Latest / Beyond Vibe Coding / #S2.12 Stop Predicting the End of Software Engineers - Raz Schweiger-Shuty (auxmoney)
Transcript
- Raz: I think we humans are really, really bad at predictions. And I want to maybe have like a dispel a bit of predictions to play a game of how we are really bad at overestimating short-term gains and how bad we are at underestimating long-term gains. Sebastian: Welcome to another episode of our new season of Beyond Vibecoding, partnering with Impala Search. André Neubauer: Die go-to-Tech- und Exekutiv-Search-Agentur in Deutschland. Sebastian: In this podcast, we explore the transformational change in software engineering and knowledge work in general. I'm Sebastian Heidemeyer zu Erpen, CTO at Ecosia. André Neubauer: und ich Andre CDPO bei Trusted Shops. Schön, Sie wieder zu Diesmal es eine philosophische Rede. Und, naja, beginne mit einer Frage. Wer weiß Amaras Laue? Sebastian: Yes. Amara's law basically describes that we usually overestimate the impact of change in the short term, but underestimate the long term impact. And the question is, is this also true for AI? That's what we discussed today with Ras Shuty. So hope you enjoy. Welcome to the show Raz. We're excited to have you on the show. And as usual, I'd like to ask you to please introduce yourself to our listeners. Raz: yeah, thanks. Thanks for having me. I'm Raz. I'm currently the group CTO of Fox Money. And I've been very much active in the Berlin tech community for a very long time as part of working in many different companies. think we probably even intersected in some where we had friends working with each other. I know I had lots of friends at Scout24, for example, but worked in Berlin for 12 years now in management roles, started at the company called Oxonata, which was a startup that sadly went André Neubauer: Ahja. Raz: under back in the 2014 days. Worked with Wikimedia from there to Deliver Hero, a very well-known company where I spent five years of my time. And now I'm already, what is it, two and a bit years at Oxmoney. And for those who don't know, because Oxmoney is not part of the Berlin community and ecosystem. if Berliners are listening. Oxmany is a Dissadolf based company, due to or post the pandemic became fully remote. And that also enabled me to basically work from Berlin at Oxmany. And Oxmany is the European biggest lending platform for those who don't know. And I think recently we had an article that ranked us in the fourth place of Fintechs in Germany, which is a nice. Let's call this a nice break from my side with companies like N26, scalable capital and so on. So Oxmoney is very much less known, I think, because it's not in Berlin. So it's less in the cool sphere, but very much one of the most successful fintechs we actually do have right now. Sebastian: Interesting. Yes, all right. And as usual, let's start with a status quo. So how are you currently working? What kind of tools are you using? What kind of workflow skills, be it in your coding or in your managerial capacity? Raz: Yeah. Fantastic question. from, let's start with my private setup. I like my private setup and I can talk about work a lot, but I actually don't get to write code almost at all unless it's for my private projects. To be honest, I've been through so many different setups right now for my private projects. And my current one that works really well is I wonder, have you Have you read this book called Goost? It's a growing object oriented software driven by tests. I love that book. Sebastian: No, I haven't, have to admit. Raz: It talks about basically architecture and mostly outside in TDD. I'm going to drop some stuff that maybe will get me killed by the audience. I was never a fan of TDD until AI. I think AI made me actually love TDD because my brain never worked well at thinking about the small units and then writing tests for them. And then from unit to grow out, I was also always better at outside in. So I wrote a really big test and then wrote my features and then the unit tests afterward. But, now with AI, it is actually a fantastic situation. So my setup is actually very much a grill me, where my, have a lot of. grill me until we're done. So I want to basically find every, every stone and turn all of them as much as I can. Then I get to have my unit tests written and I can basically go and review those structures as I go from private projects. And then from there I go to integration tests, I see implementation, and then I go to some tests on the outside. So basically I have those kind of like cleared. I read recently this really nice book that's called What was it? Specification driven development, which is kind of built into this AI framework and helped me a lot to basically use Goost and this to set up my setup. Now, how does my team work? Not necessarily this, right? This is a personal choice from my side. At Oxmoney, we try to give freedom to people. So we ended up giving many different tools. We've been using Copilot since. I believe since it was formed basically before my time. let's say 2022, 2023, Co-Pilot was already integrated at Oxmoney. We kind of believe that the idea of AI is there to support and help. I also will say another thing here. I love the name Co-Pilot, but not for the Microsoft tool. I just generally like to look at AI as my Co-Pilot or me as the Co-Pilot of AI depends on the mission. Sadly, Microsoft stole that name, every time I say copilot, I need to differentiate, it the software or is it how I look at it? But anyway, we use copilot, but later with time when I joined, we also added cursor, we have added cloud, so we allow our developers to find what setup works for them. And as long as it follows our principles and governance of how we want to work, and we have a lot of protecting systems on the outside, we actually prefer to give our devs freedom on how they would like to set up their AI interfaces. Of course, we make sure that we have all the governing side and the testing happening. But as far as do you use Cursor? Do you use Cloud? Do you use Copilot? By all means, whatever makes you more productive, we're good. It came also because a lot of my devs, for example, are PHP devs. They really like to have their environment. So many of them struggle to move to Cursor from PHP Storm. So giving them Claude and using Bash is a much easier setup. Some people actually do enjoy Cursor and how it works. It gives them more ability to, in a sense, interact with their IDE. So we've kind of given them many different opportunities to basically be as productive as they can. So that's basically, to... I hope I answered it. I didn't babble too much. Sebastian: Yes, André Neubauer: Absolut. Sebastian: thanks a lot. the the fact that you're a fintech also, and you hinted at that a little bit, makes pretty much compliance and and security much more important, right? So and you said Raz: Absolutely. Sebastian: you hinted a little bit that you have a lot of protective systems around pretty much what the engineers are using in order to develop to develop code. can you like Detail this out a little bit more without going into any trade secrets, obviously. Raz: Of course, exactly. That's what I'll do. So I think the one most important thing I'll start before, you know, the basics, everyone say, first of all, education, you cannot put a system that in the end, any system can be abused. So what you want to do is to educate. So we do give a lot of education around AI and how to use AI, whether it is external courses or internal information that we share, right? So oversharing information is always good. Let's quickly ignore that because we're talking about tools. For us as a FinTech, a lot of the important parts is honestly, let's even start with which AI can we use. So even when we're looking at a, I don't know, like a chat that you want to roll out to the entire organization to make people, you know, what they do right now, which is probably self-evaluate themselves with their chat bot or send emails and so on. You're exposing potentially company information into a whatever, a model. And it's not necessarily something that we can allow, right? Like those models are trained onto that. So if you have information going to the U S that's also not okay. So we started by actually in a harsh way, blocking every external. Churchy PT and so on and so on that we can, no access to that, but we've given them Langdoc Langdoc is basically, I don't know if you know it's at the Berlin based company and they aggregate André Neubauer: Ja. Ja. Raz: different models. and The agreement with Languc is that they don't train on that data set. The data set is only staying here in Germany and it's basically enterprise ready. You can have role-based access. You can do many things. And we also told our people you're welcome to use it as your own, you know, from your private laptop. If you want to ask about closing hours of your whatever, or you want to use nano banana by all means, OxMoney is happy to pay for that for your private use. But the idea is that you don't make that mistake accidentally where you go to Gemini. and then you put some customer information or something like that. So we are blocking all of these. So this is number one for us, the ability to access the right models. You get all the models you can on this thing, but also we tell you this is your go-to place from the organizational standpoint and from your computer. After that, we have GitLab Ultimate. So basically we see that on the pipeline itself, we want to automate as much as we can to... to avoid the hugging face type of scenario or any type of NPMs that you don't want from supply chain. So we want to have our systems working to block from any type of mistakes that could happen. So anything that goes into our production goes through a rigorous CI-CD pipeline that has a lot of automatic tests. And at the end of the day, it's again about the process. And process for me would be a 4I principle. And then AI is not a 4I principle. It's great that I can have my AI co-pilot, not the tool, but as a concept. And I work with my AI co-pilot, but the code review should be done by another set of eyes, also with the support of AI. That's fantastic. But at the end, I am the human who ships the code and there is a human that actually signs off on the code. So we test ourselves and we also have a testing period, right? So similar to any other company with a secure software development life cycle. We also make sure that there is a QA phase. We do not have QAs by the way, but we have at least a testing period. Either it is automation end to end, or if there is a bit more complexity, then the developer that takes the code also does a manual check themselves. So this is basically how we play with it. So of course, a lot of automation tests, unit testing, integration testing, a lot of security tests with GitLab Ultimate. And at the end, of course, policies that set us to what we can and can't do. according to regulations, which is for a FinTech super important because protecting the data for customers are number one priority. André Neubauer: Und das auch der Ziel? Also was du zielst? da was, wo du sagst, dass du erreichen willst, aber wir wissen nicht, wie man das Raz: To be honest, think all of us are chasing a, like we're dogs chasing an imaginary car and the car is faster than us. And also when we catch the car, what do we do with it? You know what I mean? It's AI is moving so fast. So for example, hugging face, everything that happened there was a big shocker for us, right? Like, because we didn't know it's happening and how it's there. So a lot of the thoughts we have right now is how do you minimize and how do you, make sure that supply chain attacks don't happen. And that's a very difficult task to have. So we're working on that, of course, to kind of at least bring it into a better environment from our side. And again, a lot of it will also be adding more and more systems that we can to make sure that we can deliver code faster in a more secure way. So adding more checks into the pipeline while we try to also optimize to... It's a bit of a playing like this thing with the shaders, right? You try to balance both at the same time. So we're trying to do both. We're trying to basically increase the velocity of the developers while still reducing risk of using those tools. yeah, we are far from the target state, but I also cannot tell you what the target state is because every three days the target state looks different and it's for us to always keep our eyes on what's going on. And basically at some point, I think the target state will reveal itself just basically a software design, like a garden, right? You basically plant small things and you kind of find yourself with the garden that is revealed to you rather than planning a garden in advance and then realizing that the soil doesn't fit or the sun is not coming from the right direction and so on and so on. André Neubauer: Ich denke, die Bilder mit dem Garten sind eine tolle, aber an irgendeinem nicht korrekt. Denn für den Garten kann planen, weil es viel Wissen sich und viel Erfahrung und Leute, die das bereits haben. mit AI und dort, ich absolut agree, ist es ein bisschen unfair, zu sagen, dass das das Zielstaat Man muss verändern. Das ist auch eine gute Wegweite in unsere Meat-Sektion. Es sehr schwer, erlauben, was das Zukunft für Aber vielleicht willst die Meat selber Raz: Yeah, absolutely. I think the discussion was our short-term overestimation versus our long-term underestimation. Right? So we are really, André Neubauer: dass Raz: really bad at estimation. We as humans, all of us, think, like this funny thing with, think Bill Gates talked about how, what was it like? Someday computers will have 512 kilobytes of RAM and that's like in X amount of years and We broke that by a couple of gigs. So we are really bad at underestimating short-term and also really, really bad at sorry, overestimating short-term and really bad at underestimating long-term. Sebastian: Yes, yeah, yeah, f fully agree. And and I think the same is also true for AI, right? Even though right now it Raz: So luckily. Sebastian: feels a little bit different in that I would not have predicted, and most people would not have predicted us like two years ago when ChatGPT came out, or was it even more three years or something when ChatGPT came out that we would now have like automated coding agents, right? So but but at least the The impacts on the labor market are less clear and predictable, right? In the sense that the short-term impacts are not measurable as of yet. Raz: Yes. Sebastian: If you discount pretty much the the fact that some companies are using this as a nice financial market explanation for laying off people, right? So it's it's not necessarily Raz: Yes. Sebastian: that the the the gains in productivity already would allow them to l let people go. There probably are productivity gains, but they usually would be overcompensated by the new ideas that that companies would have and then that to keep people busy, essentially, right? Raz: Yeah, on that side, actually, I'm quite an optimist. think, I mean, if you look at the idea of how the washing machine was invented to reduce labor from the home front and what happened in the end is we just do laundry more. I think most companies are realizing that if we have a superpower that can improve productivity, letting people go means I invest less in myself, but actually having the same team or growing that can do 200 % means I can actually improve my business by 200%. much faster. So at some point I believe that it will also equalize a bit. And even on that, I think if I look at AI, AI is just another, at the end of the day, another obstruction from machine language. And if you look at how machine language started, there was one company, a couple of universities and NASA that you wrote code in. So if you were a software engineer, you worked in five places. C came in. Basically brought us Microsoft and other companies. we basically spread the workforce over. I generally believe that they all just very soon. I'm just an optimist. What can I say? We'll also create many different opportunities that other people will create startups and then they will need their vibe coders or agenti coders to come in and then they'll be maybe smaller teams, but more work. I generally think that the software engineering position is not that it's just transforming. We just need to be kind of part of this transformation to figure out what are the right toolings, processes, and how to do it well. But I'm on the optimistic side, by the way. So this is my maybe great future estimation. Sebastian: And is this now an overestimation of the short term or an underestimation of the long term? André Neubauer: Haha! Raz: Yeah, this is, is, think, maybe an overestimation of the long term on my side. The short term right now, we all see that the market is a bit stalling, right? We, mean, I think it's a bit changing right now. I don't know about the two of you, but I have less positions. I have more positions open and less candidates right now. So maybe that's my own single point of view, but I think more people are hiring again. It seems, but again, I am not sure. It's hard to kind of look at it from the point of view of where we are right now without being able to zoom out. Because as you mentioned, Sebastian, generally, I also believe that many companies in the last two years use the AI as a really good excuse to say, we all went through the COVID hypergrowth. All of us, instead of optimizing for profit, we've optimized to who can hire the most people. And I think Spotify even admitted that, right? They had a layoff. said, yeah, we chased numbers of over hiring. I think other companies have not admitted that, but they're doing the same thing. So I generally don't attribute AI to this because I can share from honestly, from the Oxmoney side, we generally enjoyed having the extra ability to do more. There is a roadmap to be done. I have other C levels who work with me and as a CTO, the two of you probably can relate. You always get people coming with ideas to you. And usually as a CTO, say, I am the protector of the roadmap. The team can only do X. So if you want to add one more, I need to make sure that something gets out of the roadmap. Now we are able to say, yeah, let's give us more. it changes a bit the mindset. You know what I mean? Like back in the day, we wanted a lot of data points. Should we or shouldn't we invest in this? Because this will be seven weeks to prototype to even figure out if it makes sense to build. Do we really want to invest a team today? We can prototype that quite quickly. So it gives me the ability to say, Hey, Why don't we try it out and throw it to the trash if it doesn't work. It's an easier negotiation way for me right now than before. So again, I think if we change a bit the ways we're thinking, it's quite an optimistic future here. Again, this is me over estimating, I think for us, but the optimistic in me sees at least on my day to day how working with my chief growth officer and my CFO and my COO, when they come and they want something, I'm able to give them some quicker results and then we can test the water if it's worth it or not, rather than back in the day. So why would I want to cut my workforce if my workforce can do that and drive the business forward? Sebastian: so that two two thoughts in my head right now, but let's b double click on that one second. So number one, maybe how how big is the the engineering workforce at at Augs Money currently? Raz: yeah, fantastic question. We are about including our data team, our AI team, the technology and infrastructure, approximately, what is it, close to 100 people, I would say, with the AI. Sebastian: Yeah, all right. And how do you pretty much play innovation in in this way, specifically in this regulated fintech environment, right? S coming from you said, Yes, tr testing something, prototyping something is incredibly quicker now possible than than it was before. So it's a lot cheaper. Nevertheless, you probably need to still if you bring it in front of customers, it there there needs to be some some guardrails around okay, what kind of innovation needs Raz: Absolutely. Sebastian: like s security upfront? What kind of innovation maybe not, because it's just an like, I don't know, marketing tool or what whatever. It's not not that relevant, right? So yeah, how how do we do that? Raz: It's exactly strategically where you can and you cannot, right? I'll distinguish POC is not MVP. So we can prototype things that doesn't mean that we take this prototype and put it in production. It's more for us to test something. And then if it looks good and it behaves well and it makes sense, then we can build the product, which might have different rigorous process to how to build it. And then comes to also the strategic part. One thing that we've done, I think it's not new. Everyone is doing it right now is a chatbot. But the chatbot is an easy thing to basically build quickly, have the ability to let your customers get easier information before they need to talk to a human being. And it's something that relatively is, you know, it's there. So we've experimented with it. We did a quick prototype. We saw that it worked and then we said, okay, let's build it into the app. Right. And that's where we saw, this is how it works well. This is also where we separated and actually made the security part work quite well. and after we did that, we saw it works well. So again, let's build a front end in the web part. So those things were able to do relatively quickly. I think it took us a month to basically prototype, do a lot of tests to see that you cannot inject a kind of AI into it and so on. Make sure the answers look good. And then. build it properly from basically prototyping it with a chat, with a LangDoc chat, to then put it properly in Bedrock under proper security that is available for the customers, and then build the UI for the mobile and then later for the web. That was quite simple for us because in the end of the day, the goal was to first let people play with it. We saw that it makes sense. Let's let people also see that you cannot abuse it. And then we slowly opened it properly, meaning We don't put it in a long book check. It becomes a bedrock. It becomes part of our AWS infrastructure. It's become, it's under our API gateway and so on and so on. So of course security comes first, but the logic itself, you know, that, that chat agent and how it played, we already knew how it's going to work from day one because we basically use our FAQs. We trained it with our information. We trained it what it can and cannot say. We put the right boundaries that gave us this in two days. the ability to let everyone in the company kind of play with it and see if it works well before we even put it in development. And then development was relatively quick. But again, this is where if you have your core, right. And you have a tree coming from the core, it's easier to do those things at the, at the outskirts, but it's not something I will plug in the center of my system. That's, that's where the reality is really tricky. That's where I wouldn't prototype things and be like KC. But anything you kind of plug on the satellites or basically, I don't know if you have a moon to your planet, these are much easier things to do on the outskirts. Sebastian: Yeah. And then pretty much the the way you're doing it is you build this BOC, you make sure that that it's actually working, right? You test it internally, which means that you have the audience under control. You know it's people who work for you and who wouldn't abuse the system, right? And then you once you validate that it makes sense internally, then you have strong signal that it also makes sense the further investment in order to, yeah, pretty much secure it and bring it on the production infrastructure. Yeah. Raz: Absolutely. And the other part you can do is automation, right? Because this is again, I think software engineers is one of the best things that we have done is not just building things, it's automating processes. You can look at the loan process that what Oxmoney did was take a, go to a bank, you talk to a person, their paperwork's done to an automated process. If you look at it also as internal concept, AI is really helping us also automate processes for the company. So basically we can look internally. and then have the AI team see how team X, Y, or Z work at the company and figure out how can we support them to be more productive, to be able to basically address more cases with the customers, or how do we improve flow of similar to what we do in software engineering, the velocity of the team through the flow and use AI in those steps. And I think that's also been a fantastic way for us to... Let our employees experiment because now even people who are not using or were not using AI back in the day are now trying because they see, I can now play with some things, talk to the AI, come with a concept and then go to talk to the team and say, Hey, I've done this thing. And with my Lungdoc chat, I was able to potentially optimize here 20 % of the work. Can we automate that? Can you take this and make it into Salesforce so I can basically have it automate for the team? And that happens. And that's a fantastic thing for us where people who are less tech savvy are able to come in with tech savvy ideas because they basically have a solution architect that can prototype stuff for them and they just talk to the thing and then it gives them really good solutions and they know the context. And that's one of the most fantastic things that is happening in my opinion. Sebastian: Yeah, I I think that that's a huge leverage point for every company actually that has software engineering c capabilities, right? Using this internally for for productivity gains. So do you have teams that focus specifically on internal stakeholders and customers? Raz: Well, we technically our structure is built in a way where we have teams that are more on the back office side and teams that are more on the front facing side. So we have our customer acquisition that is more trying to bring people in. And for example, our customer management, which is the app and what, how do we train our customers? But of course we have internal tools and those internal tools are built with our software engineers and our own. So those teams are focusing inside. And we also have an AI and automation team and the AI automation team are solely focusing on looking at what we can improve with AI. They have software engineers and AI engineers. So basically it's in a sense, it gives us a cross-functional structure, even in the AI team. And then they can go and support a team that works with Salesforce, for example, how to automate their processes or the risk analysis team, and also focus on the chatbot and help the software engineering team, for example, with properly setting up the guardrails of the AI chat. Sebastian: Yeah. Let's let's bring this back a bit more to the to the meat, right? yeah, so what are some of the thoughts that you have when you think of short term overestimation and long term yeah Raz: Yes. Sebastian: underestimation? Raz: My favorite one is I think last year people came to me and said in about a year or two from now, we will not need software engineers anymore. And I am a year in since that statement was given and many people said software engineering is dead. I think this is a laughable reality. Not just because I, you can see that you still need software engineers. It's generally also a reality where. At the end of the day, whether you want it or not, someone needs to understand the specifications. Someone needs to understand the interplay of systems. And I generally don't think we will not need software engineers. We'll just need to rethink what is the role of a software engineer. Maybe. But, I don't see us in the next. I'm going to make my prediction here by 2030, software engineers will still exist and, probably we'll have more software engineers looking for jobs. like more software engine is working than less. That's my general stupid prediction that I can have an egg on my face maybe if I'm wrong. André Neubauer: Für mich der Punkt mehr, was die Verantwortung was die Verantwortung Raz: Absolutely. André Neubauer: eines Software-Engineers das wird dramatisch Ich denke, Raz: Yes. André Neubauer: wir sind einig. Wo ich einen schweren Zeitpunkt also ich bekomme die Frage, ob ich kurzen und langfristigen Werte und denke, gibt auch eine法 für das. Und Amara. Sebastian: Jevan's paradox. I d no no, it's it's not exactly that. pardon? André Neubauer: Ja, ich glaube, gibt eine Ämere-Linie. Roy Amara. Sebastian: yeah. André Neubauer: habe das ein bisschen finde, das eigentlich nötig. was ich ... Also, obwohl ich verstanden hätte, sehe eine Herausforderung. Vielleicht kannst mir Russ. Ich denke, wenn du Du überestimierst den kurzen Impakt, das kann zu langsamen. Ich das kann gefährlich Denn am Ende, obwohl der kurze Impakt unterstimiert wie es in der Amara.Law kompounst Es ist eine hohe Chance, dass einem bestimmten Punkt hinter sich Man wird übernommen, weil man zu viel Deswegen frage Ich denke, wir alle wollen richtig aber vielleicht das hier nicht der Punkt. Vielleicht ist es wichtiger, vor dem Kurs oder zumindest mit dem Kurs, als zu langsam zu und vielleicht richtig Aber wenn du nicht richtig denk oder aus dem Spiel Raz: I agree. I don't think I'm, I'm disagreeing with you. think I'm 100 % agree with you here. I think it's great to set goals. Basically you shoot for the moon and in two years, I think the general idea of stating we will not need software engineers in two years is a wishful thinking, but someone needs to say it to keep 99 % of André Neubauer: Gehen wir! Raz: the people basically on their edge and saying, okay. It's a big event and I need to change. I agree with you a hundred percent. just also. think that saying, I don't know, when you liquid those posts, and in two years, no one will need software engineers. It's also for me, it was laughable. Sadly, the problem with that is there is a full new generation of students that will probably be afraid to actually take that role. And we might, and we are good at that, that is an industry in the tech industry. We might treat ourselves in defeat to not get the new generation of developers because we are basically kind of promoting this, we will not need you who don't come in. And we've seen through 50 plus years of software engineering that every time something became easier, it actually opened the market for more and more people to join. And that's where my biggest fear is we're kind of scaring away the students, the juniors from coming in. But don't get me wrong, I think setting those goals and shooting for the moon is the only way to go because you're either moving forward and you chase everyone that are having high ambition goals, or you kind of stay behind and you die. And that's, that's the reality of stagnation in technology. It moves so fast that if you're not trying your hardest to remove all software engineers in two years, you're basically out of the race. Even though this is an attainable goal. Sebastian: So basically, then the ramification of the statement, right, or of the Amara law, Roy Amara law, is that you should like have a pragmatic view basically on things, right? You should still follow the curve and not not say, we're anyway overestimating the short term, so we can lay Raz: Yes. Sebastian: back, right? That that's nothing you should do. But I think it's it's also important for Yeah, for for us and the engineers. all all of us are impacted by this heavily right now. And we I think have a responsibility to, yeah, bring a little bit of this pragmatic view on the table and calm down people. because I fully agree, right? We'll not fire half of the team in twenty twenty six or twenty twenty seven. That if if we're doing that, then we're probably doing something wrong. Or there are other reasons, economic reasons, whatever, right? But Raz: Yes. Sebastian: but it's it it's not the goal and what we're rather tackling, and this is also has been a a stream across the different episodes that we're having, we're rather moving bottlenecks and pretty much if we remove one bottleneck, the write and code bottleneck, then we pretty much make other bottlenecks visible. following the theory of constraints, right? And then we need to address these bottlenecks in the organization. And it's usually not that we lack ideas. So that that's rarely the bottleneck. Raz: Absolutely. 100 % aligned here. And just to also say something else, because I talked about us scaring the juniors a bit, the one thing I do think is critical and maybe I'm addressing an audience here that are already believers because I guess whoever listens to this podcast is already buying into it. But it's important to say, because I've already tackled many developers who are still into this, I do not use AI, AI Slop and all those things. The reality is that this industry is changing. And I'll give my best advice here from a person that has been 20 plus years in my career. Engineering is about moving forward and advancement and being the person that was, I don't know, back in the day using cards to print out and assembly to say, I'm against C because it's not computer language would have meant you've been left behind. I think. Basically saying I want to be like those monkeys with my hands on the ears, my, and my eyes and ignore it. AI is not going to go away. And you basically need to also step up and learn how will your role shape up and what your role is now. And this is something that, you know, I've been also trying to, to convey to my team, right? Since I've joined, it was okay. If you like it, you don't like it. You have to understand that this is where we were going and this is the reality. and need to educate yourself because it's going to be a situation where you are going to compete at some point with people who are super eager and they know what they're doing and they know how to use this. And that will be their advantage. And that's something that you cannot ignore. So that's also where those goals are fair. It puts a bit of fire under our asses to also say, okay, I need to learn something new because there is a risk here. I understand why we're doing it, but again, it's a bit of a marketing ish as well into it. Sebastian: Yes, yeah, and it's sorry, just following the exam so examples I usually give is the telephone operators, right, that were plugging cables in order to connect people or the lift boys who were like making lifts go up and down, which is fully automated, or even if you look at at farming, right? So a huge amount of manual labor was was needed a hundred years ago and now everything is mechanized and now even automated partially. So This change is inevitable and it's your choice how you how you deal with it or how you react to that, how you take this change, right? Do you embrace it? Do you try to understand it and try to really follow the curve or rather be ahead of the curve? In the end, it's not AI currently, at least. It's not AI that's taking your job, it's the person who can utilize AI to be more productive. André Neubauer: ich denke, es immer gut, die andere Meinung in einem Podcast Ich sage das nicht, als würden wir 2027 Hälfte Einer der kompariierten die ich ist, dass wir uns mit dem, was wir in unserer Karriere Wenn jemand mir würde, dass die Veränderung sehr im Vergleich zu Cloud, Mikroservices oder Internet dann würde ich Vielleicht die Industrialisierung noch näher. Ich nicht sicher, aber ich war nicht da, als die Industrialisierung passiert. Ich kann beantragen. Ich weiß nicht. Man kann die Zukunft Man nicht nur nach hinten und die Punkte dort Ich bin wirklich nicht sicher. Für mich ist es, wenn man sich was mit Industrialisierung ich denke, es ist viel schneller als wir alle glauben. Aber niemand sagt, wie schnell es geht. Ich nicht schämen, aber ich finde es Das ist sehr ehrlich. Wenn Sie LinkedIn öffnen, weiß nicht, wie Ihr Fuß aussieht. Aber 99 % AI-Stuff. Dieses Framework hier, dieses Verbesserungsgerät dort, ein neuer Modell, ein neuer Wagen, neuer... Was auch immer. Ich würde gerne in der eine bessere Verständnis wie die Industrialisierung passiert Es war kein LinkedIn, also die Leute es bemerkt. Raz: Yeah. André Neubauer: Aber ist auf ein gewisser Weise überwältigend. Ich habe dieses Gefühl, dass auch Junioren oder andere Menschen in unserer Industrie Sebastian: And for good reason. Raz: Yeah. Sebastian: So I I I like I would not negate anything you just said, Andref. I I Raz: Absolutely. Sebastian: I so feel you André Neubauer: Ja, vielleicht habe es zu freundlich Und würde auch nicht mit deiner Sicht All das, ich sagen dass es eine hohe Chance wir falsch sind. Weil wir vielleicht den langen Impakt Sebastian: And and this is the this is really the thing, right? Under are we underestimating the short term now or not? Because AI to some extent, and and that's what what I tried to say in the beginning, right? It surprised me how quick it became so good. So I it in initially, right, we all were like, copilot, wow, now you can do autocomplete for for lines of code and then you could like write some comments and it would Create small functions, right? And that like you blinked with an eye and suddenly there was clawed code and and it would actually produce working software and could do TDD and and now skills and and different models and you you you named it, right? No, fully I fully agree. The the thing, and that that's something that Pip Klockner is always saying the the the speed or the theoretical maximum of speed that such a change can have is vastly different from the actual speed largely because people need time right and this is i think like for for for very small startups who are starting right now they they treat this problem fundamentally different they they will hire less Raz: Yes. Sebastian: people for sure because they can automate much more they can build the system in a different way companies that have more people have such a degree of complexity that that it will be hard, just hard, to reduce the amount of people significantly. I mean, Elon pulled it off with Twitter, then named it X and I don't know, it's still working, right? Even though he did many things in order to keep get the load down, right? Because there's less people on on it now, I think, and less people who are advertised. But anyhow, I mean there there's this This efficiency is in all companies, right? And it it always depends on where you're at at the growth curve, right? So do you have growth ahead of you? Then probably you have many ideas and and have appetite to invest more, and you rather want to go for productivity gains in order to free up more of this capacity. If you, however, are in an economically challenging situation, then yes, it it probably makes sense to look at it from a different angle. and I think that this is really different f from company to company. However, the the the huge cuts that due to AI productivity gains, yeah, we might we might see this in 2027, right? So we we don't know. Predicting the future no predictions are hard specifically if they are about the future. Raz: Yeah, exactly. can tell you what, Sebastian, I think you can also, for me, you hit the nail on its head because at the end, what you just described is how a company runs with or without AI, right? So at the end of the day, if you have a growth trajectory, then in general, you will leave the organization as is, or you will try to grow your organization to basically reveal more opportunities. If you're in an economic struggle, then you will cut the organization because you want to optimize for profit and to keep your head above water. I think this didn't change the rules of business. I'll be very nerdy here. I love Star Trek and DS9 and Star Trek, Deep Space Nunders Quark who continues to say, at the end of the day, are 224 rules of acquisition and quotes them. So at the end of the day, part of the rules of economy is you do not want to pay a price higher than the profit you can make from it. That's the general idea. And this is where we are right now. And you kind of hit a really good point in my opinion, where new companies will come up and they will be very efficient, probably with lesser teams, but it will also democratize the idea of building. in my hope, many more companies will come up with many more ideas. And then maybe we'll not have thousands of developers in one company, but rather we'll have many, different developers in many, many different companies spreading. similar to how software engineering was a NASA problem and a university then became. know, IBM and Microsoft and so on. And then we had the new internet boom of the 2000 with Java and C sharp or whatever we had back then. I think it's still used today. We'll find ourselves now with many, many companies, maybe with smaller teams, but I generally think that you will still need a pilot or a copilot to your AI. And that's where I'm very hopeful, but again, I have no clue really what will happen tomorrow. I've made every mistake possible trying to predict the future. I'm really good at analyzing the past. I'm good at the retrospective side. Sebastian: All right, I think we need to come to the end of this episode, So what are what the fuck moments you still experience and maybe what are wow moments that you still experience with AI. Raz: question. so on the what the fuck moments I find myself, I still see AI is not great with a very large code base and context. So you sometimes produce something with AI and I can see the issue and the thing keeps telling me, guest lighting me that this is not the issue and ignores basically something as simple as a junior developer will find it. And it's always make me kind of lose my mind about it where I can, tell the AI, no, I, this is the issue. This is what you've done wrong. I get it. Yes, you're absolutely right, but it's not here. It's somewhere here. I'm like, no, just listen to me for a second. but honestly, what wowed me, I sat with my son a couple of weeks ago and he described to Claude in my home, a game he wants to create. basically I, it's a nine year old, right? I sat down, wrote everything with a grill me and he created a web game that he loves. And that's like a wow moment for me because as a nine year old, when I learned how to write Pascal, I learned how to do, I don't know if 10 is bigger than nine, then you return true or false. That's like basically where I was. My son created a web game. So this is where I was wowed how how much it actually opened everyone to try stuff out and actually democratize the idea of building, which is fantastic. And that was a, as a parent, at least that was a moment for me is to see how my son can do amazing things with AI. André Neubauer: Thanks for sharing. Sebastian: Great stuff. Yeah. Thank you, Ras. it was great chatting with you. yeah, thanks for coming on the show and bye bye. André Neubauer: Bye bye. Raz: Bye bye. Thanks for having me.