Vibe Consulting & Bonus

Transcript

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

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

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

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

00:00:24KD for thought, for a smile, and above all for discussion.

00:00:33Today is a very special episode.

00:00:35In the first part, Jens and I have a guest, Volker Brun, from Adesso.

00:00:41He will introduce himself shortly.

00:00:43And speaking of Adesso.

00:00:45I had the opportunity to attend the Adesso Digital Day 2026 in Düsseldorf

00:00:52and took the chance to record a live episode together with Stefan Kämpf.

00:00:59We talk about skills, about Agent Harness, and the question Meg Orbei.

00:01:04From that side, today is something special. You’re getting a double episode, a special episode.

00:01:12Have fun!

00:01:13A wonderful good day.

00:01:16I was about to say from the broadcasting station, but I realize that

00:01:20a bit of my remaining vacation is still hanging over me. Welcome to ThinkDiffin,

00:01:24ThinkAI and in good old tradition, Jens and I have today once again

00:01:29not ventured into an episode alone. We want to spend a

00:01:33bit of time discussing the topic from Vibe Coding to Vibe Consulting.

00:01:39How could one get invited there, and I’m really excited to have someone

00:01:44as a guest with whom I’ve had the pleasure of speaking a few times on the

00:01:49podcast. I didn't know how to get this better. With me, or with us today in

00:01:56our virtual studio is Volker Kuhn, and Volker, I think it's really nice

00:02:03that you are here. Thanks for coming. Yes, Mark, Jens, thank you very much. Have a nice evening as well.

00:02:10I’m also glad to be here and I’m looking forward to our conversation. The funny thing is, though, there's

00:02:15the ash at the Mainhaub. I sent an invitation link and it was to a completely different

00:02:19podcast where I was once a guest, so thank you, Volker, for taking it so professionally.

00:02:24And before we dive into the topic, Volker, I mean I know you, but our

00:02:29listeners may not. Would you like to say two or three sentences about who you are,

00:02:33What you do? I like doing that. Volker Grun, Dortmund computer scientist, a dual

00:02:39work life, in my heart job and in my main job, I am a university lecturer at the University

00:02:43of Duisburg Essen. I have a teaching position for software engineering, so I take care of

00:02:47the question and I am also interested in how software systems are developed and operated.

00:02:51I've been doing something similar for a long time, about 30 years at various universities and

00:02:57I co-founded a side activity, a side job at media.desus.edich nearly 30 years ago

00:03:02and currently, I am the chairman of the supervisory board there.

00:03:06Thank you.

00:03:09I think it's nice when we have such a side activity, I briefly considered, I have

00:03:13published a book on the side and I've already had quite enough to do, from that

00:03:17side, very nice.

00:03:18We want to talk a bit today about the topic of Vibe Coding and Vibe Consulting.

00:03:25Maybe we should start with Vibe Coding and Jens, you're here too, right?

00:03:29So not that I'm here now...

00:03:31I'm here too, I'm just happy right now, so I'm still waiting for my entry,

00:03:35because my heart is of course laughing at the subject of Dortmund, because I am honored

00:03:39Dortmund, and then I'm glad that I can talk about this introduction you just made

00:03:42which was to put Dortmund in the spotlight afterwards.

00:03:44Good city, a good university in that area, right? A lot is happening in Dortmund, isn't it?

00:03:50Accordingly, yes, a warm welcome to Onen.

00:03:53From my side regarding the podcast.

00:03:55Yes, Mark, we probably want to briefly touch on the topic of what Vibecoding actually is and such.

00:04:00That we might want to talk about this again before we dive into the deeper topics,

00:04:05on how to move from Vibecoding to Vibe Consulting.

00:04:09Because for some out there who might not know exactly what it is,

00:04:14it’s a bit about how I can make normal inquiries, which I can of course also make here in AI in the area of,

00:04:21I want to know something, I want to search for something or I want to research something,

00:04:25or I want the AI to perhaps book an appointment for me.

00:04:27Somewhere I can of course use Vibecoding, that’s why it involves whole computer programs, apps, software,

00:04:34the SaaS provider solution or build something else, so I can go through the whole potpourri

00:04:39of things I can do. That’s a bit what lies behind Vibe Coding,

00:04:43so it’s really about saying, I just prompt, I don’t even have to have

00:04:47programming experience in quotes, for the AI to build decent solutions for

00:04:52me that I’ve thought of. And that’s a bit hidden

00:04:55behind Vibe Coding. There’s already the first transition,

00:05:00so if I look at Vibecoding, Jens said what he understands by it and that's

00:05:06all so cool. I mean, I also know that people come up to you and say,

00:05:11well, make or buy, this is a completely new world to evaluate compared to before, because making is

00:05:18suddenly so much easier. But how do you see it, Volker? So when you say,

00:05:22okay, you write software with the help of, I write a prompt, Jens just said,

00:05:27something perfect comes out. Is that everything or is there perhaps a bit

00:05:31more to it? And how might this develop, if you think about it, okay,

00:05:37if everything is reduced to a word like I wrote a prompt for a

00:05:41great software solution?

00:05:42Only if such a prompt is ultimately a complete specification of the software system, then

00:05:48there is still no difference. And so I don't want to disturb the word first of all

00:05:50but promise behind it that software development will actually be significantly

00:05:56simplified. And I think there's something to it. I don't believe that through

00:06:02Wipe-Coding people can develop large pieces of software if they are not used to, used to,

00:06:11thinking in software structures and abstracting, structuring, generalizing somehow for

00:06:16a daily business, how they are heard, or so belongs to core competencies. Then I believe,

00:06:21I believe, I no longer believe that. I think that's still for some toy application. I believe that through

00:06:25Wipe-Coding many applications have been created so far that you could have bought in the app store for

00:06:291.9.19 a month and then spend the night with it and tokens

00:06:34to gild, let's say, still a high price and the transfer to say, also

00:06:39look at the capacity for the management of my birth, the gift list and that's why

00:06:43I can also build an ERP in this way.

00:06:47I was annoyed by that.

00:06:49Yeah, well.

00:06:51As I said.

00:06:53Yeah, it is, I felt a bit reminded of my youth. Long, long ago.

00:06:57Back then it was said, when the internet came up and websites.

00:07:02There was once the saying, I don't want to offend any profession,

00:07:06but the saying was back then, no matter if you are a baker or a butcher,

00:07:10today you are a perfect website writer,

00:07:12because if you learned a bit of website coding in continuing education,

00:07:16you immediately had your project and were born to a higher.

00:07:22And it feels like that right now,

00:07:27because if you're not able to say what do I need, how would I like it

00:07:30and see AI as an enabler, then you quickly end up with something,

00:07:36that might look pretty cool, but looking cool alone doesn't make it okay,

00:07:42With an ERP system, I wouldn't use the phrase look cool, even with the fewest products

00:07:46in my mouth, because crossing the bridge is probably difficult.

00:07:50But you present something, people perceive it visually, wow.

00:07:55And then it means in reverse, Wipe Coding has saved the world.

00:07:59I mean, you know it both from software development or also from other

00:08:03things, when you see the result and the result looks nice, then

00:08:07the associated work that you used to do with your hand work,

00:08:12maybe coding software, today maybe you get a bit of help from AI.

00:08:16It's not at all identical to what other people associate with it,

00:08:20because for them it still looks nice.

00:08:21They cannot imagine what effort is behind it.

00:08:25And now with AI it's even worse.

00:08:26Now with AI it means, you've coded e-Wipe.

00:08:29You put in a prompt, something comes out.

00:08:32I do it all myself now.

00:08:33Yeah, I'm the one-man show.

00:08:34Who needs, there come the changes.

00:08:36We still need all the consultants and all the developers.

00:08:38Move aside, I have a code subscription,

00:08:41Then I'll quickly do it myself, right?

00:08:44You're doing it like this now.

00:08:46Or Volker.

00:08:47Or are you actually doing Vibecoding yourself?

00:08:51I'm not doing anything with source code, that goes wrong.

00:08:55I was a bit scared the other day.

00:08:56My own boss said he would now do Vibecoding

00:08:59and I said, damn, here comes the next one around the corner,

00:09:02until I realized that as someone from management

00:09:05I also started to write some code, yeah the poor developers, yeah. Here comes the boss

00:09:09around the corner and says, always take a look at this, I put something really ridiculous there.

00:09:12Yeah, but that's not only the case with the developers, that's true for everything. This is now,

00:09:20I recently sat in a small group with a few agencies, they also discussed

00:09:24the topic, and now the client comes along and says, I already have

00:09:27the design, the designs at this moment. Now quickly build it, because this is the

00:09:31next chain that we're talking about, now they also quickly build

00:09:35essentially an application, because my Talktide also Vibecodes over the weekend or something

00:09:39and did something.

00:09:40No, can you quickly pay before maybe you also dump it on your Big.

00:09:44I somehow read that last year alone, Apple, I believe, the number

00:09:48of apps that were registered increased by about 600,000, that there was somehow

00:09:51an 85% increase or something like that in the apps that are registered.

00:09:54This shows, of course, that a lot of AI-generated junk

00:09:59is quickly uploaded, because the engagement numbers do not increase that much at the moment, as

00:10:03you can also see on the side.

00:10:05No, no, but with Wipecoding, I struggle with the term Wipecoding

00:10:09in the classical sense, it somehow hangs here, you rely on your instinct for

00:10:13the problem and completely abstract from the solution, that's why I find it a bit difficult.

00:10:17a bit difficult with the Wipecoding, but let's say I prompt software systems

00:10:22or do it with a narrow specification that somehow comes about, maybe

00:10:27then natural language even becomes a substitute for a specification language

00:10:32at least. And then it does have benefits. Then I don't need to specify the ugly details

00:10:37of a software system because it knows from

00:10:41Assistant from the start, how the interface should look and how

00:10:44data structures are generated from it. And maybe I just need to

00:10:47specify the difficult parts. Then probably as precisely as possible,

00:10:51but maybe not right in a formal language. And then we have

00:10:55of course a huge, then we would have already made a huge step. And I mean, there are

00:10:59the first software systems, we will see that, which we put into operation without

00:11:04a person having seen the source code. Not for all systems, of course, not for all,

00:11:09and also not for all types of systems, but only for certain types, let's say,

00:11:14interface-intensive, algorithmically sparse, loosely coupled with the rest of the application landscape,

00:11:20something like that. But if you can create them without anyone needing to know programming,

00:11:24then we have long levers in our hands to improve productivity.

00:11:28Yes, and above all it is also a tool for those who can program,

00:11:33because with AI, under certain circumstances, I mean, you know the sayings,

00:11:38things you aren't interested in, you might be able to automate them away faster,

00:11:42but on the other hand, AI can, whether in Pair Programming or because you may

00:11:47really formulate tasks more cleverly, also find solutions that you might

00:11:51not have come up with alone. When I'm asked, like, who benefits the

00:11:57most from this, I still have the funny opinion that juniors can

00:12:01actually learn the most, if they engage with it, because the leap

00:12:05to a higher qualification can be guided by AI, and that the

00:12:10seniors, of course, if they engage with it, I would say, can really put some horsepower

00:12:16on the road, in my personal opinion, I believe that

00:12:20both classifications of people can progress the fastest. But I’ll say, we wanted to do a

00:12:28derivation and I'm even more looking forward to that part. That was just the entry to

00:12:32inviting our guest. Namely, now everyone talks about Vibecoding and everyone always has this

00:12:38in mind, yes, yes, software. It's getting swept aside. That's why we said Vibecoding and such

00:12:43stuff. But there are also many other areas of activity, because for AI, at the end of the day,

00:12:49it doesn’t matter whether it writes source code or documentation or other types of files.

00:12:56And I remember how we got in touch a bit over the topic,

00:13:00where I once did some vibecoding, namely my little consulting idea, where I said

00:13:05I’m trying to equip various personas with different roles, with different

00:13:12capabilities, and the system is supposed to generate something on the fly.

00:13:17And I find it fascinating, this experiment works even today, and if

00:13:24I now consider writing software is one thing, consulting is another, then

00:13:31I think that’s a really big field when you say that stuff basically jumps from

00:13:36source code creation to concept creation, maybe consulting, maybe as a sparring partner.

00:13:43How do you see that?

00:13:45Indeed, I also believe that high players are at risk.

00:13:51On the other hand, I always like to make fun of the PowerPoint artists who then disappear after they’ve drawn up PowerPoint.

00:13:58So implementation is something else entirely, I believe.

00:14:01But even with the PowerPoint drawings.

00:14:03Yes, but let's just take the part up to the PowerPoint drawings.

00:14:06It’s not over when the PowerPoint drawing is done, but then in consulting, it usually starts here,

00:14:13to bring what has been considered, so the consultant, the consultant himself has thought

00:14:17to convey to the client, to bring into their mind so that there is überhaupt

00:14:22a realization of the impulse actually occurs. And I find it to be a process that is heavily dependent on human

00:14:27interaction. So today we already receive, well, we,

00:14:31anyone who wants can get pretty good PowerPoint presentations on all questions

00:14:38created, but that doesn't mean they've internalized what it means, or

00:14:43actually absorbed the knowledge, rather they just have a PowerPoint set in their mailbox

00:14:47or elsewhere and then deduce from that what they need to do, what it actually

00:14:51comes down to, what do I really believe, what is particularly important to me,

00:14:56this is typically the result of a consulting dialogue, and that's not going to change.

00:15:01So I think the PowerPoint senders will now need less time to create the PowerPoint

00:15:05slides, but afterwards the job remains the same.

00:15:08It would just be good because they would have painted it with the help of AI and would have truly understood

00:15:14and comprehended it.

00:15:15That is of course a problem too, right?

00:15:19And the other person gets it and says, but those are a lot of phones, I'll summarize those for myself

00:15:23and then he gets the eight farmers, he could just get the prompt

00:15:27sent to him for the PowerPoints, right?

00:15:30Funnier, nektotic, I wanted to formulate an email to a company based in overseas Cupertino

00:15:37and I used Co-Pilot, saying, here this and that and

00:15:43this and that, and then I told it, please make this great English, and I

00:15:46pressed send, and what did it send? The prompt, please make the following great English

00:15:51and not the English translation. This led to my counterpart sending me back

00:15:57an email saying thankfully AI is international and then it sent me

00:16:02the response I wanted, but I actually found that very funny.

00:16:06But what you're describing is actually not so dishonest for the software developer.

00:16:12And as for vibe-coding, I would say at first glance, that is

00:16:17all quite effortless; with vibe-coding, you need to think about guardrails

00:16:21and what else it can provide you and how it helps you and how it leads you to

00:16:25a professional result, especially with regard to professional use.

00:16:29And with PowerPoints, my consulting is more than PowerPoint, but that was now

00:16:34your illustrative example, it's actually quite funny, some use PowerPoint

00:16:39to create presentations, and others use PowerPoints to gather themselves again,

00:16:42and a lot gets lost in between. Nevertheless, this is

00:16:46a field of work. I mean, we've had other professions here in this podcast

00:16:50as guests, where one must and can also ask the question

00:16:55is. Where does democratization lead to companies being able to perhaps do things themselves now,

00:17:05without covering the entire, the entire breadth of consulting business,

00:17:09but let's say do small things themselves because they perhaps no longer

00:17:13need anyone for that or can perhaps professionalize themselves in their requests,

00:17:17because they can discuss with their AI what they would like to request, where it can

00:17:22perhaps help me and whether what it offers fits my needs.

00:17:26But on the other hand, I once said in an episode that

00:17:31software developers are like water striders. What we do is the best.

00:17:35That such a thing surely exists in consulting as well. That you tell people,

00:17:39hey, watch out, this thing is not just a softwash writer. It can basically,

00:17:44if you engage with it, achieve so much more, but you also have to

00:17:50engage with it. So the research part will be different. But the research part is initially for all

00:17:57companies, all people with the same problem indeed the same. And it can be beautifully

00:18:02condensed and provided. When it comes to contextualization, prioritization

00:18:06and alignment with the individual additional objectives of the consultant, then it changes. And there,

00:18:14I believe, the path to understanding with humans is more efficient. I followed up on

00:18:23LinkedIn, Jens will talk soon, but I need to get this out,

00:18:25sorry. Jens knows this from last time when we had Thomas

00:18:28Long visiting and I somehow realized at the end that I never let Jens

00:18:32get a word in. I apologized afterwards, after

00:18:35we were done recording, but I want to mention it. If I follow you a bit

00:18:38on LinkedIn, you also occasionally emphasize the uniqueness

00:18:44of domain knowledge, industry understanding. So all that, where I would say now, both

00:18:51personal relationships as well as personal knowledge about the context of the customer, the user,

00:18:56which plays a crucial role and that, I believe, is understood,

00:19:02can also be a bit of a protective wall when it comes to the question of where you need cold AI,

00:19:09that's always pleasing to you and where you really still have a domain where a human

00:19:16provides added value, both on their own and in combination with AI. And that with domain knowledge,

00:19:21I found really exciting. I'm convinced of that. Although I believe, yes, I am

00:19:26really long convinced that even developers who develop entirely without AI,

00:19:30can develop better if they have a rough understanding of what it’s about. That is, if they have domain knowledge

00:19:34ideally. However, I believe that a change is also required.

00:19:38Basically, a change is occurring in how software development or consulting or

00:19:41the necessity of domain knowledge is structurally similar.

00:19:46One no longer needs to know the details of any

00:19:49programming instructions or algorithms while programming. One does not need to know even the

00:19:54latest APIs in architecture, but one does need to provide a rough

00:19:58structure, and I believe it is similar with domain knowledge.

00:20:01One must understand the professional connections,

00:20:04but perhaps not the last details that arise from any

00:20:07regulations in banks or insurance, as these cannot be queried well, but

00:20:13the general context and its impact on the structuring of processes, business models

00:20:18or whatever is to be structured, this general derivation requires

00:20:24real domain knowledge in a more holistic context rather than the details that the individual

00:20:32can work based on well, working objectively.

00:20:35Yes, I believe that's okay, let me dive into that quickly because it's exciting.

00:20:40I am also at this point where I've been thinking all the time, we have recently

00:20:44noticed such a shift in public discourse.

00:20:48Even Sam Altman has suddenly started saying that not all jobs will disappear.

00:20:52Recently, they were saying that everything is somehow rationalizable.

00:20:56It was all about having an AI workflow very soon.

00:20:59I have been thinking about that a bit.

00:21:01And for me, the topic is such that I say we are currently, regardless of whether we

00:21:07are now in a consulting firm or a company working with a consulting firm,

00:21:12a larger company, always facing the difficulty that we not only have formulated SOPs

00:21:18so far, i.e., any Standard Operating Procedures that we just recite

00:21:22could run in full automation because it is still a people business

00:21:28in many places where you actually, as you describe, need a certain domain knowledge,

00:21:33which can sometimes mean that I know someone who knows someone else, and I

00:21:37know how they react. In situations, maybe even in larger corporations,

00:21:43they are comparable to societies, with family structures in some cases, where you

00:21:47might have to meet the right person at the right moment, then maybe also

00:21:51know how that person behaves in that moment, so it might be smarter to

00:21:55talk to them on Monday instead of Friday afternoon. So these are also parts of this domain knowledge that simply belong to it, where I say, if I look at a consulting firm now, then it is certainly an asset, especially if I am likely going to be working with a client for a longer time and of course then basically know their domain, that is, their professional domain well. So I believe that this can actually be an asset of a consulting firm when I build a long-term client relationship, because then I essentially understand these non

00:22:25verbalized or standardized written agreements, actually,

00:22:29I can and I and I believe that as long as that is the case,

00:22:34there will always be a need for this, because if we reduce ourselves to the pure

00:22:39output, as we just talked about

00:22:43Consulting firms are certainly not just PowerPoint scrubbers and PowerPoint producers,

00:22:49but it is often reduced to that, and I believe that is exactly the same problem for a

00:22:54programmer when they are reduced to the expertise of the

00:22:56programming language, and the designer is reduced to the,

00:23:00I can basically achieve great, I can do super

00:23:03lock-in flow designs or something like that, yes, or the fryer

00:23:07would let himself be reduced to, I make the best

00:23:09PowerPoints, yes, then it is, of course, I believe, a

00:23:12topic where you will simply have difficulties in the consulting business

00:23:14in the future if that

00:23:17accidentally happened in the past.

00:23:19It doesn't come from nowhere, so these topics are

00:23:23not there, so one doesn't think that many consulting firms are PowerPoint

00:23:26firms if it wasn't partly so, to be honest, or?

00:23:29Ideas in, yes, and now we present, and sometimes that comes from the interplay between

00:23:34the consultant and the client, and original ideas come out that are not

00:23:39equally already rolled out ten times for all other

00:23:43companies on board, and that original idea falls away

00:23:47when all the mush is already created. Then everywhere comes the

00:23:50same mush out. So this really original stuff will ultimately be the differentiating factor,

00:23:55then human-made content is also really premium because there are chances for

00:24:00originality to emerge. Sometimes it is also originally wrong, no question about it, but

00:24:05sometimes also originally good.

00:24:07How long is it, well, I really said, he recently had his similar

00:24:13discussion with, but with agencies, with digital agencies, where we talked

00:24:17about how that also changes the tourism industry. And I had that thought

00:24:22in the evening, is it really still, if we now power band, just hold on

00:24:27a power band firmly, now without the tool from Microsoft, it can dive deeper

00:24:29yes, of course, but of course it is...

00:24:33Dear host.

00:24:34No, host is something else. Host is now again a text document and

00:24:38word processing tool. PowerPoint itself is a presentation tool. It was actually

00:24:42made for that. But it has been used both internally in companies, as well

00:24:46as of course also with the consulting firms externally, which then explicitly prepares these PowerPoints well for various stakeholder groups.

00:24:54Of course it has been misused to not say, I present in front of a large audience, but I transport knowledge or I uncover topics.

00:25:03Is PowerPoint actually still the right choice for this kind of work in the future?

00:25:08Shouldn't we ensure through the consulting house that we think that through?

00:25:12I also have to advise my clients again in the direction that they have to get used to it, that it is of course nonsense,

00:25:18that I generate a PowerPoint with AI, which they then analyze with their AI and extract the content.

00:25:24But that one perhaps also sets up the maintenance differently and says, how does it look?

00:25:28What is my handover format? What is perhaps the workflow that I give to the clients in the future, the MD file or something like that?

00:25:34What are your thoughts on that, actually?

00:25:36My thoughts are quite old-fashioned. I believe for many ideas or for many

00:25:41instruments of knowledge transfer, the appropriate means of choice is the German language.

00:25:47So my finger also again to the language, but natural language.

00:25:51Ideally in full sentences and if we now select them all the way to

00:25:56full paragraphs, so not just bullet point lists, but entire sentences

00:26:00one after the other with conjunctions and indicative and subjunctive in between,

00:26:05to express the diversity of human life and the inner connections. What gets lost here if I do this in PowerPoint and omit the verbs.

00:26:14Everyone can think of the right verb, and then we have it approximately, and we leave it that way, but we all feel comfortable.

00:26:21Therefore, for the hard real result, I recommend in such places where precision is particularly important, and not just the right pastel color choice, just text.

00:26:31And then I follow in front of the sentence, then someone can gladly go afterwards and say, I understood the text and I want to bring it here, and I will manage, I do it in this way in the presentation and then I might have 5 keywords on a PowerPoint again.

00:26:49Then there are Wikini keywords and not such broken texts that are broken in such a way that you can still think everything, but must understand none of it.

00:26:58While you were telling this, a completely different thought came to me, Sorryans,

00:27:03I know it myself. You sit there and get a presentation. Presentations

00:27:10are then also a bit naturally about knowledge and content delivery, but are

00:27:15a bit of self-creation, is also a bit of the, yes, the slide shouldn't

00:27:21look like night, because otherwise no one wants to watch, and then you encounter

00:27:26different people in the art of presentation. Some read, let's say word for word and

00:27:32might turn around occasionally. Some are total speakers, storytellers, and you hang on

00:27:39their lips, yes, and think, it doesn't matter what they're showing back there. At the end of the day though

00:27:44still, there is a developed content in it. It caused work. It is also completely irrelevant,

00:27:51how much work it caused and it will be presented to you in the form of a presentation,

00:27:55so that you can have it in half an hour, an hour, however the meeting is scheduled,

00:28:00to have a status update and agree on follow-up activities or I don’t know what.

00:28:04Project results or order plan, it doesn't matter.

00:28:07So, while you were just speaking and said, a detailed text, that

00:28:14could indeed come back in fashion.

00:28:16Now, independent of whether this detailed text, I would say completely manually

00:28:21or with digital support was created, but you essentially have

00:28:24the chance thanks to AI. And that’s what I wanted to derive now. Sorry. You

00:28:29get a wall of text, but you as the recipient could, I mean Notebook, LM, Claude,

00:28:36Gemini, whatever it's called, also consume it in a way that's suited for you in preparation

00:28:41or for something that’s closest to you. That helps you understand things

00:28:47and especially, if you do this over time, then you also have

00:28:52context knowledge that you piece together, and that for me circles back to Jens.

00:28:56What is the format? Sorry, I don't want to be too nitpicky, yes, but this topic,

00:29:01I trim my guys and I only have guys in the company in my team. Write Markdown where it's

00:29:06applicable, because at the end of the day, it’s about the content of this text and not about, do I have

00:29:12a nice formatting and the color determines the context and I don't know what. And that’s

00:29:18basically my segue to the last part of our meeting, where is this

00:29:25going to develop? Now you said, okay, we have text, we’ve said, okay, the

00:29:29perception might change a bit, that people will realize again,

00:29:33okay, the combination with the machine together. What should we do now both on

00:29:39the customer side from your perspective, or from the perspective of a consulting house in five

00:29:46years, how do we need to prepare, so that we can continue this conversation in five years and

00:29:52say, it was a good idea with the texts, it somehow developed. I mean,

00:29:56I don't want to know from you what the next model looks like. But what do you think,

00:29:59what do the groups need to engage with, when we take a look, at whatever

00:30:04timeline, I said five, it doesn't matter, at whatever timeline?

00:30:07First, I must say that in the last 18 months, I have been dealing with all time-related

00:30:16Speculations have completely obscured it, and the confidence level will likely be affected

00:30:21just barely over five seconds to break it down now in shaping.

00:30:23But that's how it is for everyone!

00:30:24I didn't want to know that either, but you're not alone.

00:30:27Yes, I believe that in the contexts we've just discussed, the various

00:30:33in software development and in IT consultancy, let's take them sometimes

00:30:37also in consulting in general, that job profiles will not completely disappear, that

00:30:42they are affected to varying degrees by the quantity, the simple tasks

00:30:46are obviously more so. And for the sake of preparation, I currently have no

00:30:52better ideas than to say the guys and the girls who do the job for me,

00:30:57everything related to IT, they have to either know at least one domain really well or

00:31:03be exceptionally good. And everyone has to be able to assess exceptional possibilities.

00:31:10And with that, we end up with a Forward Deployment Engineer, where you say they have to

00:31:15somehow bring it all together to then leave in the context of effect with a client only

00:31:19after they have created value. And that is of course a different

00:31:23model than what we have today, where a few people are running ahead and trying,

00:31:28to do business analysis, crime engineering, then comes a lot of software development. That's not how it

00:31:33will be, but I think you also need architects and developers and others,

00:31:37who can prepare and transfer knowledge. Should I create the PowerPoint and those who create the text

00:31:41just maybe in a different mix. I've talked for a long time and can at

00:31:45the end say, other mixes allow you everything, almost even pasta. So he somehow calls it

00:31:50conversely, I don’t know. You articulated that nicely just now and

00:31:57that reminded me of an anecdote from my life. Now, it doesn't matter whether we are

00:32:03nerdy about IT or whatever, we are talking about AI enthusiasts. And depending on,

00:32:09where you speak, it's called vibe coding. Sometimes it’s defined differently.

00:32:13We haven’t done that yet; we definitely can’t stop or be downgraded to genetics discussions.

00:32:19I want to say one more thing, then you can talk about pensions.

00:32:23I’m getting to a point. You meet people, and depending on how many you interact with, you’re talking about a village community.

00:32:31So, it’s already familiar with Hitler-Persia, but it's about skills.

00:32:36That falls under employee development. We understand something different by that.

00:32:40I built a skill that does consulting, orders the consultants. No, no, no, it’s supposed to help you prepare things.

00:32:46So everyone is sitting in the room, everyone is nodding; you know how that goes,

00:32:51maybe it used to be with an abbreviation, yes, everyone nods, no one dares to contradict, because

00:32:56everyone has a word and an understanding in the room. And I think that's also a point,

00:33:00where you need to invest in the coming years, regardless of what timeframe that is,

00:33:04to shape the language we use not whether we speak German, English, or anything else, but

00:33:11that we choose terms that don’t leave anyone behind and don’t lead to misunderstandings or

00:33:19something confusing. Yes, with that. So from that perspective, perhaps a tip for consulting,

00:33:25Speak in the language of the customer.

00:33:27What do you say?

00:33:29Agents of Software 2.

00:33:31Definitely.

00:33:33And I think speaking the customers' language,

00:33:39is what you would have thought earlier,

00:33:41when you both talked about it,

00:33:43to what extent now, I’m always

00:33:45somewhat of a UX Hong denier,

00:33:47the IC there in the discussion,

00:33:49to what extent, in principle,

00:33:51the understanding of pure software

00:33:54pure software development, so if we also consider that, because it was just, in software development

00:33:59is only understood as the purely technical part, programming the code perfectly

00:34:04or even describing the requirements, so then it’s really

00:34:11actually relatively easy for AI to replace in this quality and it will only

00:34:15get much, much better. The models will improve, we will see incredible

00:34:18Seeing things, what all works when we look at the latest effects models

00:34:22and similar things, which we briefly observe and then are gone again.

00:34:25But that means, if we look into this, it’s naturally wise to consider a bit of what

00:34:31is also important for humans, or always important, that we can connect to others.

00:34:35We can dock on to them.

00:34:36And for that, of course, language is an important medium, where I say, if I again

00:34:39use jargon and basically make myself into a specialist, and then throw around AI terms

00:34:45and then also block others, I simply will not have them in my thoughts.

00:34:49And that’s why I believe it’s important, in principle, to establish a high connective ability to other domains,

00:34:54to other disciplines.

00:34:55I believe AI can help with that much more than you could in the past.

00:34:59Market place is the one related to the topic of junior, as I mentioned earlier.

00:35:02The junior can achieve a higher quality and faster,

00:35:05by improving in their software development.

00:35:07And I believe the junior can also improve in their software development,

00:35:11by understanding the other disciplines around them better thanks to AI

00:35:15and being able to connect better.

00:35:16And I think that is also an essential part of the consulting business in the future.

00:35:19to say how well I can essentially dock with my skills that I bring, yes, then

00:35:24also connect to the agentic AI system of my company and provide the right consulting service

00:35:29actually rather there in the moment. So for you, it will be much more exciting,

00:35:32to develop skills that essentially offer a high docking capability to probably

00:35:39increasingly complex IT systems than we can imagine today.

00:35:43Yes, yes.

00:35:44Because it's not that simple anymore, right? Yes. And I think that's a

00:35:49point where I say, when I think about it, I fully agree with you, Volker, being an oracle is

00:35:55harder today than ever before, compared to 10 or 15 years ago, because we simply have with the

00:35:59oracles, what can happen in 4 to 5 years. But I believe nowadays it is even

00:36:03more extreme, because this half-life of imagination, that's already

00:36:08somehow limited to half a year, because then such drastic things happen again.

00:36:11And I think that is an essential thing. One simply has to stay on the mainnet

00:36:16and engage with the topic, always with a critical eye, and I believe then

00:36:21one is always well-positioned, whether one is now on the consulting side

00:36:25or on the corporate side in this case or also as a small software shop somewhere.

00:36:30being out and about, as long as one remains open to all these innovations and looks around

00:36:33to the right and left and not just in one's own area of expertise, although it's important,

00:36:37to have domain knowledge, but even if one looks a bit beyond the plate,

00:36:40then I believe you prepare yourself well with AI capabilities, and you said,

00:36:44your people should do the same, that we are equipped with brilliant skills.

00:36:48Yes, I think that's what Mark and I also want to reiterate here,

00:36:51no matter which guests we talk to or when we discuss like this.

00:36:55One has to engage with the topics today.

00:36:56You can't avoid it anymore.

00:36:58It's not something that will go away, to be honest.

00:37:01Yeah.

00:37:03Well, then, Volker, thank you for being here.

00:37:09It was nice to see you on the other side,

00:37:12as I said, being our guest.

00:37:14We wish you a nice evening.

00:37:16If you enjoyed it, please leave a like, a thumbs up.

00:37:20All AIs are welcome to subscribe here.

00:37:22And otherwise, we also look forward to you tuning in next time, at Zinkdifferent, Zink AI.

00:37:28Thank you and goodbye.

00:37:30Thank you all.

00:37:31Thank you, Ms. Kerl.

00:37:32Goodbye.

00:37:34We are running the podcast Zinkdifferent, Zink AI, where we always talk about the fabulous world of AI.

00:37:41And we said, why not use this auditorium instead,

00:37:44auditorium, this beautiful hall, this environment should allow us to discuss this topic

00:37:49a bit.

00:37:50While I take my seat now, I would briefly ask you to say who you are, so that

00:37:54we can dive into it.

00:37:55Yes, a wonderful good afternoon.

00:37:56I am pleased to be here today in this podcast, Stefan Kemp, here at

00:38:00ADESSO for the past one and a half years as a consultant for mobile solutions and currently in the

00:38:06intersection between mobile and AI.

00:38:09Good.

00:38:10So let's get started with SyncDivision, SyncAI.

00:38:13Stefan, we have thought about a few things today,

00:38:17that we want to talk about with the audience.

00:38:19The first topic we have is skills.

00:38:22When I talk to people about skills,

00:38:25it seems that one or the other reacts a bit when the term comes up,

00:38:29when one starts to ramble.

00:38:30I don’t know if you know that feeling.

00:38:32You walk into a room, there are a few managers sitting there,

00:38:35manager, I don’t want to gender anyone here.

00:38:37And then you stand there and say,

00:38:39yes, let’s please discuss skills.

00:38:42And then you sometimes get people who have a slight twitch in their lip or raise an eyebrow.

00:38:47You wonder why, because people who might not be living in this AI babble, IT babble say, skills?

00:38:55Those are indeed topics for employee development.

00:38:57Why are they talking about skills here when IT is in the room?

00:39:00From your side Stefan, what are skills?

00:39:03Yes, those are skills. That's sort of the fundamental question.

00:39:07I might define skills a bit differently than most.

00:39:11So I use skills first of all, to simply create a PowerPoint presentation, for example.

00:39:16And it’s important to me that the PowerPoint looks exactly how an Adesso PowerPoint should look,

00:39:22that it sounds exactly how an Adesso PowerPoint should sound.

00:39:26And it’s not just enough to write a great prompt,

00:39:29it then has to be saved and reused consistently or also adjusted from time to time,

00:39:34but you need a bit more than that.

00:39:36I put everything into a Markdown file, I say this Markdown file,

00:39:39in that case. Look here, in this folder you'll find the Adesso templates. You’ll find

00:39:45the font, the colors, etc. And I compile all of that together and every time I

00:39:50tell my agent, oh look here, we have the topic ABC, please create for me a

00:39:54presentation, it will reference exactly that skill and this entirety,

00:39:59that you just described, applies also to Wordfall, etc. But this is how I define

00:40:05muscle. I must say, it goes a bit more in the direction of

00:40:10Stropic back then when this topic of skills emerged, it has defined this whole thing, and

00:40:16I am indeed very much oriented towards that. So when you ask me, what are skills,

00:40:20we come to a similar result. It's something like when you

00:40:24onboard a new employee in a specific field, then you have a kind of

00:40:28task description. What do you have to do? An AI is inherently

00:40:33perhaps very smart, but what good does it do me if I put the smartest

00:40:37colleague at a computer, he still doesn't know what to do. So you write

00:40:42it down in a list, I call mine an Orchestrator, you write what he has

00:40:46to do. Some things he really has to execute very precisely, you can use skills, if you

00:40:53take the definition from Entropic, as you just mentioned, with appropriate scripts

00:40:58for determinism, so for tests, deploying sequences of requirements, and where the AI can play to its strengths

00:41:06because it may say, okay, you have here basically a point, a very long point,

00:41:11an even longer point. The light deterministic can also be written down,

00:41:15in the Markdown files you mentioned, and all of this can then be compressed into

00:41:20a zip file and then you can give it, for example, I think at Google it's called something like

00:41:24Jamst and at Claude it’s called Skills, at Open AI it's also called Skills, but

00:41:30to make a long story short. At the end of the day, it's a Markdown file that can also be read very

00:41:37well by humans. If you're wondering what this is good for, did you use the example

00:41:42with the slides? Yes, we also have at our place, I've not made a single

00:41:46slide for months because I've built a Skill

00:41:52for myself and my colleagues, where I say, hey, about this topic, here are the materials,

00:41:57please make it as appropriate for the company, color, pen, imagery,

00:42:02layout of the slides, in the right format, depending on

00:42:08which market we are in contact with. And it does this automatically,

00:42:13but it does this not only for PowerPoints. It does this for consulting topics,

00:42:18does a SWOT analysis for me. He does this for requirements engineering. He does it for

00:42:23where square people might have written three numbers in a back block. Interviews

00:42:28everyone for so long until he has perfected the answer, basically grilled me,

00:42:34so that I give him the answers in a way he can elaborate in German English,

00:42:38whatever came up, yes. And all of this together is skills. Can you regularly

00:42:45assess these? So, I have, this is also a skill of mine. So I have a skill that

00:42:51evaluates my agent's sessions, namely regarding whether I took the results of my

00:42:59skills or have worked on them again. So the motto is, the PowerPoint skill says

00:43:06then for example, here are the 10 slides and I told him afterwards, but I

00:43:10would like 11, because I need a summary. Then he'll learn that over time and this

00:43:14self-training ensures that the skills with me then

00:43:18unhinged and automatically participates. If I no longer want that, I write at,

00:43:22I don't want that anymore. And then over time it will learn that too,

00:43:26so that it won't offer it to me anymore. That's skills. I have, that will still be added, a,

00:43:31I call it a Devil's Advocate Skill, which I always run through when I want,

00:43:36it to look at it really closely again, but from very different perspectives. So

00:43:41it critically goes over it and then learns that.

00:43:44Why is that important before you bring up the next point?

00:43:47If you conduct the experiment and you ask JetGPT,

00:43:50is it a good idea to offer my feces for sale in the city center.

00:43:54Could the AI system be inclined to wish you your color laff the IT for happiness.

00:44:00From that side, it's a skill that might critically examine the whole thing again.

00:44:03Not completely unimportant.

00:44:05Do you want to take the next topic?

00:44:06I would like to come back to that. Does that mean you've already tested this pomp?

00:44:11Yes.

00:44:12Can I have it? Okay, skills. We have three points on the agenda, so three levels.

00:44:17Skills, next is the Harnis. What I would be interested in, as I'm currently

00:44:20dealing very deeply with this topic in the principles area,

00:44:24who of you can make sense of the term Harnis?

00:44:29Yes, that's also the Adesso colleague. She talks about the hand topics, that's understood

00:44:32of course.

00:44:33I wonder if that was an Adesso colleague?

00:44:34Look, don't worry, we don't have handheld mics now, or else we'd be passing around, right?

00:44:38But now the question to you, because this is the point that I think few know.

00:44:44What is a Harnis, an agent Harnis?

00:44:47I always find it quite amusing, you know? When you also consider what it might

00:44:50mean in German. Most of you actually interact with an

00:44:54Harnis and an AI system. For some, it's the co-pilot interface, for others,

00:44:59it's this Codex app from OpenAI, Cloud Code, Co-Work app. Accordingly, the

00:45:06distribution of Harnis, if you ask me, is the environment in which a model is embedded.

00:45:11Because the model alone, some say, okay, the model alone with a system prompt,

00:45:17that's only half the battle. Depending on where you call up the model,

00:45:21you get completely different results. That’s also the reason why something like

00:45:25Cloud Code. When you look around in social media and so on, when people say,

00:45:30I have my own model at the back, then it acts on this own model, so those in the back

00:45:35do not get a job from the house, quite differently with Cloud Code,

00:45:40than if one hooks the model into something else. From that perspective, we at our company

00:45:46are building our own harness because we say the model is one thing and that you have

00:45:52tools. Then very quickly terms like A2A, MCP come into play, which aren’t even on

00:45:56our agenda today. From that side, we could also film for an hour. But a harness that gives me

00:46:01insight into how the model thinks? What do I provide to the model? How do I ensure that the results

00:46:08of the model are auditable? How can I later prove that when I execute a skill,

00:46:13that the result was indeed produced by this skill and not, let’s say, the user

00:46:19changing three things or using an outdated skill or using a completely different skill

00:46:23and saying why? I used that skill. The same goes for token costs.

00:46:28I recently saw a report again where someone wrote, look,

00:46:31creating slides for 7.95, that’s not expensive at all, that’s true. Creating slides

00:46:37for 7.95 is not expensive, because no matter which consulting house or company one works in

00:46:42for 7 euros, no decent slide has ever been produced. But if I now consider,

00:46:47what can I provide in advance through a harness so that the token consumption

00:46:53is reduced?

00:46:54Without expecting the user to specifically prompt and first use such

00:46:58strange tools like Caveman or whatever is circulating on social media, I have

00:47:03the opportunity with a harness to ensure that I manage this so-called aging loop

00:47:09before it starts, after it’s done, while it’s working, that I tell it which

00:47:14models, which depth, which efficiency. And for me, that’s the harness that brings it all together.

00:47:20Application unites and basically provides a home, even if home is not the translation.

00:47:26I remember when we were writing the book together and for the first idea

00:47:30we had of what we would put in the book, what this Hannes might look like and we simply

00:47:35couldn't find any visuals because the word view, I have to honestly admit, I want to say

00:47:41problematic, but difficult. Above all, difficult to visualize and

00:47:44conveys somehow a different content than the whole topic might mean.

00:47:49But if you're Hannes, I just said Hannes, so we sit down and build

00:47:53ourselves one, develop one. This quickly leads to the question,

00:47:58why are you building a Hannes? We heard it today, this morning in the

00:48:01keynote as well, if you are proud of building something yourself,

00:48:05certainly also with external support from colleagues. This whole question of

00:48:10make or buy. That's something where it feels like the cards are shuffled. If a consultant

00:48:17comes to my place tomorrow and says, let's implement this ourselves, then it's not that

00:48:22you, I mean, it's not a secret, right? You go back a bit

00:48:26into the past and say, building it yourself? Are you joking? Do you even know what that

00:48:32means? I mean, if you do something like that on-site, it takes months or even years,

00:48:37partly built into the products. Certainly, if you have software with

00:48:42Agenting-Baudia, Vibecoding, Vibe-Engineering, whatever those

00:48:45whole terms are. This is of course more than just

00:48:49I played a prompt. Even if the result looks really cool after a

00:48:52prompt, to put it that way, what does it mean?

00:48:55Always? Holfig means that it doesn't necessarily mean that it's

00:48:59security compliant data protection, that everyone has already thought about that and

00:49:03about everything, that when I fire the same prompt again

00:49:05I get the same result, yes from the side this make still has

00:49:11challenges where you can certainly approach it with either your own

00:49:16or external support so that you can say to yourself

00:49:20where you don't bring the problems upon yourself even faster because

00:49:24for example from my own experience I have teams that do agency

00:49:29software development and when they used to do two sprints every two

00:49:34weeks, the colleagues now do two sprints in one week, because the

00:49:40backlog curiously is empty on Monday evening, and accordingly you always have to

00:49:44look again at how to evaluate Melk or Bye, but it’s just not the way it was before.

00:49:49I would say that it still is like before, but we certainly notice that

00:49:54too.

00:49:55I notice that in customer conversations, the question comes up more frequently, can we

00:49:59do it ourselves. And my answer is, theoretically yes, do you have the people for it?

00:50:06Yes, especially, as I said, to make sure that the thing doesn’t break

00:50:10down internally. But, I would have another consulting area, depending on which company or

00:50:15what people's opinions are, based on a YouTube Clo-Code Consulting service, to actually

00:50:19build things yourself. This is a huge consulting area, because someone has to clean it up later

00:50:23what might have been created through vibe coding. Could happen. I

00:50:28also hear that we have hit an emotional nerve with the audience.

00:50:31Very nice, yes.

00:50:32From this side.

00:50:33Unfortunately, we only have limited time on this side.

00:50:37So I thank you.

00:50:38What has remained now?

00:50:39The things that have stuck with me, what I would also wish you'd take with you,

00:50:43is once again the matter, take skill seriously.

00:50:45Really build skills that match how you work,

00:50:49how you would like to work and can work in your environment.

00:50:53Please, please, please, ensure that you have a harness around it.

00:50:57Because I believe, we had just mentioned, in the end the model is almost irrelevant.

00:51:02If the harness is right, the model could be swapped out with a few small

00:51:06edges and corners here, where you have to tweak a bit.

00:51:09But if the harness is right, if so, then it works.

00:51:13I think that's for speaking, it could not have been finished better.

00:51:16Mark Zimmermann and Stefan Kämpf!

00:51:28from Mark and Jens. Two technology-loving minds who not only talk about artificial

00:51:33intelligence but live it. Here there are clear classifications, real practical insights

00:51:40and a fresh perspective on what is possible. Understandable, critical, and always with

00:51:45a wink. HI for thinking, for smiling, and above all for engaging in conversation.