kids, now hold on, Agentic is the hot topic that's currently being driven through the village

00:52:15and it's working wonderfully. But when that started to bubble up for me,

00:52:20was heading towards Agentic, someone stood next to me and said, oh, we absolutely need to

00:52:25talk to the company ABC. They presented us with such an orchestration layer

00:52:29and by now I have this strange feeling that

00:52:34the big companies that exist seem to be positioning themselves with their services,

00:52:40whether it's consulting firms, software companies, or I don't know

00:52:44what.

00:52:45Even they are moving incredibly fast when there's another Outropik, a Perplexity, a Google,

00:52:51a Wehrhoch, always releasing something and their nice marketing slides about this is our

00:52:56big orchestral piece, it has hands everywhere and you know about cost control and arse

00:53:01the robbers and then a release comes and everyone looks at you like that

00:53:05at a light, when a car comes, yes, with big eyes. And you think, too bad. It could have been so nice, it could have been so nice.

00:53:13There was a straw. Straw, they clean the straw hats when they do that too.

00:53:19Yes, maybe a positive message for you out there at the radio wave receivers.

00:53:24To the radio wave receivers, man.

00:53:27Yes, I'm not that old yet. But that's actually ...

00:53:31Yes, indeed. If you sometimes feel like everything is moving too fast,

00:53:39I don't know if that's a comfort. It is so. But it is. The big companies are feeling it too,

00:53:44the experts out there are feeling it too. The speed is insane. Therefore, I believe it is important,

00:53:49as we said earlier, to always break it down a little for yourselves,

00:53:52where can AI help you out there and then just boldly try a little.

00:53:58Please read the material, ask the inhabitants of the AI how you might

00:54:01do it safely when you install or try something, but as so often, just try

00:54:07it out.

00:54:08I don’t think there’s anyone who has a Plan A.

00:54:10So, I can’t...

00:54:11So, Mark, both of us, as I said, do think a lot about these topics, I would

00:54:15not claim in any way that I make some kind of prediction,

00:54:20about what will happen this year, what a Plan A could still be this year.

00:54:24can be.

00:54:25I can really only have as many Plan Bs as possible that don’t hurt when you

00:54:31fail with them.

00:54:32You can quickly move on again, and that makes my designer heart

00:54:37chuckle a bit, because we are now in a situation where it’s really

00:54:41about better recognizing where the problems are and less about thinking about what a solution could

00:54:47be because solutions will always become easier.

00:54:51So we can focus much more on tackling many more problems that we see in the world and then addressing them faster.

00:55:00And that is of course a nice thing, because the solution, honestly, if such a perplexity, as you just described,

00:55:07can take hours, days, or weeks to produce a certain solution, then it’s really not expensive anymore.

00:55:14Expensive, you know, then it costs me a few tokens, but it does not cost me hours,

00:55:18weeks, workshops, or any other stories, rather it’s just after the

00:55:23problem statement and clear problem recognition, it’s just the next step,

00:55:30it’s just solution production and nonsense and solution validation and adjustments,

00:55:35touches, maybe changing problems and so on. It’s just a kind of acceleration curve,

00:55:40that we are currently moving into, which is totally cool.

00:55:42Now I would actually take it as a positive direction towards the end,

00:55:47even if we say here, what, along the lines of nobody can tell you where the journey is headed,

00:55:52I would completely agree with that. Everyone learns from this, and it applies that

00:55:58anyone who has an idea should share it with others, Sharing as Carrying, I believe this

00:56:02applies even more to AI than to everything we have experienced in our professional careers so far.

00:56:07And what you said, if AI really allows us to achieve things faster,

00:56:14then that might be beneficial for everyone, for one person it might be the birth certificate,

00:56:19okay, fair point. There, one might say, the societal benefit might be

00:56:23limited. Although it depends on the regime, where you live, that could be the case, or

00:56:27what?

00:56:28Yes, but that doesn't matter. If the solution costs little, in quotes. We

00:56:36all know the resource issues we have, but if we get there and

00:56:41many of the problems will also be solved by AI.

00:56:46That's what I was getting at, that's what I was getting at.

00:56:49Then it's simply, as I said, the birth certificate is a legitimate problem,

00:56:53that should be solved quickly, just like perhaps discovering a cure

00:56:58for all different types of cancer.

00:57:00That's just how it is, and that's the idea, like Samt-Kaiz.

00:57:03You stole my thunder.

00:57:05That's what I was getting at, I wanted to start with the birthday card,

00:57:09but then wanted to move on to the big problems of the world.

00:57:10Everyone does that.

00:57:11Of course, also where AI may not have the direct solution,

00:57:15but can help people who are able to find the solution.

00:57:18represents a tool that needs to be created.

00:57:20And I already have the opinion, I notice this in my environment,

00:57:23that you are engaging with other things, which becomes an interest,

00:57:26that people specialize in this topic, are knowledgeable about it,

00:57:29maybe on a different subject,

00:57:31perhaps the specialization is no longer desired in that position,

00:57:34but in other areas. On that front, there is a constant change. And as I was told earlier,

00:57:38when I started entering the professional world, you can't hear it enough,

00:57:42you learn throughout your life. And with AI, that is even more important. Jens, if you don't want to

00:57:49veer off into a completely new topic, I want to point out with the Pläger about

00:57:53the time, we are coming to the end of this episode. I want to give a little teaser

00:57:59bring. I want to bring the teaser that we have a lot of requests from

00:58:02guests who would like to come back to us. This means there will be

00:58:07guest episodes again in the next few episodes. Not just yet, probably the next one, but it's

00:58:10coming soon with guests. I'm excited about that. During the

00:58:14preparation, we even thought about our first live episode, when we want to

00:58:17do that. In the Cologne area, maybe we can announce it when we

00:58:21know that the city approves; we still need to find a date. With an audience,

00:58:24that's right, that's also for the audience. Live, we are already doing it,

00:58:27but then with an audience. Then comes the proof that we

00:58:30basically do not edit. Except for groans, breaths, long pauses of thought while people

00:58:37go to the bathroom and others make another coffee to continue. No, it’s coming

00:58:40not pre-fun, page. From Jens's side, thank you for being here. Thank you for

00:58:44listening. It has been a bit of a longer episode again. I hope,

00:58:49many tears have remained. Otherwise, the pro tip. You can

00:58:52also listen in stages. And in this sense, leave a like, a comment,

00:58:57Tell your friends and acquaintances that you learn a lot about the world of AI with SinkDifference, SinkAI, stay curious, and hopefully have an interest in stepping into this exciting future with us.

00:59:11Ciao!

00:59:12Ciao, ciao!

00:59:29There are clear classifications, real practical insights, and a fresh look at what is possible.

00:59:35Understandable, critical, and always with a wink.

00:59:39KDI for reflection, for a chuckle, and above all for discussion.