Latest / Beyond Vibe Coding / #S2.21 From Prompts to Operating Model: What It Takes to Work With AI - Bastian Buch
Transcript
- Bastian: What we talk about here is not so different from what we spoke about 15, 20 years ago when Agile was a thing. Everyone was talking about project management and writing stories and doing DODs and having boards and so on. But what it was actually all about is enablement, a different attitude towards work, a different decision making. process and so on. And we have kind of the same thing today. Everyone talks about the next model. Is opus better than fable? Is gemini better than whatever? What kind of guardrails do we have? All of this is important. Don't get me wrong. But what I learned is that When we talk about AI transformation, it's actually the hard part of changing your attitudes and changing the kind of essence of how collaboration and work is being done in an organization. Sebastian: Welcome to another episode of our new season of Beyond Vibe Coding partnering with ImpalaSearch. André: The go to tech and executive search agency in Germany. Sebastian: In this podcast, we explore the transformational change in software engineering and knowledge work in general. I'm Sebastian Heidemeyer zu Erben, CTO at Ecosia. André: And I'm Andre, CTPO at Trusted Shops. Great to have you back. Today we are talking to Bastian Buch again. And we have talked with Bastian in April on the transformation, the AI transformation he ran at the Getaway Group. Sebastian: Exactly. And this is actually an update episode. He'll share his learnings and specifically what needs extra focus during the AI transformation. So we hope you enjoyed the episode as much as we did. Welcome to the show, Bastian. second time you're an old friend of this podcast. Last time we spoke in German, obviously, this time in English. And yeah, last time we already discussed a little bit of transformation that you've been starting in your current company, right? So there have definitely been some learnings. We'll hope we'll be able to discuss with you today. But before we jump into the main section, please repeat your intro again for our listeners, for those of you who have not heard the German episode as usual. Bastian: Yeah, hi. I'm super happy to be back. Let's see if we can make it also third time. Unless it's not Spanish or Polish or any other language, I might come back even André: You Bastian: with third time. yeah, a bit about me. I'm in tech since 2002. Actually, it's 1999 or so. That's where I started programming. And I moved to Berlin 2002 and worked in all the different tech companies here from smaller companies. ProMarketOnline, maybe some of you still know that. Then ImmoScout, Omeo, Zalando Delivery Hero. And recently, I've been in a project for two and a half years where I started as a fractional CTO, modernizing and then AI transforming a group of travel brands to today's worlds, let's say. But there's big news. I'm currently leaving the company. The project is over, the chapter is closed. We have some strategic changes in how the company will be set up in the future. Also the kind of focus of the company in terms of more marketing and so on and so on. And yeah, based on that, we together took the decision that this chapter is over. It needs a different profile from now and I moved to new things. And currently, I'm working with a few startups that I also can name right now because I'm doing more like due diligence, vendor DDs, and these kind of things. And I'm helping a founder in the space of Rectech, regulatory tech, in the field of banking, regulations, to build their first platform version right now. So there's a couple of things coming up. And I'm super happy to be back here in your show. Sebastian: Awesome. Yeah, that sounds exciting, indeed. And yeah, we you know the deal, right? So we have a first section in our podcast where we talk Bastian: Yeah. Sebastian: about the status quo, right? You like explained your very sophisticated setup already last time. So we don't want to repeat everything, just for those who haven't listened to it, you are an active user of OpenClaw from from the very early on, basically. after I don't know, it was out One or two weeks you already had been using it very extensively, have a pretty sophisticated second brain setup with several agents that are sorting through the business and updating it and building connections, right? So just just for for our listeners. and what would I be curious? So I I also do have that and and it automates part of my life, but I'm using it way, way less than you. So just out of curiosity, how how much time would you say you save an average on a on a daily basis due to this setup? Bastian: I don't know if it even saves me time. And I also think that's not the point in all of these, because it's not about automation and saving time. I also think this efficiency thing is not what we should talk about here. But it makes me wiser. It makes me make better decisions. And it gives me a lot of insights about myself, about my daily routines, and so on that make me just better in what I do. To give you one example, which I really, love, is these little gadgets where I can record meetings and so on, the plot. I think there a couple of those, pocket plot. I have a plot. And I use that every evening, almost every evening, to journal. I always wanted to journal, but writing is actually not a practice that I like. So what I'm doing right now is wherever I am in the car, in bed, wherever, I do a journal entry. And all of these things, will be picked up by my open claw setup together with what I have on my watch, which is the garment, you know, activities, sports, nutrition, and these kind of things with my calendar and all of this. And then it puts a narrative into my journal about myself. So I'm able every morning to read a chapter almost out of a book, which is my last name, about myself, which is extremely great to for reflection, right? Because it tells me about patterns that it observes and so on. And this is really great. In addition to that, it also builds me some dashboards for my sports routines, my habits, and these kind of things. And just looking into those makes me just better. Sebastian: So a lot of personal use and note to my open claw. Need to check out the Garmin integration. All right. Bastian: That's really good. You should do that. Sebastian: thanks a lot. Yeah, but now let let's go into the main section, the mead Bastian: Yeah. Sebastian: as we call it in in this podcast, right? So you last time as announced you we're talking about a fundamental transformation of the product and tech organization at the getaway group, right? And We could say you made some learnings along the way, right? And that's I think what you want to discuss and and share with our listeners today. And yeah, maybe let's start by you introducing us to the this the the main the headline of this learning and then we'll engage in a discussion about the details. Bastian: So it's actually something that we can speak about or talk about now for a couple of hours, I think. I was triggered when I listened to one of your recent episodes with Stefan, where it was, I think the point was something like transformation takes time. Right. And then I thought about, it a technical transformation? Or is it actually something different? And I, today, think, and that's a big learning, that's actually something different. What we talk about here is not so different from what we spoke about 15, 20 years ago when Agile was a thing. Everyone was talking about project management and writing stories and doing DODs and having boards and so on. But what it was actually all about is enablement, a different attitude towards work, a different decision making. process and so on. And we have kind of the same thing today. Everyone talks about the next model. Is opus better than fable? Is gemini better than whatever? What kind of guardrails do we have? All of this is important. Don't get me wrong. But what I learned is that When we talk about AI transformation, it's actually the hard part of changing your attitudes and changing the kind of essence of how collaboration and work is being done in an organization. When it comes to getaway, set without going into too much detail, but This is exactly what we saw, right? Because we cannot just go into engineering and product and say, we're building an agentic setup right now. And then everything is fine. First of all, we have not only people who like that and who want to work in a more holistic way and then kind of orchestrator, director way, but actually want to execute. That's their full right because that's how they started their jobs, right? So it actually takes time to transform that or to find different jobs for those people. But then there's also other departments and other functions and they by nature are different, right? So we have sales, we have finance, have HR, you have all these different people who are part of an organization and we have to take care of them. And just saying, hey, you all work with AI now, it's not what this is all about because... Most people use AI as a tool, right? It's like they go to a church EBT and ask for a text. get a get a text and then they think they're doing AI. It's completely different. And that is what people have to learn. Yeah. And that is my biggest learning that this is actually holding us back from AI. Or you could actually turn it around and say AI is an amplifier for how the culture in an organization is like. Sebastian: Yes. I mean that essentially as you mentioned already, we discussed fifteen years ago as well, right? The Potter change management. Bastian: Yeah, exactly. Sebastian: it it's essentially it's change management at its best and and even probably even more so than when we introduced Agile because it's changing the the core of the job of software engineers, right? Much much more than Agile, because Agile you you still were coding and and writing software, but you ha the the the collaboration with other engineers was was changing. This time it completely changes the job, right? So it's it's even even more difficult. But also Bastian: Yeah. Sebastian: for for other functions it's it's very very difficult yeah. André: I think this is one of the main differences. I liked the comparison. I like it a lot. I also can remember back in the days when we three were at Mobius and at least I can tell for myself, I was very much into the technical aspects, like how to do estimations, like comparing velocity, all that stuff. Over time, I learned, well, it's more about the people, the culture, the mindset. And so I think you absolutely have a point here, Bastian. But, I think there, AHA is maybe the wrong comparison. What we see with AI is it affects like the entire knowledge work, right? It's not just Bastian: Exactly. André: us on the technical side. It's actually everyone in the company and everyone is dealing differently. I would say the scale we see today or the challenge we see today is actually larger. Bastian: Absolutely. mean, there's a lack of comparisons, right? Because humanity never have seen something like that, right? And in Agile, André: Yeah, yeah, always. Yeah, yeah, true. Bastian: just like aspect of learning and aspect of reflection and all this was the new thing, right? And now you have something completely different, which is becoming part of an ecosystem where humans and machines are able to think. Right. this is, this is, yeah, just imagine someone would have been in your podcast 10 years ago, if you have, would have this podcast already and would come with a story of, you know, in 10 years we will have this and nobody would have believed us. Right. So this is really, really huge. And you, you absolutely name it, Andre. It's, it's not for tech. It's, for everyone. And, coming back to the biggest learnings, right. It's, it's also how, how an executive team works, how the board works. What do we expect from people? How do you actually drive decisions through the organization when everyone is becoming super fast, but you have this kind of need of becoming close or staying close to where decisions are made, but you are just a few people in the management team and the rest is like everyone else, right? If you have a hundred people, a hundred decision makers, but you are just, all of this makes it super difficult and we have not learned this as knowledge workers, as humanity. Sebastian: Yeah, and I think it has several dimensions. The first one is that every one of the knowledge workers needs to be able to engage or needs to learn how to engage with AI in order to become more productive and and with more productive. Also I'm pretty much referring to what you said before. I also do believe that efficiency is one part of it. So the the knowledge gathering is much more efficient. The the the ability to make informed decisions based on more data collated and collected in in a very quick way. also in software engineering, it like just producing code that actually works is like a lot faster, right? But it's the the even bigger lever is what's possible right now, right? So the the amount of projects that I've started in the last nine months in in areas that I wouldn't have thought about before, be it GIS be it audio processing whatever or image processing for for that matter is is just staggering right so that is that is one part but the other part is actually the next step is then how to do it how to still collaborate right because you now have have your I don't know assistant or your your the the the partners that you're working with the AI agents basically but how how do you do this then with other A other people's AI agents and other people still. So this is completely unsolved for me. So in software engineering teams but also in other knowledge areas. But anyway, it's a Bastian: Mm-hmm. Sebastian: it's a very philosophical point, I guess. Bastian: I probably it's not that philosophical in the end. So one realization that I came to is that we probably have to accept that not everyone is equal in terms of how they see work and what kind of talents and skills they bring to the table. Right. And saying that everyone should use AI is probably not the right thing. Because then you say everyone has their own agent and everyone opens cloud code and everyone has to somehow think about. The difficulties of it, giving context, setting goals, setting guardrails. When you have sales teams, cannot build these, for example, hooks to be secure and so on. So there's limitations to it. But then you could realize in this, could say, it's our saying that we are able to build and work with AI. It's our obligation to build a system. that makes everyone else, humans and the agents and juniors who come to the table after university, able to work in that environment. And this is actually, I think, what we should do. We should really have something like, there were these enterprise architects back in the days, 15, 20 years back. And probably you need really... André: Yeah, but they still exist. Bastian: I know. I tried to. Sorry, sorry for that. André: Just kidding, sorry for interrupting. Yeah, yeah. Bastian: But now you need a group of people who are not tech people, actually tech, but they also understand the human side of things and the AI side of things to build a software, sorry, an operating model, an operating system for an organization where agentic intelligence. human intelligence can work together in processes that don't overwhelm especially the humans but also the agents. André: Yeah, I think we may have now discovered a new role. I'm absolutely with you. You can't throw the complexity of building something on the entire organization. think it's overwhelming people. think people will, they like, I thought if you never had the chance to build because complexity was so high because you have no clue about software engineering and now you can just transform with a prompt or can build something with a prompt. think that is also. Fascinating, right? So that's why people also like to do this, but you absolutely have a point. You cannot throw the complexity in like in the production environment on the entire organization. You may need something like AI operating engineers or something like that, right? So people building this operating Bastian: Yeah. André: model, this operating system where, which others basically just can use, but like maybe also really do not feel the heaviness of doing it right. So providing the right prompt, providing the right context and all that stuff here. Sebastian: Which is I would say also exactly what's happening right right. we in engineering we're talking about harnesses because the harness is essentially the system that provides the right context at the right time, right? last time André: Yeah, good point. Sebastian: discussed with Rodrigue from Delivery Hero about their harness, but Bastian: Mm-mm. Sebastian: on the on the general knowledge work, they're like the Frontier Labs are utilizing forward deployed engineers, right? And actually big companies are doing the same thing. They have these internally deployed AI engineers that that are yeah taking care of exactly that. So I think it is happening. Still the fact is that it's there's no best practice yet out, right? How to how to do this. Bastian: There's no best practice. Sebastian: So we're we're all trying to find this out as we go, as we speak basically. Yeah. Bastian: And we should share and speak about it, even if we know that tomorrow it will be solved differently or that the kind of pace of things is just insane and it will be hard to... follow through with a lot of these ideas anyways. But maybe as one example, what I just did was the startup that I mentioned before. the founder, obviously, any founder, wants to work on the front end of that application. There's a lot of ownership in it. And up to now, she was doing mocks in this Google tool, Google Paint tool, or whatever it's called. So now I thought about, OK, how can I make this? An exercise for her that she works with AI without working with AI and she can kind of do what she's doing without understanding coding or without understanding Figma or all of these tools. So what I was building is quite simple and pretty straightforward is introducing pen.dev. It used to be called pencil, I think, into our repo because that's code, right? And she has a local installation of pen on her machine. She can see the kind of design system of our application with tokens, with atoms, molecules and domain components and so on. But that's not, that's just for her to see what we, what she is working with is linear. So there is now a project where the design agent actually set up issues with all the different design tokens and components to review. The agent sends a screenshot of those and she can go in and just write in whatever format she likes. I don't like the gray or I have a screenshot here of something different and then she just delegates this to linear which spawns a Claude instance itself. This then knows all the harness of how the application works and how the pen.dev design system works and how it's translated then into the application. So the only thing that she's doing is chatting, not with me, but with us, which is me and the agents within Linea within a project management tool. And it gets done. And what is the result for her is that after a few minutes, the agent sends back screenshots of the result into the issue and if she's fine with it she just marks it as done and then it will be built automatically into our front-end code which I review in GitHub and then it's live. So it's a pretty straightforward process where intelligence and human validation comes together without the need of everyone installing cloud code and trying to figure out all the complexity. Sebastian: That's basically the dark software factory, right? So you you have Bastian: In a way it is, yeah. Sebastian: you have two interfaces. It's exactly like also Rodri from from Delivery Hero described it. You have two interfaces. One is the issue tracker, where you pretty much yeah, define what Bastian: Exactly. Sebastian: you want, right? And then on the other hand you have you reviewing the outcome. And once it's approved, then it can go live. Bastian: And then it's also André: Yeah. Bastian: in the other way around. So the results or the work of the agent is made in a way that at each PR has a section which says for the founder. And then it's written in non-technical terms. It's almost a log of this was changed. This is why we do things and so on. Right? So there's a track record of that. there's taking care, like a taking care aspect built into what the agent is doing to connect the dots. And with that, I hope that trust evolves and there's just like, okay, now I understand what this is all about and it gets better and better. But yeah, this is at least my approach at the moment. Sebastian: really interesting approach and I think very yeah p I think for specifically what you're doing right now, very small team building the POC, exactly how you should do it nowadays. Yeah. Bastian: Yeah, of course, after when, three, four people joined and obviously things become more complex and complicated at the same time. André: In the moment where you move out of engineering territory, I would say this is where I would be fascinated to better understand how this is going to work out. Because as Sebastian said, actually in engineering, we are transforming this wire like a solid harness. But like what is the harness for organization? Right. And I think there also, the use cases are so different. I think there is not one harness to like to rule them all, to rule all non-technical work. And I was also wondering if, yeah, like harness for me is still something like a process, right? It's something which is giving you guidance along a process. Maybe it's Bastian: Mm-hmm. André: also more like maybe an interface rather than a process, which you may use. Not sure. I haven't, I haven't. had the time yet to look at CloudFlare OS. Have you heard about that? I think that is going a bit in that direction, like giving the organization a place where you can collaborate with AI in a less complex manner. I think at least the way it looks, it might be something worth exploring, but absolutely get your point. You don't want to throw... the complexity of making AI work for your work on the organization. Bastian: In the end, think one thing that also I underestimated is, I think UI is not going to die. UI is the interface for us human beings. So André: Absolutely. Bastian: there will be colors, there will be boxes, there will be things to touch. André: Yeah. Bastian: They will change because of the nature of how software looks like. But this is the interface to us, and it's probably not a chat. Not in all cases, at least. Sebastian: Yes, yes. Yeah, I agree. So not Bastian: and Sebastian: in all cases. Yeah, I agree. Bastian: Regarding the harness, if I like, I can talk a bit about an experiment that I did this week after I thought about exactly what we might talk about here in this episode. And one of the realizations, think I mentioned this before, is AI is not a producer of text or content. That's what 99 % of those users actually use it for currently. But what it is is actually a plug-in intelligence, Probably intelligence is not even the right word. It's more like a plug-in thinking that you have, right? And what I experimented with is writing a book, novel. So there were a lot of failures in these. in this regards, right? I even, think, books being published in Amazon, being written by AI. But how these were working is actually asking Chachi Bitti or these agents, agentic setups without giving any context and harness to write me a book. What I did is actually setting up and harness for writing a book, which consists of, for example, an agent for characters. there can be within the process up to 20 agents of those kind can be spawned. There is another agent, which is a director. There's a stage builder. There is a narrator. There is a continuity manager. And all of these things have different roles in the setup. And then all of this comes together. And one of the key elements that I tried is actually starting these profile agents. with their own background, with their motives and so on and let them talk with each other on a specific environment, on a specific stage. And I can tell you the first two books that I produced last night, they are just Sebastian: Books. Bastian: like page... They're actually page turners. They are really, really, really good. Really interesting, well written, very poetic, deep. They have many characters with deep relations and so on. I will write about this experiment soon, guess. this is actually what it is, right? So not using it as a producer, but actually thinking about what do I actually need when I write a book? I someone who thinks about the stage, someone who thinks about continuity and so on, right? And this is what we need to bring to people in all kinds of work. Sebastian: I think this is also a great example of playing to the strengths of the LLMs, right? Where what we often do, oftentimes do is we either give too little context, so or like average users either either give too level to too or too little context, asking it to just write me a book about topic X, right? And then yeah, something Bastian: Exactly. Sebastian: average will come out, or give too much context, and then there's either competing pieces of information in the context, or it's just too much for the LN to process and Bastian: Exactly, exactly. Sebastian: keep track of every little detail, which we know they are not very good at. But by dissecting it into these kind of specializations that you did, you keep every one of these focused on a specific job. And that's that's actually also I'm not sure if you also follow Matt Pocock in his latest yeah, I would say advancements to a skill library, he is saying We should not bloat the code generation up too much by giving it too much context about how the code should Bastian: Exactly. Sebastian: look like and how it should be structured, but rather do it after the fact. So first produce the code that solves the problem and then after the fact fact make the code look nice due to several specialized agents that take care of the reviews and the changes afterwards. Bastian: Exactly, exactly. And then it's a circle. It's not that different from writing code. Because writing code is all about abstractions, right? So when we were, software engineers back in the days, know, hope you remember that, when we were thinking about a problem, we were not writing every single piece of the problem, but we tried to abstract those, you know, with classes, with interfaces, with systems, with services, and so on. And now it's quite similar, right? Coming back to my book example, you could say you've write everything about the profiles. not what you want, but what you want is a casting agent that can spin up profiles just by giving it a bit of context about what kind of people you want. And then it comes up with different ideas and so on. And that puts it there, Same for goals. You don't want to write down what an agent should do step by step, but you give it a goal. And the goal itself is an abstraction. And coming back to the organization, a lot of people are really not good at setting goals. What they do is thinking about what can I do next? What's the next thing? What's the next deadline? so on. But actually thinking about a vision and a goal is the hard part. But that's what we need to do now. It's an abstraction. And I think this is actually maybe putting it into a nutshell. Sebastian: So pretty much to land the ship or the plane, whatever, is it thinking in terms of goals and not in terms of Bastian: Mm-hmm. Sebastian: the process and steps to get there, but rather in terms of which agents could be needed in order to get there or which perspectives that could be represented represented by agents? Bastian: Exactly. I think an agent is something like a bucket of thinking, if you will, right? So where you say, this is kind of, this has one goal, it has specific independent context from the others, it might have their own identity with memory and so on. And you need to think about this identity more than what it should do. This comes probably after the fact. André: Maybe a bit too philosophical, but unfortunately I do not find the LinkedIn discussion, but I read a very similar LinkedIn discussion, I don't know, yesterday Bastian: Mm-hmm. André: or the day before. And it reminded me of the phrase, the most profound technologies are those that disappear. I think that, so maybe it actually also on your point, Sebastian, maybe it also goes beyond agent, right? So at the end, want to give it to someone. doesn't matter whether this person exists or is virtual, but you hand this over and you get the result, right? You do not care about how it just gets done. And I think it's on us or someone with a good technical and business understanding to build this kind of operating model. And then I absolutely agree on it's maybe the Bastian: Absolutely. André: agent, but something, but something which is not overwhelming complex. Bastian: Absolutely, I think it's not even philosophical. It's actually coming back to what we have had in the beginning with these agile teams. Wasn't this the same thing? You said an engineer does more than just taking a plan and execute it. They have to sing. An agile team with a product owner has the ownership in place, which is decision making to the edges of the organization. A lot of those things were never really done in most teams. There was this going back André: True, true. Bastian: to engineering managers and really hierarchical organizations. But the actual idea of Agile and new work and all of these movements was exactly that. Having elements of entities, abstract entities where work can be done and you give it goals and that's it. Sebastian: Since it did not change every company, to put it mildly, right? Maybe most companies even reverted back to some form of what happened before. Can we all just sigh in relief and say, the agile thing sorry, the AI thing will also go away? just just just sarcastic comment. it will probably not and and it needs to we need we all need to yeah. see how we can actually and and this is then the the leadership question, right? It's not just again building a system that people can then use in order to yeah set goals in the right way and make progress in the direction of these goals, but also keep these goals aligned to one another, another in a company that that everything pays into the same thing, but rather yeah, that enables people on the way to get there and and still collaborate and and be a social go back to the the socio technical thing that a company is, right, where it's n it's just not technical. Just not only technical, it's also social. Bastian: Yeah, exactly. I think, by the way, that what you just said, most companies won't do it right now. It will take years and years. The big question is if those companies who do it or that do it... are so powerful because of the footprint of this change, right? That they can overtake everything that those companies who fall behind will do. And this is something to figure out. I don't have any hypothesis on this, but you have like these big banks, have Deutsche Bahn, you have other big companies, they experiment with this. But from what I know about how this looks internally, it's like far away from what we discuss here, for example, in this podcast. Sebastian: Yeah, for sure. But but also this again then there comes in the reality of time just takes change, right? Specifically if humans are part of this change that needs to be pe carried out needs to be carried out. Bastian: Exactly. Did you see there was an article that also went through LinkedIn, I think last week, where it said something about the intelligence of the others, where it compared the intelligence of the Houthis using Claude to build an intelligent missile system, while in Berlin the kind of authorities tried to manually figure out what happened in the attack that happened, I think, a couple of weeks ago. So where Sebastian: The data breach. Yeah. Bastian: They were not allowed to use any AI and then you have a terrorist organization that actually uses it to build something really like science fiction. André: Maybe also a good example why you can't compare AI with Agile. Bastian: Exactly, exactly, exactly. André: Yeah. Yeah. The scale, the complexity actually leads me to the question you mentioned, Sebastian, that will take time, right? If I make the comparison to HRI and how long it took for companies adapting it. And I think, Sebastian, you're absolutely right. For some companies, think they probably implemented just a fraction of it. So, and now you have AI way more complex. but also way more dynamic. Hard for me to guess what it means in regard to how long will it take to adapt it. That feeling would need way longer, but it feels differently, right? So the progress you see in certain companies, in certain industries, in certain functions, I think it's phenomenal. And then you also see... people denying that or like, using it as a shed, right. To produce content. it's very unbalanced as Sebastian would say. Bastian: I would say it happens both at the same time. Welcome to the world of ambiguity, right? So that's where we're all in. It happens both at the same time. Those companies won't move that fast. Maybe not at all, but at the same time, there will be solutions for things that those companies do. Let's say, I don't know, security audits or like regulatory audits. Currently, managers have to go to meetings again and again every year, the same procedure, right? They go into those meetings and fill out Excel forms and so on. of this will be solved. So some roles will even go away. designers probably will be solved. Coders is solved, will be even more solved with like the harness that we are currently building. So you need less of them or you need more problems to solve, right? Those things will happen. So Sebastian: Just a different level of abstraction and then yeah, exactly more yeah. Bastian: all of this does not need those companies to change, but they will face the change that they have solutions where they actually don't need people or need different kinds of people or whatever. So this will come anyways, but it's probably not driven by those companies by building their own contacts and harness and so on. This will be just like coming injected by the software they're using by the nature of things that change anyways. Sebastian: So this this was actually my initial my initial assumption three years ago when when it all started, that we just need to wait for the tools to catch up and bring all the the features, right? Bastian: Yeah. Sebastian: This is counteracted a little bit by the by your statement from the beginning that you just should not think in terms of tools and and steps, but rather by goals and then agents that that can carry out or can make plans and then carry out steps in order to move Bastian: Mm. Sebastian: towards the goals, right? Fundamentally, I I also think that until like if we ever get there, until work in in in knowledge work generally will be transformed in this direction, that's a long time definitely. There will be companies that will go in this direction and and will be moving a lot faster than others. And then old companies will die that cannot adapt quick enough like i with every change basically that happens right it's just that currently the the sheer tsunami or the the the size of the change makes it so hard to grasp André: Yeah. Sebastian: if if this will transform everything, right? right now it's very constrained to the knowledge work, to the digital work, to to to to the digital world. Will it also yeah move into the physical world? Will Will all companies basically be affected by this and then André: Yeah. Sebastian: pretty much overturned by these Bastian: No. Sebastian: younger competitors that that are so much faster? André: Nice closing. Bastian: Yeah, could go, yes, that's true. We could go now go maybe for, for next episode or so, go