1 00:00:07,440 --> 00:00:10,120 Hey everyone, Ben here. Welcome to the Stellar Work 2 00:00:10,120 --> 00:00:13,280 podcast where today's episode dives into the journey from 3 00:00:13,280 --> 00:00:16,840 coding in the 90s to building AI powered platforms. 4 00:00:17,360 --> 00:00:21,200 Our guest Hendrick, legendary for his role at Spotify and for 5 00:00:21,200 --> 00:00:24,320 making agile methods visible to the world, shares how he 6 00:00:24,320 --> 00:00:27,520 transitioned from pioneering software on and organizational 7 00:00:27,520 --> 00:00:30,680 design to the frontiers of artificial intelligence. 8 00:00:31,160 --> 00:00:34,480 We explore the real origins of the Spotify model, the impact 9 00:00:34,480 --> 00:00:38,080 and the myths behind coding productivity with AI, and 10 00:00:38,080 --> 00:00:41,440 Hendrick's adventures developing AI agents that redefine what it 11 00:00:41,440 --> 00:00:43,480 means to collaborate and innovate in tech. 12 00:00:43,960 --> 00:00:47,520 This episode is packed with insights about how companies use 13 00:00:47,520 --> 00:00:51,120 AI, but the junior deaths still matter, and how architecture 14 00:00:51,120 --> 00:00:55,160 decisions are evolving in a world of code written by both 15 00:00:55,160 --> 00:00:58,200 humans and machines. So let's jump right in and learn 16 00:00:58,200 --> 00:01:03,680 with and Rick Nieberg. Hello Hendrick, welcome to the 17 00:01:03,680 --> 00:01:05,239 Stellar Work podcast. Thank you. 18 00:01:06,360 --> 00:01:08,360 Good to be here. Yeah, thank you so much for 19 00:01:08,360 --> 00:01:12,960 being here. I I saw you delve into the world 20 00:01:12,960 --> 00:01:21,000 of AI some time ago, but before we dig into that, maybe we we 21 00:01:21,040 --> 00:01:24,480 pause for a second and you introduce yourself to whoever 22 00:01:24,480 --> 00:01:27,760 does not know you yet. I assume everybody does at some 23 00:01:27,760 --> 00:01:31,760 point, but could you introduce yourself to? 24 00:01:31,840 --> 00:01:33,280 Sure. So I'm Henrik. 25 00:01:33,280 --> 00:01:37,560 I'm a Swedish guy sitting here just outside Stockholm, grew up 26 00:01:37,560 --> 00:01:42,240 in Japan, umm, and did a lot of coding. 27 00:01:42,360 --> 00:01:47,800 I've studied at a Technical University in Stockholm and umm, 28 00:01:47,800 --> 00:01:50,160 I'm a soft pro guy, fund fund fundamentally. 29 00:01:50,320 --> 00:01:52,800 But then I got dragged into a bunch of startups during the, 30 00:01:52,800 --> 00:01:55,720 during the 90s and started getting involved in the 31 00:01:55,720 --> 00:01:58,800 leadership things, building companies and startups and to 32 00:01:58,800 --> 00:02:02,160 try to understand more than just how to write good code, but how 33 00:02:02,160 --> 00:02:05,440 to organize people and, and, and the work which got, which was 34 00:02:05,440 --> 00:02:07,480 interesting. So through that work, I, I 35 00:02:07,480 --> 00:02:10,400 stumbled into what later on started becoming called agile. 36 00:02:10,639 --> 00:02:12,760 So things like scrum and extreme programming and things like 37 00:02:12,760 --> 00:02:15,160 that. And through that I started 38 00:02:15,160 --> 00:02:17,120 getting involved with companies that wanted to work in an agile 39 00:02:17,120 --> 00:02:18,680 way. So a bunch of different 40 00:02:18,680 --> 00:02:21,920 companies, large and small, and I was there helping them 41 00:02:22,040 --> 00:02:24,640 basically figure out how to work effectively, build better 42 00:02:24,640 --> 00:02:27,120 products faster and have more fun along the way. 43 00:02:27,280 --> 00:02:30,440 And of course, as a side effect, I, I, I learned a lot also. 44 00:02:31,560 --> 00:02:34,680 So yeah, I spent a bunch of years at, at the Swedish startup 45 00:02:34,800 --> 00:02:37,840 called Spotify, which then grew and became this wonderful thing, 46 00:02:38,480 --> 00:02:40,880 learned a lot from that, published some videos and, and, 47 00:02:40,880 --> 00:02:43,760 and some people started using the term Spotify model to 48 00:02:43,760 --> 00:02:45,720 describe what we were doing, which was kind of interesting. 49 00:02:46,280 --> 00:02:47,640 But what? Is that interesting? 50 00:02:47,760 --> 00:02:49,120 Yeah. And from that I learned a lot 51 00:02:49,120 --> 00:02:52,720 about scaling. And then later on I stumbled 52 00:02:52,720 --> 00:02:56,480 into a Lego which, which my, which was my first kind of non 53 00:02:56,480 --> 00:02:58,960 software company, which was very interesting. 54 00:02:59,600 --> 00:03:01,200 Umm, and then I got back into coding again. 55 00:03:01,200 --> 00:03:04,720 I ended up at, at Mojang, first as a coach, but then I, I kind 56 00:03:04,720 --> 00:03:08,320 of got sucked into the, the Minecraft team and started doing 57 00:03:08,720 --> 00:03:10,320 development again, development and design. 58 00:03:10,320 --> 00:03:12,640 So I spent about four years, umm, doing that. 59 00:03:12,800 --> 00:03:17,000 So quite, quite, you know, a mixed stuff, but and and then 60 00:03:17,000 --> 00:03:19,680 now and then when AI came out, I made a complete shift into AI 61 00:03:19,680 --> 00:03:21,840 space. But I think agile has probably 62 00:03:21,840 --> 00:03:27,040 been a bit of a, a theme behind, behind all of it and basically 63 00:03:27,240 --> 00:03:29,720 helping companies figure out how to how to work better. 64 00:03:29,720 --> 00:03:31,360 Is has also I guess been some kind of a theme? 65 00:03:34,000 --> 00:03:39,720 You, you passed for a second on on the Spotify model and I, I, I 66 00:03:39,720 --> 00:03:42,680 always thought that you were like the the mind behind it who 67 00:03:42,680 --> 00:03:44,720 promoted it. But now you said people are 68 00:03:44,720 --> 00:03:49,320 using the term Spotify model. Yeah, no, I never, I didn't coin 69 00:03:49,320 --> 00:03:53,200 the term. What what I did was basically 70 00:03:54,080 --> 00:03:57,400 make it visible. I, I started publishing articles 71 00:03:57,400 --> 00:03:59,680 about what we did and I, I didn't invent it. 72 00:03:59,680 --> 00:04:01,720 I helped shape it. But there were a bunch of people 73 00:04:01,720 --> 00:04:04,600 at Spotify that we, we worked together to figure out a way of 74 00:04:04,600 --> 00:04:06,000 working. And then I thought, this is, 75 00:04:06,000 --> 00:04:08,760 this is a cool way of working. It's a cool approach and it's, 76 00:04:08,800 --> 00:04:12,440 and it's working for us. So I started making I, I made, I 77 00:04:12,440 --> 00:04:15,080 made like AI started doing presentations mostly internally 78 00:04:15,080 --> 00:04:18,120 as we hired people to explain this is the company that we are 79 00:04:18,120 --> 00:04:22,280 trying to be kind of. And then I did a talk at a 80 00:04:22,280 --> 00:04:24,000 conference and then that kind of went viral. 81 00:04:24,000 --> 00:04:26,600 So then people at in Spotify said, Henrik, you should record 82 00:04:26,600 --> 00:04:29,280 a video because we can't have you running around all over the 83 00:04:29,280 --> 00:04:31,440 place just doing talks. You know, that's not good use of 84 00:04:31,440 --> 00:04:32,640 your time. You should record a video 85 00:04:32,640 --> 00:04:35,440 instead. And so I did that. 86 00:04:35,440 --> 00:04:37,560 And ironically, people wanted even more talks after that. 87 00:04:38,440 --> 00:04:40,080 I thought it would be like, I made a video now. 88 00:04:40,080 --> 00:04:41,600 And so now I don't have to do this talk again. 89 00:04:42,760 --> 00:04:45,800 But then at some point, the video was called Spotify 90 00:04:45,800 --> 00:04:47,800 Engineering Culture, which is what it was about. 91 00:04:48,480 --> 00:04:51,280 But then someone at some point started using the term Spotify 92 00:04:51,280 --> 00:04:54,280 model to describe the approach. And then it kind of stuck. 93 00:04:54,720 --> 00:04:56,680 But no, I I didn't come up with that term. 94 00:04:57,160 --> 00:04:59,400 Yeah, it was. It was never an intent to create 95 00:04:59,400 --> 00:05:01,280 like a model. It was more it was just a case 96 00:05:01,280 --> 00:05:05,080 study. Interesting because I, I found 97 00:05:05,080 --> 00:05:09,160 through my career at multiple stops, like German banks who 98 00:05:09,640 --> 00:05:14,080 implemented the Spotify model through McKinsey and, and went 99 00:05:14,080 --> 00:05:18,200 with that. And I was like, why, what, what, 100 00:05:18,200 --> 00:05:21,800 how, why? Like what's the speech speech 101 00:05:21,800 --> 00:05:24,640 that's looking at that? Yeah, I've been quite fascinated 102 00:05:24,640 --> 00:05:27,480 just from a social perspective, Like, OK, so this is how models 103 00:05:27,480 --> 00:05:33,160 get created, like, OK. But then to something completely 104 00:05:33,160 --> 00:05:39,160 different you, is it fair to say you moved away from lean and 105 00:05:39,160 --> 00:05:42,800 Agile topics and moved into AI or or how is that connected? 106 00:05:43,720 --> 00:05:48,440 I would say I moved away from from coaching on agile to 107 00:05:48,440 --> 00:05:50,880 coaching and training agile. I still do a bit of training. 108 00:05:52,040 --> 00:05:53,680 I didn't move away from using agile. 109 00:05:53,840 --> 00:05:58,440 I use agile all the time, but it, it didn't I, I, I left that 110 00:05:58,440 --> 00:06:01,960 as a business, so to speak. When when the when the AI 111 00:06:01,960 --> 00:06:04,720 started becoming really useful, I basically found that a new 112 00:06:04,720 --> 00:06:08,600 favorite toy and. And how did that start? 113 00:06:08,600 --> 00:06:12,000 Like did you just woke up 11 morning and you were like, oh, 114 00:06:12,200 --> 00:06:15,240 AI coding agents? That's something that I want to 115 00:06:15,560 --> 00:06:19,200 sort of jump into. Yeah, it's, I've always been 116 00:06:19,200 --> 00:06:21,840 kind of interested in AI, studied a bit of AI in the in 117 00:06:21,840 --> 00:06:24,960 the 90s, like in artificial neural networks at the Royal 118 00:06:24,960 --> 00:06:27,640 Institute of Technology here. And I was like, this technology 119 00:06:27,640 --> 00:06:30,160 is really interesting. You know, someday it'd be fun to 120 00:06:30,160 --> 00:06:32,680 work with it, but of course it wasn't practical at the time. 121 00:06:32,680 --> 00:06:37,720 It wasn't really useful. And then and also when I worked 122 00:06:37,720 --> 00:06:40,160 with Minecraft, I did a lot of AI programming in the game, 123 00:06:40,560 --> 00:06:44,760 although it's not real AI, it's more rule based procedural, but 124 00:06:44,960 --> 00:06:48,400 it's still interesting the concept of creating the 125 00:06:49,840 --> 00:06:53,560 behaviour that looks kind of like it's intelligence, although 126 00:06:53,560 --> 00:06:57,720 through driven by software. But then ChatGPT came along and 127 00:06:57,720 --> 00:07:00,480 everyone started getting excited about the fact that you can chat 128 00:07:00,480 --> 00:07:02,280 with it, like just like you chat to a human. 129 00:07:03,080 --> 00:07:05,760 And I was really impressed. I never thought I'd experienced 130 00:07:05,760 --> 00:07:08,920 that in my life, that level of kind of skill and human language 131 00:07:09,480 --> 00:07:12,400 from software. But it wasn't very, it wasn't 132 00:07:12,400 --> 00:07:14,280 very useful in the beginning because it tended to hallucinate 133 00:07:14,280 --> 00:07:17,440 and things like that. But then when GPT 4 came, I 134 00:07:17,440 --> 00:07:20,680 started seeing a lot of, you know, research around like this 135 00:07:20,680 --> 00:07:23,120 is actually starting to become really useful and really, you 136 00:07:23,120 --> 00:07:25,920 know, really smart. So I decided to explore this. 137 00:07:25,920 --> 00:07:28,760 So I, I took a week off and just kind of went off to my cabin and 138 00:07:28,760 --> 00:07:32,200 just turned off the phone kind of and just started tinkering 139 00:07:32,280 --> 00:07:35,320 on, you know, all day and all evening with this weird 140 00:07:35,320 --> 00:07:38,280 technology and tried it for stuff, tried it from, you know, 141 00:07:38,280 --> 00:07:41,760 coding, tried it for reviewing articles for planning workshops. 142 00:07:41,760 --> 00:07:45,400 So probably about half of my, my, my experimentation was with 143 00:07:45,400 --> 00:07:46,800 coding and half of it was other stuff. 144 00:07:47,680 --> 00:07:49,040 But then I was really blown away. 145 00:07:49,040 --> 00:07:51,640 It become like a kind of, in a way a life changing moment 146 00:07:51,640 --> 00:07:53,240 because I started seeing the potential. 147 00:07:53,640 --> 00:07:55,360 It wasn't perfect, it had its flaws. 148 00:07:55,360 --> 00:07:58,840 But when I realized how to use it and I started discovering 149 00:07:59,200 --> 00:08:01,760 techniques that nowadays people would call prompt engineering or 150 00:08:01,760 --> 00:08:05,520 context engineering, then I was like, OK, this is, this is the 151 00:08:05,520 --> 00:08:07,440 most powerful tool I've ever seen and it's only in its 152 00:08:07,440 --> 00:08:09,040 infancy. So then. 153 00:08:09,160 --> 00:08:11,360 So then I decided just to work with it full time. 154 00:08:12,240 --> 00:08:16,080 What have you built during that time when you when you started 155 00:08:16,080 --> 00:08:18,680 playing around with it? So the, the first thing I built 156 00:08:18,680 --> 00:08:23,520 was actually a chat bot, which then became a platform for chat 157 00:08:23,520 --> 00:08:25,560 bots. The chaplet I built is called 158 00:08:25,560 --> 00:08:28,560 Egbert. Egbert is a bit of a, a persona 159 00:08:28,560 --> 00:08:31,520 that I invented, a sarcastic little kind of cartoon 160 00:08:31,520 --> 00:08:33,919 character. And I and I made a virtual 161 00:08:33,919 --> 00:08:36,640 Egbert and then I added him to my Minecraft server and to my 162 00:08:36,640 --> 00:08:39,760 Discord server with my friends. So Egbert would kind of comment 163 00:08:39,760 --> 00:08:41,840 on what's happened and what's happening inside our Minecraft 164 00:08:41,840 --> 00:08:44,360 server and on Discord. And then I kept adding stuff. 165 00:08:44,360 --> 00:08:47,000 So I started adding memories. Egbert would remember things so 166 00:08:47,000 --> 00:08:50,000 he can recall things using techniques that now people would 167 00:08:50,000 --> 00:08:52,120 call rag or retrieval augmented generation. 168 00:08:52,680 --> 00:08:55,000 And Egbert became my little, you know, platform for 169 00:08:55,000 --> 00:08:56,520 experimentation. And then a friend was like, Hey, 170 00:08:56,520 --> 00:08:59,320 I want to add another persona. So then I made the platform more 171 00:08:59,320 --> 00:09:02,480 generic and added the ability to add any kinds of of, of 172 00:09:03,320 --> 00:09:06,840 characters, which is, I guess that the embryo of what I work 173 00:09:06,840 --> 00:09:09,880 with now, which is a agents. But it was pretty fascinating to 174 00:09:09,880 --> 00:09:13,720 see like how kind of live this character seemed. 175 00:09:14,280 --> 00:09:16,440 You know, you would log into the Minecraft server and he would 176 00:09:16,440 --> 00:09:18,440 say some sarcastic comments like, oh, it's you again. 177 00:09:18,440 --> 00:09:20,080 Are are you going to fall into a hole today like you did 178 00:09:20,080 --> 00:09:21,360 yesterday? Right. 179 00:09:21,360 --> 00:09:24,040 And then and just had these long arguments with them. 180 00:09:24,880 --> 00:09:26,720 Very fun. But the, but then also during 181 00:09:26,720 --> 00:09:30,600 that whole experience, I used, I used a lot of AI for, for 182 00:09:30,600 --> 00:09:31,880 writing the code for that platform. 183 00:09:32,520 --> 00:09:35,800 And I was really blown away by the fact that I could write such 184 00:09:35,800 --> 00:09:39,440 a so, so like 10 times faster in terms of development. 185 00:09:40,680 --> 00:09:42,880 So just from a productivity perspective, I was like, whoa, 186 00:09:42,880 --> 00:09:44,640 this kind of changes the whole equation. 187 00:09:45,080 --> 00:09:49,120 Yeah, you're just jumping into the next question that I that I 188 00:09:49,120 --> 00:09:53,040 prepared and I was wanted to ask you where do you see the use 189 00:09:53,040 --> 00:09:57,000 cases in company beyond like let's say summarizing a bunch of 190 00:09:57,000 --> 00:10:00,360 emails, is that generating code? Is it? 191 00:10:00,440 --> 00:10:02,800 Is it that? I I think generating code is 192 00:10:02,800 --> 00:10:05,520 probably the the strongest use case right now. 193 00:10:05,520 --> 00:10:07,680 That's where the technology has come the furthest in terms of 194 00:10:07,680 --> 00:10:12,640 being like useful right now. But there's a lot of other use 195 00:10:12,640 --> 00:10:15,000 cases now that are that are kind of starting to reveal 196 00:10:15,000 --> 00:10:17,520 themselves. But, but even the basic things 197 00:10:17,520 --> 00:10:19,920 like summarizing a meeting is, is of course a very useful use 198 00:10:19,920 --> 00:10:23,080 case. But what we've noticed and the 199 00:10:23,080 --> 00:10:25,720 reason why I started the, the company I work for now 200 00:10:26,400 --> 00:10:30,120 abundantly is is because we noticed that AI becomes a lot 201 00:10:30,120 --> 00:10:32,920 more useful when you give it some level of autonomy and let 202 00:10:32,920 --> 00:10:35,560 it do stuff on its own and not just be stuck inside a chat. 203 00:10:36,560 --> 00:10:40,200 So example of use cases there. So we made a platform where you 204 00:10:40,200 --> 00:10:42,520 can build colleagues, essentially AI colleagues that 205 00:10:42,520 --> 00:10:45,360 have a job and a mission and tools and they can do do things 206 00:10:45,360 --> 00:10:48,400 on their own, but within, within a scope that you give them. 207 00:10:48,840 --> 00:10:52,480 So typical things they would do then is for example, screening 208 00:10:52,480 --> 00:10:54,600 an investment company we work with. 209 00:10:54,600 --> 00:10:56,880 They need to look at, they need to do research for thousands of 210 00:10:56,880 --> 00:10:59,480 companies every year. And this agent just takes that 211 00:10:59,480 --> 00:11:02,240 spreadsheet and does, you know, parallel research tracks for all 212 00:11:02,240 --> 00:11:05,440 1500 lines. And and it looks at the website, 213 00:11:05,440 --> 00:11:09,000 does research, digs online and does follow up research and then 214 00:11:09,000 --> 00:11:12,080 basically classifieds it into, yes, this company is, is 215 00:11:12,320 --> 00:11:13,520 interesting for you to invest in. 216 00:11:13,800 --> 00:11:15,200 This is a no, this one's a maybe. 217 00:11:15,880 --> 00:11:20,000 So those kind of things or things like a triaging support 218 00:11:20,000 --> 00:11:24,720 tickets or yeah, there's just so many different use cases where, 219 00:11:25,560 --> 00:11:28,320 you know, lower level work that, that kind of waste people's 220 00:11:28,320 --> 00:11:30,880 time. And by by having an agent do 221 00:11:30,880 --> 00:11:32,680 things like business intelligence research, right, 222 00:11:32,680 --> 00:11:35,240 then that frees up time for people to work kind of on higher 223 00:11:35,240 --> 00:11:37,720 level stuff. So that's, that's, that's 224 00:11:37,800 --> 00:11:40,160 there's a lot of use cases popping up around that. 225 00:11:40,160 --> 00:11:42,960 But what one of one of the simple ones that all companies 226 00:11:42,960 --> 00:11:46,360 need to find is basic research. So what, what, what are our 227 00:11:46,360 --> 00:11:47,840 competitors doing? Have there? 228 00:11:48,080 --> 00:11:51,520 Are there any new laws being, being being created within our 229 00:11:51,520 --> 00:11:53,520 space? I mean, any changes, any, any 230 00:11:53,520 --> 00:11:55,880 news events that that is relevant for me and my team. 231 00:11:56,280 --> 00:12:00,160 So go out, look for that, proactively alert us to what we 232 00:12:00,160 --> 00:12:02,360 need to know in a format that's useful to us. 233 00:12:02,360 --> 00:12:05,520 So not just a spammy newsletter, but very targeted like this is 234 00:12:05,520 --> 00:12:06,920 something that should be on your radar. 235 00:12:07,800 --> 00:12:11,600 Interesting. But maybe let's come back to to 236 00:12:11,720 --> 00:12:15,480 the AI coding. There were a bunch of case 237 00:12:15,480 --> 00:12:19,440 studies that showed you could achieve like like that. 238 00:12:19,440 --> 00:12:22,440 I think it was a get up case study that that published like 239 00:12:22,440 --> 00:12:26,800 you achieved 20% performance increase when you use coding 240 00:12:27,920 --> 00:12:31,320 like AI supported or AI guided coding. 241 00:12:31,840 --> 00:12:34,760 I think it was. That study is pretty old by now, 242 00:12:35,080 --> 00:12:37,800 but do do where do you think those performance indicators 243 00:12:37,800 --> 00:12:40,880 could go? So this is something it's, I'm 244 00:12:40,880 --> 00:12:43,840 kind of fascinated by it because since I'm, I'm living this now 245 00:12:43,840 --> 00:12:46,920 since the past two years coding everyday with AI support and 246 00:12:46,920 --> 00:12:51,120 before using AII did coding at a professional level on and off 247 00:12:51,120 --> 00:12:53,480 for about 30 years. So I have the comparison kind of 248 00:12:53,480 --> 00:12:56,480 before and after. And for me it's easily 10 times 249 00:12:56,480 --> 00:12:59,000 more effective, sometimes 20 times or even 100 times. 250 00:12:59,000 --> 00:13:02,040 It's, it's ridiculous on average. 251 00:13:02,040 --> 00:13:04,080 So in some specific cases, sometimes I have to do things 252 00:13:04,080 --> 00:13:07,600 myself or so worst case it's like it was before, but then 253 00:13:07,600 --> 00:13:10,120 suddenly I can do something that would have taken me a week and I 254 00:13:10,120 --> 00:13:14,160 do it in one hour. So, but then what I've noticed 255 00:13:14,160 --> 00:13:15,960 is that it's quite rare to have that effect. 256 00:13:15,960 --> 00:13:20,000 It's, it's only like most people I meet don't see that effect. 257 00:13:20,640 --> 00:13:23,840 And, and, and, and I've been thinking a lot about why. 258 00:13:23,960 --> 00:13:26,840 And I've worked with a bunch of teams with this. 259 00:13:27,640 --> 00:13:30,960 And what I, what I realized is you need a number of components 260 00:13:30,960 --> 00:13:32,720 to get that level of productivity improvement. 261 00:13:32,720 --> 00:13:36,200 You need a really good model. You need a really good tool. 262 00:13:36,840 --> 00:13:39,800 So by, for example, a really good model is clawed for Sonnet 263 00:13:39,800 --> 00:13:42,880 or clawed for Opus. A really good tool is cursor. 264 00:13:42,880 --> 00:13:44,600 I would say nothing compares with it right now. 265 00:13:44,600 --> 00:13:47,920 I can develop an environment and then you need really good 266 00:13:47,920 --> 00:13:50,480 skills, both like personally, how to use it. 267 00:13:50,480 --> 00:13:52,560 What in what situations do I use AI? 268 00:13:53,000 --> 00:13:56,440 In what situations do I trust the AI well, versus carefully 269 00:13:56,440 --> 00:14:00,160 review the code And in what situations do I give, do I give 270 00:14:00,160 --> 00:14:03,320 it a, a big job versus slice into tiny increments? 271 00:14:03,800 --> 00:14:06,720 And there's all this kind of skill effects. 272 00:14:06,720 --> 00:14:08,280 So how do I phrase a good prompt? 273 00:14:08,920 --> 00:14:11,560 What context do I give it? How do I what do I do when it's 274 00:14:11,560 --> 00:14:15,080 when it makes a mistake? So all that stuff is a skill you 275 00:14:15,080 --> 00:14:16,680 build up over time. But if you have really good 276 00:14:16,680 --> 00:14:20,200 models, really good tools and you've built up this this skill 277 00:14:20,200 --> 00:14:23,480 set, that's when you get the kind of radical improvements, 278 00:14:23,480 --> 00:14:26,240 which which makes the 20% improvement sound like really 279 00:14:26,240 --> 00:14:30,200 like what that's it's way. It's way beyond that. 280 00:14:30,200 --> 00:14:34,160 But I think if you're, if you're new to it, you're not really if 281 00:14:34,160 --> 00:14:36,040 you're not used to working in this way because it is a 282 00:14:36,040 --> 00:14:38,200 different way of working. So if you've been coding for a 283 00:14:38,200 --> 00:14:40,400 long time, coding with AI is different. 284 00:14:40,400 --> 00:14:44,520 It's a new way of thinking and until you to until you grasp 285 00:14:44,520 --> 00:14:47,600 that the improvements will still be there, but it'll be maybe in 286 00:14:47,600 --> 00:14:51,440 the 1020% range because it's just it's like having a junior 287 00:14:51,440 --> 00:14:53,240 assistant helping you with some very basic tasks. 288 00:14:53,280 --> 00:14:54,800 Yeah. Can you elaborate a little bit 289 00:14:54,800 --> 00:14:59,880 why why it's so different or? Well, for example, normally if 290 00:14:59,880 --> 00:15:03,120 I'm without AI help, if I'm going to solve a problem, I need 291 00:15:03,120 --> 00:15:04,880 to think about the problem is I'm going to solve and I need to 292 00:15:04,880 --> 00:15:06,280 think about how am I going to write the code. 293 00:15:07,040 --> 00:15:09,120 And I'm in my mind is very much on how am I going to get this 294 00:15:09,120 --> 00:15:12,320 code to work? Both AI, there's this meta 295 00:15:12,320 --> 00:15:16,440 thinking all the time, which is what kind of complexity is this 296 00:15:16,440 --> 00:15:18,600 problem? What kind of context would an AI 297 00:15:18,600 --> 00:15:20,280 need to solve it? So it's like an AI first 298 00:15:20,280 --> 00:15:22,840 thinking my, my default goal is AI should write every single 299 00:15:22,840 --> 00:15:25,080 line of this code. And then there's this reasoning 300 00:15:25,080 --> 00:15:28,760 happening all the time in my head, like a separate thread 301 00:15:29,120 --> 00:15:31,400 thinking about how much or how little do I do I give this 302 00:15:31,400 --> 00:15:33,320 person. So I would say you can compare 303 00:15:33,320 --> 00:15:38,160 it to coding yourself versus mentoring a new employee who 304 00:15:38,160 --> 00:15:39,760 joined your team. You're going to have a 305 00:15:39,760 --> 00:15:42,560 completely different mindset. And some you want that employee 306 00:15:42,560 --> 00:15:44,520 to write as much of the code as possible, but sometimes you need 307 00:15:44,520 --> 00:15:46,240 to help them, give them feedback. 308 00:15:46,400 --> 00:15:48,680 And sometimes you need to write the code yourself and explain to 309 00:15:48,680 --> 00:15:51,120 that employee. And if that employee makes the 310 00:15:51,120 --> 00:15:53,360 same mistake several times, you need to take a step back and 311 00:15:53,360 --> 00:15:54,760 say, what is that person lacking? 312 00:15:54,760 --> 00:15:56,240 What context do do they not have? 313 00:15:57,200 --> 00:15:59,760 Or is it, or is this kind of problem beyond their skills? 314 00:16:00,240 --> 00:16:02,760 So yeah, it's it's a different kind kind of mindset. 315 00:16:03,200 --> 00:16:05,920 You, you mentioned that you wanted to write 100% of that 316 00:16:05,920 --> 00:16:13,080 code that I mean, it sounds crazy to me, like I'm a hardware 317 00:16:13,080 --> 00:16:16,080 guy, probably I don't understand coding as well. 318 00:16:16,080 --> 00:16:19,640 But having this, I thought it would work in a way that you 319 00:16:19,640 --> 00:16:22,760 would prompt something and you would get a bunch of code and 320 00:16:22,760 --> 00:16:26,360 then you would like sort of debug it or you would like play 321 00:16:26,360 --> 00:16:29,880 with it and then push it to production. 322 00:16:29,880 --> 00:16:33,400 But you seem to work in a different way, like you would 323 00:16:34,080 --> 00:16:39,080 try to give it as good of a context that it generates 100%. 324 00:16:39,880 --> 00:16:45,080 Yeah, it's, it's like a it's a goal, but the goal is not always 325 00:16:45,080 --> 00:16:47,040 achieved. But by by having that as a goal, 326 00:16:47,040 --> 00:16:48,560 it forces me to think in a certain way. 327 00:16:48,960 --> 00:16:51,200 It forces me to break down the problem in a really good way. 328 00:16:51,760 --> 00:16:53,760 And it and that's where Agile comes into play, right? 329 00:16:53,760 --> 00:16:56,080 The ability to write a good user story, the ability to split it 330 00:16:56,080 --> 00:16:58,000 into small slices that each are testable. 331 00:16:58,400 --> 00:17:01,360 That skill gets comes into play when you're thinking about what 332 00:17:01,360 --> 00:17:03,160 would the AI need to solve this problem. 333 00:17:03,720 --> 00:17:07,200 But in practice, what we find is if you imagine a scale like a 334 00:17:07,240 --> 00:17:09,319 develop, I like to draw a picture of, I call it the 335 00:17:09,319 --> 00:17:15,079 developer laziness scale, right? So it's from 0% to 100%, so on, 336 00:17:15,079 --> 00:17:17,480 on the 100% and it's 100% laziness. 337 00:17:17,720 --> 00:17:20,240 So you, you write a prompt and then you go make some coffee 338 00:17:20,960 --> 00:17:22,599 and, and then when it's done, you just push to production. 339 00:17:22,599 --> 00:17:24,680 Don't even look at the code. So that's what people sometimes 340 00:17:24,680 --> 00:17:28,920 call vibe coding, right? But then moving left on a scale 341 00:17:28,920 --> 00:17:32,720 we have like let's say 80% laziness is a Reddit prompt. 342 00:17:32,880 --> 00:17:36,720 AI writes the code and then I just skim it high level and just 343 00:17:36,720 --> 00:17:38,920 see if anything looks weird. And then and then I push it. 344 00:17:39,720 --> 00:17:41,840 So AI writes. So basically AI does most of the 345 00:17:41,840 --> 00:17:43,880 stuff, but I'm at least aware of the code. 346 00:17:44,800 --> 00:17:49,320 And then 60% right would be, or like let's say 50% AI writes the 347 00:17:49,320 --> 00:17:53,440 code and I carefully read every line and, and, and almost always 348 00:17:53,440 --> 00:17:56,120 make a few tweaks. And then going more down to the 349 00:17:56,120 --> 00:18:01,840 down, down, down the scale, maybe 2030% would be AI writes 350 00:18:01,840 --> 00:18:04,200 the code, but I just use that to get some ideas. 351 00:18:04,200 --> 00:18:08,160 Then I write it myself. And then of course, the case 0% 352 00:18:08,160 --> 00:18:09,680 laziness. I write every line myself. 353 00:18:09,680 --> 00:18:11,880 And I, you know, I Google around, go to Reddit and, and, 354 00:18:11,880 --> 00:18:13,720 and stack Overflow to figure out things. 355 00:18:14,520 --> 00:18:18,960 And the key, the key skill I think now is as a developer is 356 00:18:18,960 --> 00:18:20,720 you need to master all parts of that scale. 357 00:18:21,440 --> 00:18:23,240 You need, you need to be able to write code yourself because 358 00:18:23,240 --> 00:18:25,080 sometimes AI will fail unless you're building a very simple 359 00:18:25,080 --> 00:18:26,160 product. But if you're building something 360 00:18:26,160 --> 00:18:28,600 complex or a platform, you need to understand what's going on. 361 00:18:28,600 --> 00:18:30,840 You need to be able to write the code yourself, even though it 362 00:18:30,840 --> 00:18:34,640 might take you 20 times longer, But you also need to be able to 363 00:18:35,000 --> 00:18:37,520 write, craft a good prompt. And, and the most important 364 00:18:37,520 --> 00:18:40,880 skill, I think is the meta level skill, which is determining 365 00:18:40,880 --> 00:18:42,960 which part of that scale I should be on right now. 366 00:18:43,800 --> 00:18:47,560 Does that, does that make sense? And, and I find it if, if you 367 00:18:47,560 --> 00:18:50,320 can do that, that that's when you really get, you get the 368 00:18:50,320 --> 00:18:51,480 best, you get the best everything. 369 00:18:51,480 --> 00:18:55,960 You get AI generating the code when it can, but you also don't 370 00:18:55,960 --> 00:18:58,280 get the technical debt, which you can get if you just do fire 371 00:18:58,280 --> 00:18:59,920 and forget and don't even check the code yourself. 372 00:18:59,920 --> 00:19:03,160 Yeah, I think it's really interesting because it reminds 373 00:19:03,160 --> 00:19:07,720 me of Q1 this year when all the tech CEOs claim that so and so 374 00:19:07,720 --> 00:19:10,000 percentage of their code is written by AI. 375 00:19:10,480 --> 00:19:13,160 I don't know if you, you, you follow the press there and I, 376 00:19:13,160 --> 00:19:17,720 but I think it was Microsoft and Satya started with like 30% of 377 00:19:17,720 --> 00:19:20,560 our code is written by AI. And then Mark Zuckerberg went 378 00:19:20,560 --> 00:19:24,560 down with like 50% of our code is written by AI And traffic CEO 379 00:19:24,560 --> 00:19:27,840 is like by the end of this year, we have 90% of AI mine. 380 00:19:28,080 --> 00:19:30,520 Is bigger than yours. Yeah, I'm just waiting for Elon 381 00:19:30,520 --> 00:19:33,080 Musk saying 200% of the code is written by AI. 382 00:19:33,520 --> 00:19:35,440 Yeah, The funny thing is, if you asked me, like how much of our 383 00:19:35,440 --> 00:19:37,720 code is written by AI, can't answer it because it's like if, 384 00:19:37,760 --> 00:19:40,280 if you're in a group of people doing mod programming, how much 385 00:19:40,280 --> 00:19:43,240 of the code was written by Joe? It's a collaboration, right? 386 00:19:43,880 --> 00:19:45,560 Even he wrote some code. I looked at it, I made some 387 00:19:45,560 --> 00:19:49,200 changes, I wrote some code. So, so AI to me is just like a 388 00:19:49,200 --> 00:19:50,520 colleague. It's it's a team member. 389 00:19:50,920 --> 00:19:53,240 And just like with any team member, you kind of split up the 390 00:19:53,240 --> 00:19:54,240 work. Sometimes you work together, 391 00:19:54,240 --> 00:19:55,600 sometimes you review each other's code. 392 00:19:55,920 --> 00:19:57,800 So it's, it's not even a, a relevant question. 393 00:19:57,800 --> 00:20:00,400 So I would say in my case, I would say how much is AI 394 00:20:00,400 --> 00:20:02,800 involved in the coding? I would say it's involved in 395 00:20:02,880 --> 00:20:05,040 probably 90% of all code in one way or another. 396 00:20:05,560 --> 00:20:07,000 But then how much is actually written? 397 00:20:07,000 --> 00:20:08,600 Well, that to be different, right? 398 00:20:08,800 --> 00:20:15,160 Case by case. Yeah, oh gosh, I I found a very 399 00:20:15,160 --> 00:20:18,400 stupid question that I got here on on my prep sheet. 400 00:20:18,640 --> 00:20:20,240 They're they're they're is. It possible? 401 00:20:21,200 --> 00:20:24,560 Is it possible to vibe code lovable at lovable? 402 00:20:26,840 --> 00:20:31,080 Wow, OK, here's my guess. I've used lovable quite a bit 403 00:20:31,800 --> 00:20:33,720 and my guess is no. OK. 404 00:20:34,160 --> 00:20:37,120 In fact I would say with certainty no, not yet. 405 00:20:37,800 --> 00:20:40,040 But in the future almost certainly yes. 406 00:20:40,160 --> 00:20:41,680 The question is how long that is a future. 407 00:20:41,680 --> 00:20:43,160 Don't know. OK. 408 00:20:44,720 --> 00:20:47,800 OK, a really interesting question, but yeah, lovable is 409 00:20:47,800 --> 00:20:51,600 is incredible, but it, but it's optimized for fairly simple web 410 00:20:51,600 --> 00:20:53,320 apps with a front end and a database. 411 00:20:53,320 --> 00:20:56,840 And for that it's it's amazing. But to build an actual platform, 412 00:20:57,440 --> 00:21:00,120 then I would say at least at the moment it's you really can't. 413 00:21:00,680 --> 00:21:05,640 OK, have you seen, Do You know the Pragmatic Engineer podcast 414 00:21:05,640 --> 00:21:08,400 or or newsletter? No. 415 00:21:08,800 --> 00:21:12,520 So it's a really, really cool newsletter from Gergley Auros. 416 00:21:12,800 --> 00:21:18,760 And in one of his interviews, he interviews Farhan Tawar from 417 00:21:18,760 --> 00:21:26,000 Shopify and they would go through the, the history of AI 418 00:21:26,000 --> 00:21:29,520 at Shopify. They were one of the, the beta 419 00:21:29,520 --> 00:21:32,600 testers from Microsoft Copilot even before it was called 420 00:21:32,640 --> 00:21:35,880 Microsoft Copilot. And they have some interesting 421 00:21:35,880 --> 00:21:39,800 things going on there. So, and, and I'm curious what, 422 00:21:39,800 --> 00:21:44,200 what you would say about that, because let's start with this 423 00:21:44,200 --> 00:21:49,280 one here. They use AI in interviewing 424 00:21:49,280 --> 00:21:53,760 processes like when they hire someone, regardless if if that's 425 00:21:53,760 --> 00:21:57,600 an engineer or it's a manager, they would have a coding 426 00:21:57,600 --> 00:22:02,080 interview because they claim hey with AI everyone should be able 427 00:22:02,080 --> 00:22:03,480 to code now. Right. 428 00:22:03,480 --> 00:22:05,440 They would, I I didn't understand they would use AI. 429 00:22:05,440 --> 00:22:09,000 How? They would give you a like a 430 00:22:09,000 --> 00:22:12,560 coding interview when when you apply a Shopify, even if it's 431 00:22:12,560 --> 00:22:15,360 for a senior management position where you're not involved in 432 00:22:15,360 --> 00:22:17,400 technical details. OK, so so they want to see if 433 00:22:17,400 --> 00:22:18,840 you're able to basically vibe code. 434 00:22:19,000 --> 00:22:22,080 Yes, correct. How do you feel about that? 435 00:22:24,360 --> 00:22:26,520 Both positive and negative at same time. 436 00:22:26,520 --> 00:22:28,800 I'll explain why. Umm I I think everyone should 437 00:22:28,800 --> 00:22:30,320 know how to buy code. At my company. 438 00:22:30,320 --> 00:22:32,080 Our CEO is coding in our platform. 439 00:22:32,080 --> 00:22:34,560 He's not a coder. I I set up an environment for 440 00:22:34,560 --> 00:22:35,960 him. So when he wants to make a 441 00:22:35,960 --> 00:22:38,720 change in our platform, mostly not customer facing stuff, it's 442 00:22:38,720 --> 00:22:41,920 mostly internal admin stuff. But he needs some, some graph or 443 00:22:41,920 --> 00:22:44,440 something that it's over usage of our platform. 444 00:22:44,680 --> 00:22:46,840 He doesn't come to us. Instead he just makes APR. 445 00:22:47,720 --> 00:22:50,080 And in the simple cases, we just, we just, you know, merge 446 00:22:50,080 --> 00:22:52,840 it sometimes we look carefully. It saves a lot of time for us 447 00:22:53,120 --> 00:22:55,760 and it also empowers him a lot. He doesn't have to put stuff in 448 00:22:55,760 --> 00:22:57,400 our backlog and hope we get around to it right. 449 00:22:58,160 --> 00:23:00,200 So it's, it's amazing. So so my my goal is that 450 00:23:00,200 --> 00:23:02,760 everyone at our company will have a set up so they can code 451 00:23:03,560 --> 00:23:06,040 but to during an interview I think doesn't make sense. 452 00:23:06,040 --> 00:23:08,000 What Why is it important that someone already knows how to 453 00:23:08,000 --> 00:23:09,600 vibe code before they even join your company? 454 00:23:09,600 --> 00:23:11,280 You're going to lose a lot of potential people that way. 455 00:23:11,840 --> 00:23:15,760 The idea is to. To you know, it's, it's great 456 00:23:15,760 --> 00:23:17,960 that you encourage that, but I wouldn't filter someone out 457 00:23:17,960 --> 00:23:19,160 because they haven't already done it. 458 00:23:19,240 --> 00:23:20,800 However, I would check their attitude. 459 00:23:20,800 --> 00:23:22,960 Are they willing to learn this? Are they interested? 460 00:23:23,520 --> 00:23:27,480 Got it, got it. So the the way I I understood it 461 00:23:27,480 --> 00:23:31,520 was that you would be able to optimize your work flows through 462 00:23:31,640 --> 00:23:38,040 AI and code yourself some little agent helpers that help you 463 00:23:38,040 --> 00:23:41,720 along doing your job. Yeah, yeah, I think you're kind 464 00:23:41,720 --> 00:23:43,880 of like everyone should know how to use Excel. 465 00:23:44,560 --> 00:23:46,960 Yeah, right. Yeah, and this is Excel, right? 466 00:23:46,960 --> 00:23:48,560 You should be able to use this as a basic tool. 467 00:23:48,560 --> 00:23:49,960 Like I need to do some calculation. 468 00:23:50,200 --> 00:23:52,800 Don't go talk to someone else. Set up a simple spreadsheet. 469 00:23:53,560 --> 00:23:55,000 If you're going to build something more advanced, you're 470 00:23:55,000 --> 00:23:56,960 still going to need someone to help you do it probably. 471 00:23:57,440 --> 00:23:59,880 But for simple things, knowing how to use Excel gives you 472 00:23:59,880 --> 00:24:01,880 superpowers. And I think in the same way now, 473 00:24:02,040 --> 00:24:04,800 taking at the next level, knowing how to use, knowing how 474 00:24:04,800 --> 00:24:08,280 to write code for simple tasks with AI help, it gives you 475 00:24:08,280 --> 00:24:11,840 superpowers. Maybe let's take a little bit to 476 00:24:11,840 --> 00:24:14,640 the hiring process, because another question that almost 477 00:24:14,640 --> 00:24:18,640 always pops up is whether do companies still need junior 478 00:24:18,640 --> 00:24:24,400 developers or can they be replaced by AI agents doing the 479 00:24:24,400 --> 00:24:27,600 coding? So I, I think the, the junior 480 00:24:27,600 --> 00:24:28,880 senior thing is really interesting. 481 00:24:29,880 --> 00:24:31,440 I think that question is timeless. 482 00:24:31,440 --> 00:24:33,360 It's always been like, have we have we ever needed junior 483 00:24:33,360 --> 00:24:35,800 developers? So I'm going to throw that back 484 00:24:35,800 --> 00:24:38,640 at you. Even without AI, why would you 485 00:24:38,640 --> 00:24:43,440 ever hire a junior developer? To get some if. 486 00:24:43,600 --> 00:24:47,720 You can hire a senior one What? To to get some cheap volume. 487 00:24:48,480 --> 00:24:49,920 Yeah, it could be. It's there's usually a number 488 00:24:49,920 --> 00:24:51,160 reason I feel people do it right. 489 00:24:51,160 --> 00:24:53,440 One is that, well, you know, your senior developers are going 490 00:24:53,440 --> 00:24:55,560 to die off at some point. You need to get some younger 491 00:24:55,560 --> 00:24:58,080 people in also. They come in with some fresh 492 00:24:58,080 --> 00:24:59,760 perspectives. They tend to be faster at 493 00:24:59,760 --> 00:25:02,960 learning new things, a little more curious sometimes. 494 00:25:03,120 --> 00:25:05,320 Sometimes senior developers are really productive, but kind of 495 00:25:05,320 --> 00:25:06,560 stuck in their ways a little bit. 496 00:25:07,360 --> 00:25:09,320 There's all kinds of reasons that even as a senior developer, 497 00:25:09,320 --> 00:25:11,400 you will improve your skills by helping a junior. 498 00:25:11,920 --> 00:25:14,360 All those things still still apply with or without AI. 499 00:25:15,000 --> 00:25:17,320 So I think the, the key thing is not whether it's junior or 500 00:25:17,320 --> 00:25:19,560 senior, I would say, I would make sure that everyone that's 501 00:25:19,560 --> 00:25:23,320 on my team, regardless of skill level, are, are using AI as part 502 00:25:23,320 --> 00:25:26,000 of their toolkit. And and that's, that's the 503 00:25:26,000 --> 00:25:27,800 important part, but they might use them in different, in 504 00:25:27,800 --> 00:25:29,840 different ways. So I'd rather have a junior 505 00:25:29,840 --> 00:25:33,840 developer who really leans into using AI help versus a senior 506 00:25:33,840 --> 00:25:36,440 developer who doesn't. However, ideally I would like to 507 00:25:36,440 --> 00:25:38,560 have both because they bring different things to the table 508 00:25:38,560 --> 00:25:41,840 just like they always have. So yeah, I'm, I get a little bit 509 00:25:41,840 --> 00:25:44,440 concerned when people are like, yeah, we don't, you know, we, 510 00:25:44,640 --> 00:25:47,400 you know, we, we, we, we don't have any, like junior developers 511 00:25:47,400 --> 00:25:50,240 won't have any job. Maybe that's going to be true. 512 00:25:50,240 --> 00:25:52,200 I don't, I don't know. But I really, I really hope not. 513 00:25:52,200 --> 00:25:54,920 So I think it's that would be really sad. 514 00:25:56,200 --> 00:25:58,880 I mean, in the in the last quarter we saw a bunch of tech 515 00:25:58,880 --> 00:26:03,760 layoffs and positions that didn't get filled, filled and 516 00:26:05,680 --> 00:26:10,520 less jobs being put out on the market for developers. 517 00:26:10,800 --> 00:26:14,560 So the times were were really rough and everyone was claiming 518 00:26:14,560 --> 00:26:18,840 like, hey, this is the AI coding is the reason for that 519 00:26:18,840 --> 00:26:21,400 development. And looking at the numbers, now 520 00:26:21,400 --> 00:26:24,600 that they publish, there is a slight uptrend in the 521 00:26:24,600 --> 00:26:30,040 Magnificent 7 and they're hiring more and more engineers, not a 522 00:26:30,400 --> 00:26:33,440 not big numbers, but the the trend is going back up again. 523 00:26:33,720 --> 00:26:35,320 Yeah, Yeah, some thoughts about that. 524 00:26:35,320 --> 00:26:38,040 One is it's easy to make the, you know, mistake of causation 525 00:26:38,040 --> 00:26:41,320 versus correlation, right? You know, is it because of AI or 526 00:26:41,320 --> 00:26:43,040 not? It's hard to know for sure. 527 00:26:44,040 --> 00:26:47,360 But if you look historically, what AI does, it gives as a 528 00:26:47,360 --> 00:26:50,040 developer, as a junior developer, I would say as long 529 00:26:50,040 --> 00:26:54,600 as you don't define your job as writing lines of code, then 530 00:26:54,600 --> 00:26:56,480 you're in a probably then you're probably fine. 531 00:26:56,480 --> 00:26:58,560 But but if you define your job as the person who types the 532 00:26:58,560 --> 00:27:00,560 lines of code, then yes, that job is gone. 533 00:27:01,200 --> 00:27:05,920 It's disappearing very quickly. But historically, we invent the 534 00:27:05,920 --> 00:27:08,560 technologies that have, you know, vastly improved the 535 00:27:08,560 --> 00:27:11,800 productivity of developers by going from writing machine code, 536 00:27:11,800 --> 00:27:13,840 assembly code to writing higher level languages. 537 00:27:14,120 --> 00:27:17,280 And now with modern ID ES that do code completion, you know, 538 00:27:17,400 --> 00:27:20,040 one shortcut, refactor and read something through the whole code 539 00:27:20,040 --> 00:27:25,080 base, automated testing, even before AI, there, there are 540 00:27:25,080 --> 00:27:27,720 several step changes in, in the software development community 541 00:27:27,720 --> 00:27:29,960 that made everyone like, I would say at least twice as 542 00:27:29,960 --> 00:27:33,640 productive, but maybe even more. So you'd think that, OK, so now 543 00:27:33,640 --> 00:27:36,120 that developers are twice as productive, we only need, you 544 00:27:36,120 --> 00:27:38,480 know, have as many developers. Now we can fire all the juniors. 545 00:27:38,480 --> 00:27:40,440 And why do we need them? Because we have these modern 546 00:27:40,440 --> 00:27:43,000 development environments, but the effect has been more 547 00:27:43,000 --> 00:27:44,480 developers. The software industry has always 548 00:27:44,480 --> 00:27:46,840 been growing. So, so, so in that sense, from a 549 00:27:46,840 --> 00:27:50,040 macro perspective, I don't, I would, I would be surprised if 550 00:27:50,040 --> 00:27:53,280 this is the end of development. I think that trend is going to 551 00:27:53,280 --> 00:27:54,680 continue. We made people more productive. 552 00:27:54,680 --> 00:27:57,120 There's going to be more people doing development than ever. 553 00:27:57,600 --> 00:28:00,200 However, maybe not many people that have that as their defined 554 00:28:00,200 --> 00:28:02,680 specific job. And instead it's everyone is 555 00:28:02,680 --> 00:28:05,400 coding. And what real kind of 556 00:28:05,400 --> 00:28:07,920 professional coders do is not focus so much on writing the 557 00:28:07,920 --> 00:28:09,440 code because AI is better than that. 558 00:28:09,760 --> 00:28:12,320 In the same sense that, you know, in the past, you need to 559 00:28:12,320 --> 00:28:14,640 worry very, very much about allocating memory, about garbage 560 00:28:14,640 --> 00:28:16,560 collection. That was an important part of 561 00:28:16,560 --> 00:28:18,760 your skill. But now compilers do that better 562 00:28:18,760 --> 00:28:20,720 than humans. So you don't work, you work at a 563 00:28:20,720 --> 00:28:22,640 higher level. So I think as long as you as a 564 00:28:22,640 --> 00:28:26,240 developer are thinking in terms of I'm a problem solver, I solve 565 00:28:26,240 --> 00:28:31,360 problems with software and AI is 1 tool in doing that, then then 566 00:28:31,360 --> 00:28:33,680 then you definitely have a lot of value to to bring to bring to 567 00:28:33,680 --> 00:28:36,720 the table. And part of it is, I think what 568 00:28:36,720 --> 00:28:40,880 you mention is, is pretty well summarizing the Jevons paradox. 569 00:28:42,520 --> 00:28:46,040 What's that? That is, when something takes 570 00:28:46,040 --> 00:28:49,840 less resource to create, there is a higher demand for that. 571 00:28:51,000 --> 00:28:53,120 Right. Yeah, that's probably part of 572 00:28:53,120 --> 00:28:57,720 it. Another technical question 573 00:28:57,720 --> 00:29:02,720 maybe, and you have to explain it to me like I'm a 5 year old 574 00:29:03,880 --> 00:29:09,720 because as I said, I'm a hardware guy just strolling 575 00:29:09,720 --> 00:29:13,240 around in a software world. What about architecture? 576 00:29:13,840 --> 00:29:17,720 It it seemed to me that some, at some point the AI is decided 577 00:29:17,720 --> 00:29:21,480 making architectural decisions for you and overhauling them. 578 00:29:21,480 --> 00:29:26,360 Whenever it I think it is the correct word for that, it thinks 579 00:29:26,360 --> 00:29:29,200 it it should be overhauled. So so how is it? 580 00:29:29,200 --> 00:29:32,040 How is it possible to not own the architectural plan behind it 581 00:29:32,240 --> 00:29:34,200 or or is there a way to do it? Yeah. 582 00:29:34,200 --> 00:29:38,200 I, I think what changes at least the how things are now, I find 583 00:29:38,200 --> 00:29:42,600 that in, in like for example, for us, we are in 100% control 584 00:29:42,600 --> 00:29:44,240 of our architecture. We're not going to outsource 585 00:29:44,240 --> 00:29:47,160 that to anybody because that's a core part of our business. 586 00:29:47,960 --> 00:29:50,800 So what what happens now is we spend a lot more time in 587 00:29:50,800 --> 00:29:53,280 architecture than we ever would have done in the past because we 588 00:29:53,280 --> 00:29:56,280 have time for it, because we're not as much stuck in the weeds 589 00:29:56,280 --> 00:29:57,720 with, with getting the code to work. 590 00:29:59,000 --> 00:30:01,440 But I think it's a really a really important part of the 591 00:30:01,440 --> 00:30:05,120 decision as as a developer is deciding what part of the 592 00:30:05,120 --> 00:30:08,560 architecture do you care about? And then you pay attention to 593 00:30:08,560 --> 00:30:11,360 that and now you have the option to outsource some part of the 594 00:30:11,360 --> 00:30:14,280 architecture. For example, maybe I haven't, 595 00:30:14,280 --> 00:30:16,920 let's say I have an admin interface, an admin web. 596 00:30:17,720 --> 00:30:20,840 Where would you use internally to kind of follow up on things 597 00:30:20,920 --> 00:30:23,960 on what's going on? And that's maybe not a crucial 598 00:30:23,960 --> 00:30:25,800 part of our business. It's not customer facing. 599 00:30:26,440 --> 00:30:30,040 And if an AI, if I can use an AI to, to build that in a day, the 600 00:30:30,040 --> 00:30:35,080 whole thing, then as long as it works fine, maybe I've, I, I may 601 00:30:35,080 --> 00:30:36,840 decide that I don't care about that architecture. 602 00:30:37,320 --> 00:30:39,640 I can look at a high level, OK, it's a front end, it's a back 603 00:30:39,640 --> 00:30:40,720 end. It's a react thing. 604 00:30:40,720 --> 00:30:43,680 Sure, it's using a few frameworks, but it's working. 605 00:30:43,720 --> 00:30:47,000 And I don't really care why or how that's a choice you can 606 00:30:47,000 --> 00:30:48,480 make. It's not always the right 607 00:30:48,480 --> 00:30:50,080 choice, but it's also not always the wrong choice. 608 00:30:50,080 --> 00:30:53,400 I think it's the key thing is to kind of think of it as you 609 00:30:53,400 --> 00:30:56,160 decide which parts of the system that that you want to be in 610 00:30:56,160 --> 00:30:58,720 control of and it's in it's a it's a continuum. 611 00:30:58,720 --> 00:31:00,920 It's not like either you control it or the AI control. 612 00:31:00,920 --> 00:31:03,920 It's, it's you decide how much. So maybe it's this. 613 00:31:03,920 --> 00:31:07,400 I complete, I control completely this I control control mostly 614 00:31:07,400 --> 00:31:10,120 myself, but I use a lot of AI help that part. 615 00:31:10,120 --> 00:31:12,200 The AI does most of it, but I'm at least involved. 616 00:31:12,200 --> 00:31:14,760 I'm I'm at least aware of the architecture and this part of 617 00:31:14,760 --> 00:31:16,400 the black box. I don't care as long as it 618 00:31:16,400 --> 00:31:17,480 works. I don't even want to look at the 619 00:31:17,480 --> 00:31:20,480 code or the architecture. But of course, over time, this 620 00:31:20,480 --> 00:31:24,320 is shifting because as AI improves in capability and it is 621 00:31:24,320 --> 00:31:26,560 improving at pretty much an exponential rate, it's, it's 622 00:31:26,560 --> 00:31:28,920 quite mind boggling. There are things I can do now 623 00:31:28,920 --> 00:31:30,600 that I wouldn't dreamed of doing for, for a year ago. 624 00:31:30,600 --> 00:31:32,920 So it's kind of like you bought this, you, you hired an 625 00:31:32,920 --> 00:31:35,560 employee, but that employee is getting smarter all the time. 626 00:31:36,160 --> 00:31:38,360 And as they get smarter, you can give them a responsibility. 627 00:31:38,720 --> 00:31:43,160 So that's also part of it. I feel a very similar question 628 00:31:43,160 --> 00:31:50,400 is regarding do you run with rather micro services to create 629 00:31:50,400 --> 00:31:53,760 a more monolithic structure in your product? 630 00:31:53,760 --> 00:31:57,360 And when I first encountered like, OK, there are AI coding 631 00:31:57,360 --> 00:32:02,280 agents now, so I thought that we would lean towards like the 632 00:32:02,280 --> 00:32:07,120 throwaway monolith and just vibe coded whatever we want from 633 00:32:07,120 --> 00:32:09,360 scratch. Yeah, this is really 634 00:32:09,360 --> 00:32:11,880 interesting. And I, I think that really is 635 00:32:11,880 --> 00:32:13,280 one of the key parts of the decision. 636 00:32:13,280 --> 00:32:15,480 Do I care about this code, right. 637 00:32:15,520 --> 00:32:19,240 And, and I think, I think there's a very clear complexity 638 00:32:19,240 --> 00:32:21,320 ceiling for how, how much you can get away with that. 639 00:32:22,080 --> 00:32:24,480 And for example, at the moment, I would say if I need to build a 640 00:32:24,480 --> 00:32:27,200 carpool sharing app or something, right? 641 00:32:27,600 --> 00:32:31,120 Or if I want to build something like, yeah, a meeting 642 00:32:31,120 --> 00:32:33,520 coordinator, find the right time for everybody based on 643 00:32:33,520 --> 00:32:36,360 everyone's schedules, fairly straightforward app, right? 644 00:32:38,000 --> 00:32:39,960 It may, it may, it may be it would have taken a few weeks to 645 00:32:39,960 --> 00:32:44,560 do manually, but with lovable or cursor, it would be maybe, you 646 00:32:44,560 --> 00:32:48,160 know, an hour to get the first draft up and then maybe a day or 647 00:32:48,160 --> 00:32:50,960 two to get to really work. Well in that kind of context, I 648 00:32:50,960 --> 00:32:53,160 would I would consider it throwaway because it took one or 649 00:32:53,160 --> 00:32:55,040 two days to build. So I can make a new one, you 650 00:32:55,040 --> 00:32:56,920 know, next month if I want. So the code doesn't really 651 00:32:56,920 --> 00:32:59,960 matter. But when you get beyond a 652 00:32:59,960 --> 00:33:02,720 certain threshold, it quickly turns into this is actually code 653 00:33:02,720 --> 00:33:05,480 I need to maintain. And then architecture matters a 654 00:33:05,480 --> 00:33:07,680 lot. And it pitfall that a lot of 655 00:33:07,680 --> 00:33:10,120 people fall into is not knowing that. 656 00:33:10,680 --> 00:33:12,840 And if and if you don't, if you don't have development 657 00:33:12,840 --> 00:33:15,040 experience, you really can't know that. 658 00:33:15,040 --> 00:33:16,920 It's, it's hard to know where that limit is. 659 00:33:17,280 --> 00:33:20,160 But with development experience, you, you have a, a feeling for 660 00:33:20,160 --> 00:33:22,200 it. Like, OK, at this point, I do 661 00:33:22,200 --> 00:33:23,560 need to do, need to care about architecture. 662 00:33:24,320 --> 00:33:27,240 And then it's easy to run into this problem where AI gets it to 663 00:33:27,240 --> 00:33:31,120 work, but it keeps adding to your pile of, of technical debt. 664 00:33:31,120 --> 00:33:33,480 It keeps adding duct tape and stapling stuff. 665 00:33:33,480 --> 00:33:36,360 And and then you finally get to this product where, oh, wait a 666 00:33:36,360 --> 00:33:39,240 SEC before I can add a new feature and 10 minutes of vibe 667 00:33:39,240 --> 00:33:41,600 coding. Now it's going to be, you know, 668 00:33:41,680 --> 00:33:44,560 3 hours of swearing because AI keeps making a new mistake. 669 00:33:45,120 --> 00:33:47,920 And it's not because AI isn't competent, it's because you've 670 00:33:47,920 --> 00:33:49,840 built the mess and even a human would have the exact same 671 00:33:49,840 --> 00:33:52,440 problem. So that that's really important. 672 00:33:52,440 --> 00:33:54,880 So, So what what I find is like in our case, we are a tech 673 00:33:54,880 --> 00:33:57,840 company and we have a platform that is part, you know, a core 674 00:33:57,840 --> 00:33:59,920 part of our business. So quality matters are locked 675 00:34:00,520 --> 00:34:03,320 and there I find that it's really the same principles as 676 00:34:03,320 --> 00:34:06,560 always as before, you know, modular architecture, clear 677 00:34:06,560 --> 00:34:10,520 interfaces, minimizing, you know, a coupling and and 678 00:34:10,560 --> 00:34:12,639 thinking about dependencies. All those principles are, are 679 00:34:12,639 --> 00:34:15,520 the same and maybe even more important than ever, because if 680 00:34:15,520 --> 00:34:19,560 you have a solid architecture nowadays, the, the best AI 681 00:34:19,560 --> 00:34:21,960 models will follow that architecture in a really good 682 00:34:21,960 --> 00:34:24,400 way. In the past, a year ago, they 683 00:34:24,400 --> 00:34:26,000 would tend to just add stuff on top. 684 00:34:26,440 --> 00:34:28,800 So your, your architecture would kind of deteriorate. 685 00:34:29,080 --> 00:34:31,600 But now I find that it, it, it keeps the architecture and even 686 00:34:31,600 --> 00:34:34,480 sometimes improves it. But if it, but if, but if you 687 00:34:34,480 --> 00:34:37,600 have a monolithic mess from the beginning, it's really hard for 688 00:34:37,639 --> 00:34:39,360 AI. It's possible to change it, but 689 00:34:39,360 --> 00:34:40,639 it's not going to do it by itself. 690 00:34:40,639 --> 00:34:43,120 It's going to tend to just add more stuff on top, very similar 691 00:34:43,120 --> 00:34:46,360 to what humans do, right. So I'm going to find now that 692 00:34:46,360 --> 00:34:49,280 that kind of architectural patterns that seem to be best 693 00:34:49,280 --> 00:34:54,560 for AI power development is Mona, Mona repos. 694 00:34:55,239 --> 00:34:58,760 So we recently went from a, from a kind of a micro service 695 00:34:58,760 --> 00:35:01,880 architecture into from, from a deployment perspective, it still 696 00:35:01,880 --> 00:35:04,440 is micro service, but in the coding environment, it's, it's a 697 00:35:04,440 --> 00:35:06,960 Mona repo because then it's a lot easier for the AI to see the 698 00:35:06,960 --> 00:35:10,200 big picture. And it, it really improved our 699 00:35:10,200 --> 00:35:12,600 productivity a lot. When I could ask right one 700 00:35:12,600 --> 00:35:15,400 prompt that say, make this change and it examines the front 701 00:35:15,400 --> 00:35:18,480 end and the back end answers a answer is being and goes, OK, 702 00:35:18,640 --> 00:35:20,320 here's all the changes we make in all the places. 703 00:35:20,720 --> 00:35:22,800 Test everything and and then it's ready to go. 704 00:35:23,200 --> 00:35:25,080 Well, in the past, I'd have to jump around between different 705 00:35:25,080 --> 00:35:27,440 repos and it'd be a lot more, a lot more work. 706 00:35:28,000 --> 00:35:31,120 So, yeah, the, the, the good old architectural principles that 707 00:35:31,120 --> 00:35:32,760 we've always, you know, the timeless architectural 708 00:35:32,760 --> 00:35:35,080 principles are still relevant, more relevant than ever. 709 00:35:35,480 --> 00:35:37,720 The one thing that maybe changes a little bit is that it becomes 710 00:35:37,720 --> 00:35:40,360 even more useful to have things like a Mona repo. 711 00:35:40,720 --> 00:35:46,800 Yeah, having that, aren't you afraid of let's say so there was 712 00:35:46,800 --> 00:35:51,520 this case about from from Jason Lemkin, who who runs a a company 713 00:35:52,280 --> 00:35:55,360 solely built on vibe coding. He used Replit for it. 714 00:35:55,360 --> 00:35:58,400 I don't know if you heard about that case, but during a code 715 00:35:58,400 --> 00:36:03,840 freeze that that Replit AI deleted the whole database that 716 00:36:03,840 --> 00:36:07,440 was in production. Yeah, I didn't hear. 717 00:36:07,680 --> 00:36:10,000 I'm just scared that those things are going to happen. 718 00:36:10,920 --> 00:36:14,240 Probably yes. It happens also with employees, 719 00:36:14,240 --> 00:36:16,400 right? It it's, it's a, it's a thing 720 00:36:16,400 --> 00:36:17,960 that sometimes happens. People make mistakes. 721 00:36:18,400 --> 00:36:21,960 A is they, I find they're less likely to make mistakes, but but 722 00:36:21,960 --> 00:36:24,200 they can be given a context that pushes them in the wrong 723 00:36:24,200 --> 00:36:26,920 direction. Or if you, if you use a cheap 724 00:36:26,920 --> 00:36:29,080 model, then it may just make dumb mistakes. 725 00:36:30,440 --> 00:36:33,920 So part of the the work, and this is, I guess, an example of 726 00:36:33,920 --> 00:36:36,480 why it's different, right? It's you thinking about, I just 727 00:36:36,480 --> 00:36:39,240 got this intern right. What environment should I give 728 00:36:39,240 --> 00:36:41,360 the intern? Should I give the intern the key 729 00:36:41,360 --> 00:36:43,920 to our production database with write access? 730 00:36:43,920 --> 00:36:45,040 Should they be able to delete it? 731 00:36:45,600 --> 00:36:48,160 And I would say, if you hire an intern and that intern deletes 732 00:36:48,160 --> 00:36:50,200 the production database, whose fault is it? 733 00:36:51,560 --> 00:36:54,720 Is it really the intern? What do you think of? 734 00:36:55,760 --> 00:36:58,680 Course not. Not someone decided to give the 735 00:36:58,680 --> 00:37:01,200 intern the ability to erase the production database. 736 00:37:01,200 --> 00:37:03,880 Why would they do that, right? You didn't create a safe space 737 00:37:03,880 --> 00:37:05,200 for that intern to make mistakes. 738 00:37:05,920 --> 00:37:07,840 So I think it's really important to have the same mindset. 739 00:37:07,960 --> 00:37:10,240 And the way I think of it is there are there are reverse 740 00:37:10,240 --> 00:37:12,400 there, there are mistakes that are reversible and mistakes that 741 00:37:12,400 --> 00:37:15,200 are irreversible. Really think about that with 742 00:37:15,200 --> 00:37:18,280 your AI setup. So making code changes is 743 00:37:18,280 --> 00:37:20,680 reversible almost always if you have version control system, 744 00:37:20,680 --> 00:37:23,280 which you do, right? So it's fine if it erased all 745 00:37:23,280 --> 00:37:25,400 the code, you just undo, right, Revert. 746 00:37:26,120 --> 00:37:29,440 So that's pretty safe. But if the tools like Cursor can 747 00:37:29,440 --> 00:37:33,680 also run shell commands, so they can do stuff like get commands 748 00:37:33,680 --> 00:37:35,080 and stuff and get commands are pretty safe too. 749 00:37:35,080 --> 00:37:36,400 They're mostly reversible as well. 750 00:37:38,000 --> 00:37:41,880 But if it's going to start, you know, installing stuff in your 751 00:37:41,880 --> 00:37:44,520 operating system, then it might break something. 752 00:37:45,720 --> 00:37:48,480 So yeah, you need to think about that. 753 00:37:49,000 --> 00:37:53,560 I, I would never give my cursor installation production in 754 00:37:53,560 --> 00:37:54,760 access to the production database. 755 00:37:54,960 --> 00:37:57,920 It shouldn't need it, so I would say that would be a a mistake as 756 00:37:57,920 --> 00:38:00,000 a user to give it stuff. I I I don't think it would 757 00:38:00,000 --> 00:38:02,480 accidentally erase it. You'd have to go quite far for 758 00:38:02,480 --> 00:38:04,720 that to happen. But the key thing is why even 759 00:38:04,720 --> 00:38:06,120 take that risk? It doesn't need access to the 760 00:38:06,120 --> 00:38:07,920 production database, at least not write access. 761 00:38:08,280 --> 00:38:13,040 Yeah. All right, I'm, I'm wondering 762 00:38:13,040 --> 00:38:16,200 like you're, you've been into the space for quite some time. 763 00:38:16,240 --> 00:38:18,400 How do you, how do you keep up with all the trends? 764 00:38:18,480 --> 00:38:22,640 Like I'm I'm interviewing you now but I seem like everything 765 00:38:22,640 --> 00:38:25,120 seems to happen so fast that we could do another interview 766 00:38:25,120 --> 00:38:28,680 within 1/4 probably and have a bunch of new topics over. 767 00:38:29,320 --> 00:38:33,680 Yeah, it's, it's crazy the rate of of like technical 768 00:38:33,760 --> 00:38:36,640 improvements. But personally, I just given up 769 00:38:36,640 --> 00:38:41,120 on keeping up kind of even even like a couple of years ago. 770 00:38:41,600 --> 00:38:43,880 I'll just conclude that, you know what, I'm going to stop 771 00:38:43,880 --> 00:38:47,000 trying to keep up because it's just stressful and it kind of 772 00:38:47,000 --> 00:38:49,120 impossible. So instead, there are, there are 773 00:38:49,120 --> 00:38:52,160 a few things that I kind of try to keep an eye on. 774 00:38:52,160 --> 00:38:53,520 I think that's what people need to do. 775 00:38:53,520 --> 00:38:56,240 They need to decide what are the, what, what subset of this 776 00:38:56,240 --> 00:39:00,200 space am I interested in? So in my case, you know, the, 777 00:39:00,280 --> 00:39:04,040 the, the foundational models clawed and, and, and, you know, 778 00:39:04,040 --> 00:39:08,200 open AI, you know, when a new model comes, I will, you know, 779 00:39:08,560 --> 00:39:11,800 look into that. Improvements to development 780 00:39:11,800 --> 00:39:13,560 tools such as cursor. I'll definitely look into that. 781 00:39:13,920 --> 00:39:16,800 Or new patterns for agent prompting. 782 00:39:18,000 --> 00:39:21,800 But, you know, then a new video tool comes along, a new video 783 00:39:21,800 --> 00:39:23,960 video generation tool comes along or some new, you know, 784 00:39:24,560 --> 00:39:28,040 some new platform. Yeah, I just accept that I won't 785 00:39:28,040 --> 00:39:30,400 know what's going on and then hope that somebody's going to 786 00:39:30,400 --> 00:39:33,040 poke, poke at me and say, Henrik, you should check this 787 00:39:33,040 --> 00:39:35,560 out. Maybe it'll hit my radar. 788 00:39:35,920 --> 00:39:38,640 But yeah, there is a continuous sense of FOMO a little bit that 789 00:39:38,640 --> 00:39:41,880 I think anybody working in this space has to just get used to. 790 00:39:42,720 --> 00:39:46,840 Yeah, probably. When do you think all of this 791 00:39:46,840 --> 00:39:51,080 will with HAT, let's let's say maybe focus on on AI assisted 792 00:39:51,080 --> 00:39:55,000 coding or or AI coding. Where do you think will this HAT 793 00:39:55,000 --> 00:39:57,880 within the next 12 months what we're going to see? 794 00:39:59,080 --> 00:40:02,760 So just looking at the trend. It seems that they're just 795 00:40:02,760 --> 00:40:05,080 improving all the time and there's 2 levels. 796 00:40:05,080 --> 00:40:08,240 There's, I guess there's three things that are improving all at 797 00:40:08,240 --> 00:40:10,840 the same time. 1 is the foundational models themselves, 798 00:40:11,520 --> 00:40:15,640 such as Claw, GPT, etcetera. The second is the tools that we 799 00:40:15,640 --> 00:40:19,240 use to access them, For example, Cursor, which is a wrapper, a 800 00:40:19,240 --> 00:40:21,120 product that sits on top of foundational models. 801 00:40:21,120 --> 00:40:23,800 So it's a code editor, but there's a lot of prompts in 802 00:40:23,800 --> 00:40:27,440 there and a lot of kind of magic going on where the developers of 803 00:40:27,440 --> 00:40:31,560 that tool are, are doing things to make AI aware of your whole 804 00:40:31,560 --> 00:40:35,800 code base, which is magical, but, but you need both. 805 00:40:35,800 --> 00:40:38,440 So if you just go to Claude and copy paste some code or go to 806 00:40:38,440 --> 00:40:41,680 GPT and copy, copy, paste code, you get a very limited 807 00:40:41,680 --> 00:40:44,400 productivity improvement. But using a tool like Cursor, it 808 00:40:44,400 --> 00:40:46,440 just kind of understands your whole code base in some weird 809 00:40:46,440 --> 00:40:49,200 magical way. And that's of course going to 810 00:40:49,200 --> 00:40:51,600 keep improving as well. Cursor gets updated every week 811 00:40:51,880 --> 00:40:54,000 and there's of course Cursor competitors that are starting to 812 00:40:54,000 --> 00:40:56,480 catch up gradually. And the third thing that's 813 00:40:56,480 --> 00:40:59,720 improving is, is the engineer's ability to understand how and 814 00:40:59,720 --> 00:41:03,800 when to use these tools. So I think as, as, as a just 815 00:41:03,800 --> 00:41:06,360 looking forward, I think it's just going to be AI is going to 816 00:41:06,360 --> 00:41:08,800 be able to take on more and more complex coding tasks. 817 00:41:09,160 --> 00:41:11,240 We're going to be able to give them bigger and bigger jobs. 818 00:41:11,640 --> 00:41:15,120 For example, nowadays I sometimes do things like write 819 00:41:15,120 --> 00:41:18,560 one prompt and then we'll have lunch and then come back and it 820 00:41:18,560 --> 00:41:19,920 made a major change in our system. 821 00:41:20,800 --> 00:41:24,520 And I would review it at a high level, ask for some small 822 00:41:24,520 --> 00:41:26,320 changes and then push it to production. 823 00:41:26,800 --> 00:41:29,080 But the work it did while I went while I went to lunch would 824 00:41:29,080 --> 00:41:32,400 basically be a whole week of, of both coding and meetings. 825 00:41:33,680 --> 00:41:36,120 And that was not possible to do for, you know, half a year ago. 826 00:41:36,120 --> 00:41:38,280 Then I would have to slice it and work in smaller steps. 827 00:41:38,800 --> 00:41:43,000 So I think things like today you can't go to a legacy banking 828 00:41:43,000 --> 00:41:48,240 system and say, can you rewrite this whole COBOL thing in to C# 829 00:41:50,120 --> 00:41:52,880 and and and, but maybe in the future, like you can do that 830 00:41:52,880 --> 00:41:54,840 with some parts of it, but you couldn't just do the whole 831 00:41:54,840 --> 00:41:56,520 thing. But in the future, my guess, 832 00:41:56,520 --> 00:41:59,080 maybe not within 12 months, but maybe, you know, maybe a bit 833 00:41:59,080 --> 00:42:00,960 longer. I think it would be perfectly 834 00:42:00,960 --> 00:42:04,920 feasible to basically write a prompt and saying rewrite this 835 00:42:04,920 --> 00:42:09,160 whole damn thing using C#. And as part of that, there'll be 836 00:42:09,160 --> 00:42:10,960 first discussion about how the work will happen. 837 00:42:11,520 --> 00:42:13,520 So it would it would maybe say things like, how about if I 838 00:42:13,520 --> 00:42:16,320 first write really good test coverage in the existing system 839 00:42:17,280 --> 00:42:20,040 and then I will let you know about any key trade-offs that 840 00:42:20,040 --> 00:42:22,160 need to be made. And we'll have points where we 841 00:42:22,160 --> 00:42:24,800 check in and we'll make an incremental plan for how to test 842 00:42:24,800 --> 00:42:26,400 this and roll it out. So we'd have a discussion 843 00:42:26,760 --> 00:42:30,400 exactly as if you hired A-Team or a company do this job right, 844 00:42:30,400 --> 00:42:32,320 you hire them. The mission is rewrite our whole 845 00:42:32,320 --> 00:42:36,080 system from cobalt to C#. Sure, that's the one prompt, but 846 00:42:36,080 --> 00:42:37,960 of course it's going to be touch points and check insurance. 847 00:42:38,280 --> 00:42:39,960 But that company is driving the work. 848 00:42:40,160 --> 00:42:41,920 And I think that's going to be, we're going to gradually 849 00:42:41,920 --> 00:42:43,320 shifting into more and more of that. 850 00:42:43,640 --> 00:42:45,880 And then again, it's as a as an engineer, it's up to you to 851 00:42:45,880 --> 00:42:47,840 decide you, you need to be like a good customer. 852 00:42:48,280 --> 00:42:50,520 You need to decide what, what job am I giving the AI? 853 00:42:50,520 --> 00:42:53,320 How am I going to follow up on it and which parts do I want to 854 00:42:53,320 --> 00:42:54,720 understand? Which parts am I willing to 855 00:42:54,720 --> 00:42:56,880 trust the model? That's going to be really 856 00:42:56,880 --> 00:42:58,600 important. But yeah, the the trend seems 857 00:42:58,600 --> 00:43:01,200 very clear that the the models are just improving really 858 00:43:01,200 --> 00:43:04,840 quickly. When people want to learn more 859 00:43:04,840 --> 00:43:07,720 about this and when people want to sort of maybe have questions 860 00:43:07,720 --> 00:43:10,720 or reach out, what would be the best way to to get in touch with 861 00:43:10,720 --> 00:43:12,880 you? So getting in touch with me is a 862 00:43:12,880 --> 00:43:17,880 little bit hopeless because I, I basically focus mostly on, you 863 00:43:17,880 --> 00:43:21,040 know, developing our company and our our platform and very rarely 864 00:43:21,040 --> 00:43:23,520 check e-mail, but sometimes on social media. 865 00:43:23,520 --> 00:43:29,000 So on Twitter and LinkedIn, I'm sometimes actor there it is. 866 00:43:29,200 --> 00:43:31,520 It is fine to drop an e-mail. My e-mail address is on our 867 00:43:31,520 --> 00:43:35,160 website that a bundle. I usually read emails, but I 868 00:43:35,160 --> 00:43:36,960 don't always respond. So it depends a little bit on, 869 00:43:37,000 --> 00:43:39,600 on on the context. But yeah, generally speaking, 870 00:43:39,600 --> 00:43:42,400 I'm pretty lousy at offering, you know, kind of free support 871 00:43:42,400 --> 00:43:47,440 to the world. I, I, I do post things sometimes 872 00:43:48,200 --> 00:43:51,160 and if people write comments or questions, I will sometimes 873 00:43:51,480 --> 00:43:54,320 respond in that context. I like to engage in that way. 874 00:43:54,320 --> 00:43:57,040 And what I like about that is if someone sends me an e-mail and I 875 00:43:57,040 --> 00:44:00,480 answer, it's only us 2. But if I post a a topic and 876 00:44:00,480 --> 00:44:02,640 someone writes a comment and I answer there, then it's kind of 877 00:44:02,640 --> 00:44:05,200 in public. And there's a lot more use out 878 00:44:05,200 --> 00:44:10,160 of that than than. Individual awesome and thank you 879 00:44:10,160 --> 00:44:12,520 so much for jumping onto this episode with me. 880 00:44:12,560 --> 00:44:15,760 It was really, really interesting and looking forward 881 00:44:15,760 --> 00:44:19,400 to your work at the bundle. Cool, thank you, this was super 882 00:44:19,400 --> 00:44:21,640 interesting. Likewise, bye. 883 00:44:23,480 --> 00:44:26,280 Thanks for tuning in and being part of today's deep dive with 884 00:44:26,280 --> 00:44:29,000 Hendrick Nieberg. It's always a privilege to share 885 00:44:29,000 --> 00:44:31,720 such inspiring stories of tech transformation and learning. 886 00:44:32,160 --> 00:44:34,840 If you enjoy this conversation, don't forget to follow and join 887 00:44:34,840 --> 00:44:38,800 Stellar Work on our social media channels, LinkedIn on Twitter, 888 00:44:38,880 --> 00:44:42,880 Instagram and Blue Sky. Gosh, join Blue Sky where 889 00:44:42,880 --> 00:44:45,600 discussion continues and you can connect with other industry 890 00:44:45,600 --> 00:44:47,920 leaders. Your feedback and stories shape 891 00:44:47,920 --> 00:44:50,800 the podcast, so make sure to reach out, leave a review and 892 00:44:50,800 --> 00:44:54,720 stay tuned for our next episode. And to them, keep experimenting. 893 00:44:54,720 --> 00:44:56,000 See you soon. Bye.