1 00:00:00,000 --> 00:00:02,740 If a company lays off developers, it means they don't 2 00:00:02,740 --> 00:00:05,934 have enough ideas to feed a development engine, that's now 3 00:00:05,934 --> 00:00:08,610 five times more productive. Product is a bottleneck. 4 00:00:08,640 --> 00:00:10,988 In a lot of companies, the product management needs to keep 5 00:00:10,988 --> 00:00:13,798 the developers developing. The features become shallow just 6 00:00:13,798 --> 00:00:16,193 to keep developers working. But now with AI, this is 7 00:00:16,193 --> 00:00:17,828 breaking down. Today's guest is Stephan 8 00:00:17,828 --> 00:00:21,332 Schmidt, a CTO coach and the author of The Amazing CTO's 9 00:00:21,332 --> 00:00:23,915 Missing Manual. With 45 years of software 10 00:00:23,915 --> 00:00:26,090 engineering experience, he specializes in transforming 11 00:00:26,090 --> 00:00:29,600 organizations into AI first teams and helping CTOs navigate 12 00:00:29,600 --> 00:00:32,941 the challenges of AI adoption. How do you think will be a good 13 00:00:32,941 --> 00:00:34,994 strategy for organizations to start rolling out AI 14 00:00:34,994 --> 00:00:37,355 effectively? Business is not focusing enough 15 00:00:37,355 --> 00:00:40,660 on AI usage for strategy, vision, features, and KPIs 16 00:00:40,660 --> 00:00:44,004 before it gets to a developer. You need to have a vision where 17 00:00:44,004 --> 00:00:46,464 you want to be and how AI helps you in that vision. 18 00:00:46,495 --> 00:00:49,292 Not just being reactive and say, okay, let's lay off 20% of 19 00:00:49,292 --> 00:00:50,394 people. That's not a strategy. 20 00:00:50,425 --> 00:00:52,689 Do you think a software development team needs to convey 21 00:00:52,689 --> 00:00:53,875 something different to the executives? 22 00:00:53,934 --> 00:00:56,833 Business expects everything to be twice as fast, or 10 times as 23 00:00:56,833 --> 00:00:58,934 fast, which is the kind of difficult to deliver. 24 00:00:58,995 --> 00:01:02,925 And then the CTO needs to somehow create a structure and 25 00:01:02,925 --> 00:01:05,199 also managing the expectations. If you're only thinking about 26 00:01:05,199 --> 00:01:06,555 generating code, you're doing it wrong. 27 00:01:06,765 --> 00:01:09,889 The struggle that the CTOs have for creating five time more 28 00:01:09,889 --> 00:01:11,955 output is the architecture is not in place. 29 00:01:12,015 --> 00:01:14,385 The processes and the control structures are not in place. 30 00:01:14,385 --> 00:01:18,512 AI is a great opportunity for CTOs to shine in the top 31 00:01:18,512 --> 00:01:20,584 management. You could do things earlier than 32 00:01:20,584 --> 00:01:22,460 other people. And that's a way to shine. 33 00:01:22,520 --> 00:01:25,596 Having five prototypes for great ideas per week that wasn't 34 00:01:25,596 --> 00:01:27,679 possible before. That's now possible with AI. 35 00:01:44,460 --> 00:01:46,906 Hey, quick pause. My goal with Tech Lead Journal 36 00:01:46,906 --> 00:01:49,490 is simple. Learn from the best in tech so 37 00:01:49,490 --> 00:01:52,860 we can all grow together. If this resonates with you, hit 38 00:01:52,860 --> 00:01:56,154 subscribe to follow the channel. It's the biggest way for you to 39 00:01:56,154 --> 00:01:58,725 support the show and help us keep bringing great guests and 40 00:01:58,725 --> 00:02:01,176 insights to you. Thanks for being here, and let's 41 00:02:01,176 --> 00:02:04,075 get back to it. Hello everyone. 42 00:02:04,075 --> 00:02:06,770 Welcome back to another new episode of the Tech Lead Journal 43 00:02:06,770 --> 00:02:09,079 podcast. Today I have with me a repeat 44 00:02:09,079 --> 00:02:12,133 guest, uh, Stephan Schmidt. Um, so if you still remember, 45 00:02:12,133 --> 00:02:15,565 it's about a year, maybe a few months ago, that we talk about 46 00:02:15,565 --> 00:02:18,756 The Amazing CTO's the Missing Manual, from Amazing CTO, right? 47 00:02:19,056 --> 00:02:22,807 So Stephan today is coming back to continue the theme of talking 48 00:02:22,807 --> 00:02:26,950 about how to become amazing CTO. But this time I'm sure all of us 49 00:02:26,950 --> 00:02:29,464 know that what is happening these days is about AI. 50 00:02:29,674 --> 00:02:33,364 So probably today's topic will be predominantly about AI based. 51 00:02:33,714 --> 00:02:35,184 Welcome back Stephan to the show. 52 00:02:35,184 --> 00:02:37,314 Looking forward for this exciting conversation. 53 00:02:37,750 --> 00:02:40,192 Thank you, Henry. Looking forward for this ex, uh, 54 00:02:40,192 --> 00:02:42,269 to this, conversation and thanks for having me again. 55 00:02:43,386 --> 00:02:45,246 Right. So Stephan, I think like one 56 00:02:45,246 --> 00:02:48,224 year ago, if you still remember, we talk about the missing manual 57 00:02:48,224 --> 00:02:51,289 for amazing CTO. So what made you wanna come back 58 00:02:51,289 --> 00:02:53,276 this time? Uh, is there something else 59 00:02:53,276 --> 00:02:58,172 still missing from your manual? I'm rewriting parts of it and 60 00:02:58,172 --> 00:03:02,697 especially I'm adding a chapter on AI because that's the 61 00:03:02,697 --> 00:03:06,826 elephant in the room, obviously. And that's from my CTO coaching 62 00:03:06,826 --> 00:03:11,353 also one of the biggest challenges currently, how to 63 00:03:11,353 --> 00:03:15,504 structurally introduce AI and transform software development 64 00:03:15,504 --> 00:03:17,854 organizations into AI-first organizations. 65 00:03:17,854 --> 00:03:19,314 That's a challenge for a lot of people. 66 00:03:20,240 --> 00:03:22,400 So maybe if you can share a little bit from your customers, 67 00:03:22,400 --> 00:03:26,200 what are some of the main theme that they are talking about, 68 00:03:26,200 --> 00:03:28,730 they're asking you, or they're confused about, I guess? 69 00:03:30,045 --> 00:03:33,169 Their situation is basically, they get a lot of pressure from 70 00:03:33,169 --> 00:03:37,302 the CEO, from business to introduce AI, uh, and to push 71 00:03:37,302 --> 00:03:40,889 really hard. Then they have developers who 72 00:03:40,889 --> 00:03:44,919 are experimenting with AI. Uh, they are using Cursor or 73 00:03:44,919 --> 00:03:46,649 Claude Code or Copilot, something. 74 00:03:47,327 --> 00:03:51,869 And then the CTO in the middle needs to somehow create a 75 00:03:51,869 --> 00:03:56,225 structure from the engineers that experiment and do stuff on 76 00:03:56,225 --> 00:03:58,579 different speeds. Some are very, very fast, some 77 00:03:58,579 --> 00:04:01,553 are very, very slow. And the business pressure to 78 00:04:01,553 --> 00:04:03,647 deliver and also managing the expectations. 79 00:04:03,647 --> 00:04:07,367 Like business expects everything to be twice as fast or 10 times 80 00:04:07,367 --> 00:04:10,557 as fast, uh, which is the kind of difficult difficult to 81 00:04:10,557 --> 00:04:12,377 deliver. So this is where they find 82 00:04:12,377 --> 00:04:13,581 themselves in. Yeah. 83 00:04:13,586 --> 00:04:15,855 So I guess if we still remember back then when we talk about 84 00:04:15,855 --> 00:04:19,060 the, one of the role for CTO is actually to become a bridge 85 00:04:19,060 --> 00:04:20,451 between business and tech, right? 86 00:04:20,870 --> 00:04:24,629 Um, so I guess this time, I would imagine the leaders, the 87 00:04:24,629 --> 00:04:27,566 executives seeing all the news, the crazy thing that, you know, 88 00:04:27,566 --> 00:04:30,226 a lot of AI vendors are kind of like promising. 89 00:04:30,720 --> 00:04:32,760 So I guess the pressure is kind of real, right? 90 00:04:32,760 --> 00:04:36,583 We, they expect the development to be, I dunno, twice, 10x 91 00:04:36,583 --> 00:04:39,155 faster. So first of all, do you think 92 00:04:39,155 --> 00:04:42,630 this is something that is, you know, true, that development can 93 00:04:42,630 --> 00:04:47,360 actually be that much faster? Um, yes and no. 94 00:04:48,410 --> 00:04:52,284 Uh, sorry for that answer. Like I'm currently, I'm also 95 00:04:52,284 --> 00:04:56,387 doing a lot of AI stuff on my own, in private, not being a 96 00:04:56,387 --> 00:04:58,397 coach. But I write myself, for example, 97 00:04:58,397 --> 00:05:01,102 a coaching operating system to make my coaching smoother and 98 00:05:01,102 --> 00:05:05,050 the operations of that smoother. And I don't look at the code for 99 00:05:05,050 --> 00:05:08,352 some time now and just let Claude Code write the code. 100 00:05:08,786 --> 00:05:12,287 And I feel like, yeah, I think it's five times faster or even 101 00:05:12,287 --> 00:05:14,698 more. I'm very, very, very productive 102 00:05:14,698 --> 00:05:17,961 on one hand. On the other hand, there are two 103 00:05:17,961 --> 00:05:20,836 things. First, um, it's very, very 104 00:05:20,836 --> 00:05:23,498 straining. I feel like as an engineer, 105 00:05:23,498 --> 00:05:27,118 doing like four features a day or something compared to one 106 00:05:27,118 --> 00:05:31,343 feature a day and that's, that has a high cognitive load and a 107 00:05:31,343 --> 00:05:34,761 lot of stress. And also managing AI is kind of 108 00:05:34,761 --> 00:05:37,200 stressful on one hand and the other. 109 00:05:37,410 --> 00:05:42,065 I think it works for me because I have the right guardrails in 110 00:05:42,065 --> 00:05:45,215 place in my new project. And the architecture of the 111 00:05:45,215 --> 00:05:50,330 project is also done in a way that works for AI because it was 112 00:05:50,330 --> 00:05:52,970 built, kind of built with AI in mind. 113 00:05:53,540 --> 00:05:58,431 The struggle that the CTOs have by for creating five times more, 114 00:05:58,909 --> 00:06:02,674 five times more output, is the architecture is not in place. 115 00:06:02,854 --> 00:06:05,604 And also the processes and the control structures are not in 116 00:06:05,604 --> 00:06:07,605 place. You know, the, so control is 117 00:06:07,605 --> 00:06:09,642 also very, very, challenging thing with AI. 118 00:06:10,269 --> 00:06:13,795 So business needs to also adapt. Yeah, you mentioned something 119 00:06:13,795 --> 00:06:16,796 very interesting, right? So when we have kind of like the 120 00:06:16,796 --> 00:06:19,111 process, the guardrails, maybe the rules, right, the 121 00:06:19,111 --> 00:06:21,737 architecture in place. Probably AI can be a good 122 00:06:21,737 --> 00:06:23,431 leverage, but I would imagine it's not. 123 00:06:23,791 --> 00:06:26,107 I mean, it's, it could be a luxury in so many software 124 00:06:26,107 --> 00:06:27,901 development team to have all these in place, right? 125 00:06:28,262 --> 00:06:30,953 People are still expecting the development team to actually 126 00:06:30,953 --> 00:06:33,112 still speed up no matter what the situation. 127 00:06:33,442 --> 00:06:36,460 So do you think, um, software development team needs to be 128 00:06:36,460 --> 00:06:38,842 able to convey something different to the executives? 129 00:06:38,872 --> 00:06:42,077 And if so, how should they actually communicate it back 130 00:06:42,077 --> 00:06:44,457 about, you know, the managing the expectations? 131 00:06:45,474 --> 00:06:49,616 So first, yes, I think they should communicate differently. 132 00:06:49,646 --> 00:06:53,995 And what they should do is... I'm a big fan of framing. 133 00:06:54,025 --> 00:06:56,005 You know, there is this book, Don't Think of an Elephant. 134 00:06:56,335 --> 00:06:59,280 I think that framing is a very, very important skill for 135 00:06:59,280 --> 00:07:02,272 managers to have. And the frame would be, for me, 136 00:07:02,272 --> 00:07:06,357 would be our architecture is not yet AI ready and we need to get 137 00:07:06,357 --> 00:07:09,888 it AI ready. And we are not a greenfield 138 00:07:09,888 --> 00:07:11,963 project. We are, we have a existing code 139 00:07:11,963 --> 00:07:13,997 base and we need to put stuff in place. 140 00:07:14,237 --> 00:07:18,869 So yes, we can benefit from AI and from generating code faster 141 00:07:18,869 --> 00:07:21,827 and more features, but we need to get into position. 142 00:07:22,325 --> 00:07:24,320 That's one thing. And the other thing there is 143 00:07:24,320 --> 00:07:27,971 stuff that can be done. One thing you can do is have 144 00:07:27,971 --> 00:07:30,368 much more prototypes. That's something like, people at 145 00:07:30,368 --> 00:07:32,342 Google called, I think called, prototype first. 146 00:07:32,402 --> 00:07:35,275 There's been discussion about this on LinkedIn, on other, in 147 00:07:35,275 --> 00:07:38,517 other places. Don't do requirements first or 148 00:07:38,517 --> 00:07:41,189 don't do ideation first. Go for prototype first. 149 00:07:41,459 --> 00:07:44,407 And I think this is something where CTOs can deliver already 150 00:07:44,407 --> 00:07:49,019 if you have a pipeline of prototypes and also MVPs. 151 00:07:49,319 --> 00:07:53,307 Um, you can show the benefit of like having, I dunno, having 152 00:07:53,307 --> 00:07:55,829 five prototypes for great ideas per week. 153 00:07:56,509 --> 00:07:58,349 That's something that wasn't possible before. 154 00:07:58,679 --> 00:08:02,979 That's now possible with AI. So you can deliver on some 155 00:08:02,979 --> 00:08:08,235 promises of AI to the business and keep them happy until your 156 00:08:08,235 --> 00:08:12,205 architecture is more in place to enable this boost in 157 00:08:12,205 --> 00:08:15,154 productivity. Yeah, so definitely I think 158 00:08:15,154 --> 00:08:18,860 everyone needs to be kind of like be trained, I guess like in 159 00:08:18,860 --> 00:08:22,062 terms of how to use AI because there are so many spectrum of 160 00:08:22,062 --> 00:08:25,014 experience in terms, for example, senior, juniors, right? 161 00:08:25,074 --> 00:08:27,128 They all could benefit differently, I would think, 162 00:08:27,128 --> 00:08:28,764 right? And especially if you mentioned 163 00:08:28,764 --> 00:08:31,374 there's no guardrail, then it becomes, uh, much more important 164 00:08:31,374 --> 00:08:34,216 for the seniors or maybe the executives, the CTO to actually 165 00:08:34,216 --> 00:08:37,193 put them in place. But talking aside from coding 166 00:08:37,193 --> 00:08:40,150 for a moment, right? So what do you think are some of 167 00:08:40,150 --> 00:08:42,633 the other use cases? I know that people are talking 168 00:08:42,633 --> 00:08:45,058 to ChatGPT or Gemini, you know, asking some questions. 169 00:08:45,058 --> 00:08:48,256 Are there some other use cases that you think in, you know, in 170 00:08:48,256 --> 00:08:50,878 your client situation that they could benefit from AI. 171 00:08:51,833 --> 00:08:54,310 Yes. I tell all my clients, uh, if 172 00:08:54,310 --> 00:08:57,240 you only think about generating code, you're doing it wrong. 173 00:08:57,577 --> 00:09:00,790 And I also do a lot of workshop, AI workshop, helping in the 174 00:09:00,790 --> 00:09:03,594 transition. Mostly helping with the 175 00:09:03,594 --> 00:09:07,224 motivation of developers and motivating developers to use AI 176 00:09:07,224 --> 00:09:10,237 because there are some who don't, a group of developers who 177 00:09:10,237 --> 00:09:13,006 don't want to use AI and a group who are very eager. 178 00:09:13,066 --> 00:09:17,525 And so most of my workshops are around motivating people. 179 00:09:17,525 --> 00:09:22,511 And I show a lot of examples. You can do a lot of security 180 00:09:22,511 --> 00:09:25,500 scanning, bug scanning. You can use prompting to get 181 00:09:25,500 --> 00:09:28,092 into a new code base to understand things. 182 00:09:28,512 --> 00:09:32,440 I think, um, Claude is very good at finding bugs with the right 183 00:09:32,440 --> 00:09:33,640 prompts and the right guardrails. 184 00:09:33,941 --> 00:09:37,181 So there's a lot of stuff that, and some stuff that was been, 185 00:09:37,241 --> 00:09:40,967 has been tedious before for developers, that's now much 186 00:09:40,967 --> 00:09:42,971 easier. Like sometimes fixing a bug or 187 00:09:42,971 --> 00:09:45,733 some, some other stuff. Creating some documentation, 188 00:09:45,733 --> 00:09:49,019 creating some tests. If you do it right, I think 189 00:09:49,019 --> 00:09:52,009 Claude Code is good at, even better at creating tests than 190 00:09:52,009 --> 00:09:54,259 developers are, if you're doing it the right way. 191 00:09:54,629 --> 00:09:57,170 So there's a lot of things that developers can do and 192 00:09:57,170 --> 00:09:59,560 engineering can do besides generating code. 193 00:10:00,051 --> 00:10:05,374 And on the other hand, I show a little, it's very naive perhaps 194 00:10:05,374 --> 00:10:10,495 example, but if I connect GitHub issues, Zendesk with MCP into 195 00:10:10,495 --> 00:10:14,920 Claude and the code and everything, and I ask Claude 196 00:10:14,920 --> 00:10:18,464 Code what features should I develop, you know? 197 00:10:18,464 --> 00:10:22,348 And then it looks at Zendesk and customer requests and bugs and 198 00:10:22,348 --> 00:10:25,788 issues and all of these things. And, perhaps a strategy document 199 00:10:25,788 --> 00:10:30,500 in, uh, in Notion, and then it comes up with a list of features 200 00:10:30,500 --> 00:10:34,717 that you should implement. Or you can say, well, this is my 201 00:10:34,717 --> 00:10:38,096 strategy, this is my roadmap. Does the strategy map to the 202 00:10:38,096 --> 00:10:41,119 roadmap or not? Or does the roadmap fulfill the 203 00:10:41,119 --> 00:10:43,394 strategy? So that, I think that's where 204 00:10:43,394 --> 00:10:46,430 business needs to adapt. They obviously adapt in 205 00:10:46,430 --> 00:10:49,800 marketing and sales, but that's where business needs to adapt in 206 00:10:49,800 --> 00:10:52,291 product development. Uh, where I know something 207 00:10:52,291 --> 00:10:54,228 about, I dunno, about sales or marketing. 208 00:10:54,568 --> 00:10:58,323 But, um, I think in product development, business is not 209 00:10:58,323 --> 00:11:05,088 focusing enough on AI using or AI usage for strategy, vision, 210 00:11:05,088 --> 00:11:08,907 features and KPIs. Finding KPIs that if we build 211 00:11:08,907 --> 00:11:11,755 this, how can we find out if it works? 212 00:11:11,905 --> 00:11:14,215 This kind of stuff that's pre-development. 213 00:11:14,605 --> 00:11:17,621 Um, I think there is a huge potential for business to use AI 214 00:11:17,621 --> 00:11:21,385 in product development before it gets to a developer. 215 00:11:21,415 --> 00:11:26,279 And just focusing on generating code and saying, okay, we want 216 00:11:26,279 --> 00:11:29,254 to use developers to generate more code with Claude. 217 00:11:29,924 --> 00:11:33,204 With Claude, I think that's too narrow focused. 218 00:11:34,403 --> 00:11:37,005 Yeah, so I think even like for me, when you mentioned about, 219 00:11:37,005 --> 00:11:38,806 you know, checking your strategy, vision. 220 00:11:38,806 --> 00:11:42,271 Even communicating that strategy and vision itself can benefit 221 00:11:42,271 --> 00:11:46,066 from using AI to maybe structure your sentence, make sure it 222 00:11:46,066 --> 00:11:48,766 aligns with your strategies, your KPIs, like you mentioned, 223 00:11:48,766 --> 00:11:50,854 right? I think definitely it's, um, up 224 00:11:50,854 --> 00:11:52,876 to your creativity on how you use AI. 225 00:11:52,906 --> 00:11:56,176 Uh, sometimes definitely it can give you some hallucination or 226 00:11:56,176 --> 00:12:00,457 something that is totally wrong. But it's really up to you at the 227 00:12:00,457 --> 00:12:03,292 end to actually criticize or maybe accept the suggestions 228 00:12:03,292 --> 00:12:06,799 done by AI. You mentioned about MCP and I 229 00:12:06,799 --> 00:12:10,319 think many people might benefit a lot from using MCP, right? 230 00:12:10,319 --> 00:12:13,267 So even I read, like I think Shopify mentioned about, you 231 00:12:13,267 --> 00:12:16,399 know, the usage of AI everywhere within the company and they 232 00:12:16,399 --> 00:12:17,849 connect everything through MCP internally. 233 00:12:18,119 --> 00:12:21,516 So tell us what kind of power that someone can leverage in an 234 00:12:21,516 --> 00:12:25,256 organization if let's say they have so many MCP available 235 00:12:25,256 --> 00:12:27,474 systems or sources that they can connect to. 236 00:12:28,294 --> 00:12:31,909 So first of all, I think integrating systems with MCP is 237 00:12:31,909 --> 00:12:36,146 a low hanging fruit. Introducing, uh, MCP servers or 238 00:12:36,146 --> 00:12:40,037 MCP bridges proxies, API proxies, is a low hanging fruit. 239 00:12:40,037 --> 00:12:43,306 It's a low effort, low effort, big gains. 240 00:12:44,146 --> 00:12:47,288 So that's the first thing. The second about MCP is there 241 00:12:47,288 --> 00:12:51,226 are a lot of use cases that also my clients use. 242 00:12:51,885 --> 00:12:55,612 The very first one is just connect your data warehouse with 243 00:12:55,612 --> 00:13:00,634 MCP to a chat bot or to Claude Code, and then show the CEO that 244 00:13:00,634 --> 00:13:04,575 he can ask questions about the data and about what's the best 245 00:13:04,575 --> 00:13:08,106 customer and what should we do, how can we upsell things. 246 00:13:08,106 --> 00:13:12,792 And so there's a huge benefit of connecting your data sources 247 00:13:12,792 --> 00:13:15,794 with MCP. But then there are also some 248 00:13:15,794 --> 00:13:18,954 other things like, simple, very simple example, but if you 249 00:13:18,954 --> 00:13:23,369 connect GitHub, perhaps not for MCP but you can also use it for 250 00:13:23,369 --> 00:13:27,246 MCP, GitHub to Claude Code. A build process fails, you can 251 00:13:27,246 --> 00:13:29,150 just say, okay, the build process fails. 252 00:13:29,920 --> 00:13:31,901 What would we need to do to fix it? 253 00:13:32,251 --> 00:13:36,907 And then if Claude Code comes up with the right plan, you say, 254 00:13:36,907 --> 00:13:41,117 okay then execute the plan. So you can also do a lot of 255 00:13:41,117 --> 00:13:44,307 operational things, um, with MCP and integrating into Claude. 256 00:13:44,577 --> 00:13:47,577 And something that some of my clients do is also do a lot of 257 00:13:47,577 --> 00:13:51,422 compliance work with MCP. Like you connect, uh, various 258 00:13:51,422 --> 00:13:54,992 sources, your processes, your implementations, code and all of 259 00:13:54,992 --> 00:13:57,778 these things. And then you can pre-screen for 260 00:13:57,778 --> 00:14:01,447 compliance before your audit and say, okay, am I compliant or am 261 00:14:01,447 --> 00:14:03,826 I not compliant? And then probably, the AI agent 262 00:14:03,826 --> 00:14:07,708 will come up with a list of stuff that you're not compliant. 263 00:14:07,708 --> 00:14:10,074 There is a ticket that is not approved, there is this and this 264 00:14:10,074 --> 00:14:13,062 and this. So, uh, it really helps you stay 265 00:14:13,062 --> 00:14:16,772 compliant if you do this iteratively and fast or 266 00:14:16,772 --> 00:14:18,824 continuously. Yeah. 267 00:14:19,154 --> 00:14:21,254 So I think some good tips there, right? 268 00:14:21,254 --> 00:14:23,777 So I like the one that you mentioned connecting your data 269 00:14:23,777 --> 00:14:26,294 warehouse or maybe even database, a simple database, 270 00:14:26,294 --> 00:14:28,058 right? That lets the non-technical 271 00:14:28,058 --> 00:14:30,949 users to query it using so-called natural language, 272 00:14:30,949 --> 00:14:32,739 right? And this is something that I 273 00:14:32,739 --> 00:14:34,712 think maybe some people realize it or not, right? 274 00:14:34,982 --> 00:14:38,852 So by leveraging MCP combined with the chat interface, right? 275 00:14:39,122 --> 00:14:42,321 Actually you can kind of like execute a lot of things through 276 00:14:42,321 --> 00:14:44,999 natural language and AI would somehow be able to translate 277 00:14:44,999 --> 00:14:48,593 that to kinda like the API calls, maybe summarizing and 278 00:14:48,593 --> 00:14:52,745 making inferences and all that. So definitely MCP is a good 279 00:14:52,745 --> 00:14:55,553 thing, especially for non-technical users, I think. 280 00:14:55,833 --> 00:14:59,123 Because previously maybe it's difficult to actually get the 281 00:14:59,123 --> 00:15:02,183 data source, analyze, use the tools and all that, but probably 282 00:15:02,183 --> 00:15:03,623 now it's getting much, much easier. 283 00:15:04,224 --> 00:15:07,976 So I think one mention, one mention you, uh, initially said 284 00:15:07,976 --> 00:15:12,430 is you have some jobs to motivate developers to using AI. 285 00:15:12,820 --> 00:15:16,334 So I'm a bit intrigued by this, because I would assume many 286 00:15:16,334 --> 00:15:19,559 developers would love to use AI to help them, you know, maybe 287 00:15:19,559 --> 00:15:21,380 writing some code and things like that. 288 00:15:21,874 --> 00:15:25,144 Tell us the case why some developers are not motivated? 289 00:15:25,992 --> 00:15:29,442 There are several reasons I, so I don't know. 290 00:15:29,922 --> 00:15:32,202 You know, I'm not in the head of the, of developers. 291 00:15:32,202 --> 00:15:35,442 I'm not... I can't read the mind of people, 292 00:15:35,442 --> 00:15:38,110 but from what I see and from what... 293 00:15:38,110 --> 00:15:40,329 I've been writing code for 40 years. 294 00:15:40,629 --> 00:15:44,718 So from what I see and what I see in people, I have some 295 00:15:44,718 --> 00:15:48,105 ideas. One idea is the company comes to 296 00:15:48,105 --> 00:15:53,464 the developer and says, use more AI so you can have a more 297 00:15:53,464 --> 00:15:58,447 stressful job, produce five times the features and kicker, 298 00:15:58,447 --> 00:16:02,644 same salary. So you know that, you know, 299 00:16:02,644 --> 00:16:04,149 that's where reticent makes sense. 300 00:16:04,149 --> 00:16:06,459 In which world does this make any sense? 301 00:16:06,489 --> 00:16:08,688 I don't know. I think it doesn't make any 302 00:16:08,688 --> 00:16:10,229 sense. So that's one thing. 303 00:16:10,913 --> 00:16:13,763 The other thing is look, because everyone benefits. 304 00:16:13,763 --> 00:16:16,436 The product manager benefits, the CTO benefits, the CEO 305 00:16:16,436 --> 00:16:19,997 benefit, everyone benefits basically from the simple 306 00:16:19,997 --> 00:16:22,916 generate code setup. The developer does not. 307 00:16:23,816 --> 00:16:27,393 And so that's one thing. The other thing is, the other 308 00:16:27,393 --> 00:16:31,023 thinking is if your value proposition is you're a senior 309 00:16:31,023 --> 00:16:35,097 developer with five years or 10 years of React - I hope React is 310 00:16:35,097 --> 00:16:39,980 that old but I guess - so if you have 10 years of React 311 00:16:39,980 --> 00:16:44,018 experience, and then you should use AI and Claude Code is quite 312 00:16:44,018 --> 00:16:47,962 good at, if you tell it to read this source and this 313 00:16:47,962 --> 00:16:50,242 documentation, this documentation, it's quite good 314 00:16:50,242 --> 00:16:54,446 at doing some hooks and components and all of that stuff 315 00:16:54,446 --> 00:16:56,747 that you thought, that distinguished you. 316 00:16:57,047 --> 00:17:00,782 The knowledge has distinguished you from other developers and 317 00:17:00,782 --> 00:17:04,522 that's taken away. So AI is taking a lot of things 318 00:17:04,522 --> 00:17:06,707 away that distinguishes you from other developers. 319 00:17:07,127 --> 00:17:11,599 So why should you like AI? It's leveling the game, the 320 00:17:11,599 --> 00:17:14,539 playing field. And if you're at the top, that 321 00:17:14,539 --> 00:17:17,342 does not make a lot of sense either, you know? 322 00:17:17,642 --> 00:17:21,403 And the third thought is, that's how I see it is. 323 00:17:21,883 --> 00:17:24,583 There are coders and there are creators. 324 00:17:25,204 --> 00:17:28,250 When I was a kid, probably, uh, I told you last time, but when I 325 00:17:28,250 --> 00:17:30,215 was a kid, I wanted to write video games. 326 00:17:30,215 --> 00:17:33,221 So I learned, I teach myself programming in a department 327 00:17:33,221 --> 00:17:36,671 store as a tool to write video games. 328 00:17:37,174 --> 00:17:40,702 I didn't want to be a coder. I want to create a, be a creator 329 00:17:40,702 --> 00:17:44,460 of video games. And over time I like coding, I 330 00:17:44,460 --> 00:17:49,150 liked the puzzles, I liked the intrinsic beauty of code. 331 00:17:49,210 --> 00:17:53,390 That's all the things I like. But nevertheless, I'm a creator. 332 00:17:53,390 --> 00:17:55,640 I want to create things with tools. 333 00:17:55,760 --> 00:17:58,917 And if you define yourself, I think if you define yourself as 334 00:17:58,917 --> 00:18:04,670 a creator who's using tools to create things, you flourish with 335 00:18:04,670 --> 00:18:08,519 AI. But if you define yourself as a 336 00:18:08,519 --> 00:18:12,061 coder who writes code and you're in the business because you 337 00:18:12,061 --> 00:18:15,251 like, love writing code, you don't care too much about what 338 00:18:15,251 --> 00:18:17,656 you create. You care a lot about the beauty 339 00:18:17,656 --> 00:18:20,277 of the code. And I like a little bit beauty 340 00:18:20,277 --> 00:18:22,557 of the code, but mostly I'm a creator. 341 00:18:22,957 --> 00:18:26,621 But if you're a coder, then I think that's a challenge for you 342 00:18:26,621 --> 00:18:29,122 to, uh, to accept AI as a new tool. 343 00:18:29,950 --> 00:18:34,547 So these are three reasons that I see that developers resist AI 344 00:18:34,547 --> 00:18:37,430 adoption. Yeah, very, very interesting, 345 00:18:37,430 --> 00:18:39,470 uh, reasons that you mentioned there. 346 00:18:39,680 --> 00:18:42,370 I think it could also be, like a threat, like when you mentioned, 347 00:18:42,370 --> 00:18:44,720 right? It could be seen as a threat to 348 00:18:44,720 --> 00:18:47,992 their kind of like skillset, existence, job existence, I 349 00:18:47,992 --> 00:18:50,356 mean. We know that a lot of people are 350 00:18:50,356 --> 00:18:52,521 saying, you know, we need lesser developers, right? 351 00:18:52,521 --> 00:18:54,921 Team size is gonna shrink. We see a lot of layoffs. 352 00:18:55,293 --> 00:18:58,623 Personally, what do you think? Would, um, developers, a lot of 353 00:18:58,623 --> 00:19:01,513 developers lose their jobs? Or something else could happen 354 00:19:01,513 --> 00:19:05,353 with the usage of AI? So a little bit, I'm a little 355 00:19:05,353 --> 00:19:07,969 bit gloom and doom. Uh, so I think developers will 356 00:19:07,969 --> 00:19:12,834 lose their job. The reason is not AI per se, but 357 00:19:12,834 --> 00:19:18,444 what I've been saying for last 20 years that product is kind of 358 00:19:18,444 --> 00:19:22,603 a bottleneck. In a lot of companies I've seen, 359 00:19:22,603 --> 00:19:27,760 product is a bottleneck and that manifests in a way that product 360 00:19:27,760 --> 00:19:30,788 creates very shallow features. They always need to... 361 00:19:30,788 --> 00:19:34,404 The most important thing in a lot of companies I see is that 362 00:19:34,404 --> 00:19:36,982 product management needs to keep the developers developing. 363 00:19:37,072 --> 00:19:40,751 That's a primary goal, you know. The product fails if developers 364 00:19:40,751 --> 00:19:44,354 have nothing to do when the Scrum, when the sprint starts, 365 00:19:44,354 --> 00:19:47,949 you know. So so there are a lot of work 366 00:19:47,949 --> 00:19:50,686 that product does to keep developers typing. 367 00:19:51,616 --> 00:19:56,259 And so they are already, I think seeing product management is 368 00:19:56,259 --> 00:19:59,702 already thinned. They don't have deep thoughts 369 00:19:59,702 --> 00:20:02,824 and great features. The features become shallow just 370 00:20:02,824 --> 00:20:04,994 to keep developers working. That's what I see. 371 00:20:05,579 --> 00:20:08,171 So in this way, product management has been a 372 00:20:08,171 --> 00:20:10,277 bottleneck. And I tell startups, you need 373 00:20:10,277 --> 00:20:12,089 more product managers, not more developers. 374 00:20:12,089 --> 00:20:15,087 You know, you need to have higher better ideas, not faster 375 00:20:15,087 --> 00:20:17,399 development. But it worked. 376 00:20:18,179 --> 00:20:22,306 But now with AI, this is breaking down, you know, so the 377 00:20:22,306 --> 00:20:23,999 product is becoming a bottleneck. 378 00:20:24,059 --> 00:20:27,737 The number of good ideas that they're coming up with ideas is 379 00:20:27,737 --> 00:20:31,949 the bottleneck. And I say if a company lays off 380 00:20:31,949 --> 00:20:34,559 developers, it means they don't have enough ideas. 381 00:20:35,259 --> 00:20:39,564 You know, it's not like we are going, we become more efficient. 382 00:20:39,984 --> 00:20:46,104 Yes, but bottom line is you don't have enough ideas to feed 383 00:20:46,104 --> 00:20:50,043 a development engine that's now five times, two times, five 384 00:20:50,043 --> 00:20:53,954 times more productive. Um, so the doom and gloom in me 385 00:20:53,954 --> 00:20:58,014 thinks, yeah, there will be layoffs, but the reason is not 386 00:20:58,014 --> 00:21:01,651 AI by definition, but because companies are limited on great 387 00:21:01,651 --> 00:21:04,120 ideas. Very interesting because I 388 00:21:04,120 --> 00:21:06,104 rarely hear this kind of perspective, right? 389 00:21:06,104 --> 00:21:08,354 So I think it kinda makes sense in a way, right? 390 00:21:08,384 --> 00:21:12,423 Because if you don't have too much so-called innovation, 391 00:21:12,423 --> 00:21:14,819 product development, like good features to build, right? 392 00:21:14,879 --> 00:21:18,824 Obviously, if we can automate lots, like big parts of our 393 00:21:18,824 --> 00:21:21,510 software development, you probably won't need a lot of 394 00:21:21,510 --> 00:21:24,483 developers, right? So I think, um, definitely it's 395 00:21:24,483 --> 00:21:28,619 very interesting perspective. And I find that a lot of 396 00:21:28,619 --> 00:21:32,931 developers actually think that AI is also taking a lot of their 397 00:21:32,931 --> 00:21:36,866 jobs in terms of, you know, like now, you know, I don't need so 398 00:21:36,866 --> 00:21:40,770 many juniors anymore. Or now I don't need to hire like 399 00:21:40,770 --> 00:21:42,106 different stack developers anymore. 400 00:21:42,446 --> 00:21:44,588 So what do you think? Is this something that is valid 401 00:21:44,588 --> 00:21:47,276 and what do you think should juniors do then if let's say 402 00:21:47,276 --> 00:21:50,245 that's the case? I think there is a huge 403 00:21:50,245 --> 00:21:52,046 opportunity for juniors. Why? 404 00:21:52,876 --> 00:21:55,626 Some of my clients and I think parts of the industry are moving 405 00:21:55,626 --> 00:21:59,385 to a product engineering role, which means taking on 406 00:21:59,385 --> 00:22:02,979 engineering, managing AI but also understanding product and 407 00:22:02,979 --> 00:22:06,679 coming up with features and all the minor decisions yourself 408 00:22:06,679 --> 00:22:10,100 without a product manager. Because I wrote a large article 409 00:22:10,100 --> 00:22:13,388 about Amdahl's Law and how it relates to product and AI, but 410 00:22:13,388 --> 00:22:16,738 that comes into play as a driver, uh, as one of the 411 00:22:16,738 --> 00:22:19,868 drivers that we might see the industry moving to a product 412 00:22:19,868 --> 00:22:22,891 engineering role. And I think like that's, I see 413 00:22:22,891 --> 00:22:25,159 it more like, you know, when I... 414 00:22:25,159 --> 00:22:27,757 I've been doing this for 40 years or 45 years. 415 00:22:27,999 --> 00:22:31,436 I've seen so many transformations in our industry. 416 00:22:31,436 --> 00:22:36,260 Not as big as AI, but like moving from machine language to 417 00:22:36,260 --> 00:22:40,532 C, moving from C to Java, moving perhaps from Java to Python, 418 00:22:40,532 --> 00:22:45,424 though for me it was more Perl, Python, Java, not the other way 419 00:22:45,424 --> 00:22:48,348 around. It's just like these ancient 420 00:22:48,348 --> 00:22:50,341 Fortran developers, they don't change. 421 00:22:50,341 --> 00:22:53,834 It's not Fortran developers who jump into Java. 422 00:22:54,301 --> 00:22:57,341 It's basically, if you look at it, it's junior Java developers. 423 00:22:57,631 --> 00:23:01,831 It's juniors coming from university who want to demand to 424 00:23:01,831 --> 00:23:05,075 jump into Java, who want to use Java jobs. 425 00:23:05,075 --> 00:23:11,432 Whereas senior C developers had a great resistance moving to 426 00:23:11,432 --> 00:23:15,037 Java. So there was always an 427 00:23:15,037 --> 00:23:20,707 opportunity for juniors to jump into new technologies to get a 428 00:23:20,707 --> 00:23:24,757 headstart in new technologies, because the incumbents, mostly a 429 00:23:24,757 --> 00:23:26,797 lot of incumbents resist the change. 430 00:23:26,827 --> 00:23:30,471 So I think there is a huge opportunity for juniors, to 431 00:23:30,471 --> 00:23:34,462 become proficient in AI, to know about AI, to become product 432 00:23:34,462 --> 00:23:36,062 engineers. And then they will become 433 00:23:36,062 --> 00:23:38,073 senior. They will not become senior Java 434 00:23:38,073 --> 00:23:41,189 developers or senior Python developers, but they will 435 00:23:41,189 --> 00:23:43,001 become, senior product engineers. 436 00:23:43,061 --> 00:23:46,398 So I think there is a huge opportunity for juniors. 437 00:23:47,417 --> 00:23:49,847 Yeah, I think that, that's, again, very another unique 438 00:23:49,847 --> 00:23:52,031 insight, right? So I think the key message for 439 00:23:52,031 --> 00:23:54,491 juniors is don't get discouraged, right, about this 440 00:23:54,491 --> 00:23:57,286 AI. They can actually become a much 441 00:23:57,286 --> 00:24:00,472 better AI native engineers. And in fact, in some random 442 00:24:00,472 --> 00:24:02,362 posts or articles that I've seen, right? 443 00:24:02,422 --> 00:24:05,799 And even personally myself, I've seen some of my juniors actually 444 00:24:05,799 --> 00:24:08,722 leverage AI differently than what I could think about. 445 00:24:08,932 --> 00:24:12,421 Simply because, you know, they are trained using AI in the 446 00:24:12,421 --> 00:24:15,044 first place, right? Like they are exposed to AI. 447 00:24:15,157 --> 00:24:17,647 I always think of it like the social media era, right? 448 00:24:17,887 --> 00:24:20,808 Sometimes, when you are not exposed from the very beginning 449 00:24:20,808 --> 00:24:23,400 about social media versus the youngsters who are very savvy 450 00:24:23,400 --> 00:24:26,531 with social media, the way that you can leverage on social media 451 00:24:26,531 --> 00:24:29,212 is so much different than, you know, the older people. 452 00:24:29,507 --> 00:24:32,304 So I think, don't get disheartened, uh, use AI to the 453 00:24:32,304 --> 00:24:34,102 best of what you can do, actually. 454 00:24:34,570 --> 00:24:36,093 I think again, this is a great insights. 455 00:24:36,543 --> 00:24:40,497 So speaking about leveraging AI a lot for developers, right? 456 00:24:40,647 --> 00:24:44,315 I personally, myself, feel that, after a few time, right? 457 00:24:44,345 --> 00:24:48,485 I get addicted myself and I feel every time I wanna start work, I 458 00:24:48,485 --> 00:24:52,808 would just leverage on AI first. My worry is that a lot of people 459 00:24:52,808 --> 00:24:56,126 fall into this same trap, and I think the portion of thinking 460 00:24:56,126 --> 00:24:59,502 becomes lesser and lesser, and we kind of like outsource a lot 461 00:24:59,502 --> 00:25:03,016 of thinking and decisions to AI. So do you think this is a valid 462 00:25:03,016 --> 00:25:05,415 concern? Uh, will there be any issue with 463 00:25:05,415 --> 00:25:06,640 our critical thinking going forward? 464 00:25:08,148 --> 00:25:12,103 I think on one hand, critical thinking is, um, is important. 465 00:25:12,103 --> 00:25:15,403 You should think about what you're doing and especially why 466 00:25:15,403 --> 00:25:18,265 you're doing things. I'm a huge proponent in asking 467 00:25:18,265 --> 00:25:23,932 why you do things. That said, I think, yes, you're 468 00:25:23,932 --> 00:25:26,733 losing. Like as I just mentioned, I made 469 00:25:26,733 --> 00:25:29,219 a transition. I started in a little bit in 470 00:25:29,219 --> 00:25:32,623 dabbling in BASIC, but then I did most of my programming was 471 00:25:32,623 --> 00:25:35,829 in machine code or in Assembler and partially in machine code. 472 00:25:36,537 --> 00:25:39,464 And then, on 8-bit and 16-bit CPUs. 473 00:25:40,054 --> 00:25:44,034 And when I moved to C, I forgot all of that. 474 00:25:44,662 --> 00:25:47,447 All of the really, really cool stuff that you could do with 475 00:25:47,447 --> 00:25:52,423 registers and address modes and all of these really cool skills 476 00:25:52,423 --> 00:25:55,273 I had. Optimization machine code and 477 00:25:55,273 --> 00:25:57,589 Assembly optimization skills in my head. 478 00:25:57,919 --> 00:26:02,869 I lost this, but I learned new things and C enabled me to do 479 00:26:02,869 --> 00:26:04,009 things I could not do in Assembler. 480 00:26:04,279 --> 00:26:07,741 Same thing with Java. And I think with AI, we see the 481 00:26:07,741 --> 00:26:11,668 same thing. My decisions move one, one step 482 00:26:11,668 --> 00:26:14,266 up. So I do not make the decision, 483 00:26:14,266 --> 00:26:18,130 should I create a caching layer here or should I do this here or 484 00:26:18,130 --> 00:26:20,968 there, you know? But I think it comes, like, I 485 00:26:20,968 --> 00:26:24,248 just move one level up. Uh, it's just one meta level. 486 00:26:24,248 --> 00:26:27,842 I'm meta, making meta decisions and then I'm making meta meta 487 00:26:27,842 --> 00:26:30,172 decisions. So I think I just move up the 488 00:26:30,172 --> 00:26:33,128 food chain. And I make different decisions 489 00:26:33,128 --> 00:26:35,433 and have different kind of thinking. 490 00:26:35,974 --> 00:26:38,593 And not just when I need to write C code, I have a different 491 00:26:38,593 --> 00:26:41,247 kind of thinking than compared to writing Assembly code. 492 00:26:41,307 --> 00:26:45,146 So that's how I see it. We just move up the food chain. 493 00:26:45,146 --> 00:26:48,056 It's not like some people say they become dumber. 494 00:26:48,056 --> 00:26:51,890 Yeah, I, yes. I mean, I can't remember, um, 495 00:26:51,890 --> 00:26:54,306 phone numbers. Does it make me dumber? 496 00:26:54,476 --> 00:26:57,562 I don't know, perhaps, but I can do so much more with my phone 497 00:26:57,562 --> 00:27:02,134 and do so much more thinking and I think it enabled me a lot so I 498 00:27:02,134 --> 00:27:05,127 can live with that. I don't remember any phone 499 00:27:05,127 --> 00:27:07,240 numbers except that of my parents. 500 00:27:07,360 --> 00:27:10,399 So that's, uh, yeah. Yeah. 501 00:27:10,771 --> 00:27:13,697 I also don't remember a lot of phone numbers, uh, a lot of 502 00:27:13,697 --> 00:27:16,006 birthdays, a lot of mathematics operations. 503 00:27:16,396 --> 00:27:19,428 So I think, uh, in one sense, yeah, you could say a little bit 504 00:27:19,428 --> 00:27:21,611 dumber, but at the other end, like what you mentioned, there 505 00:27:21,611 --> 00:27:23,976 are a lot of possibilities, how you can leverage the technology. 506 00:27:24,588 --> 00:27:27,378 I like that you mentioned that we are moving one level up, 507 00:27:27,378 --> 00:27:31,046 right moving to more the, like, higher level abstraction, right? 508 00:27:31,046 --> 00:27:34,209 So thinking about, you know, not just the code itself, the 509 00:27:34,209 --> 00:27:36,716 programming language, but thinking maybe about design, 510 00:27:36,716 --> 00:27:40,024 architecture and maybe features, outcomes, tests, and all that. 511 00:27:40,564 --> 00:27:43,554 But I think one of the counterargument is like some 512 00:27:43,554 --> 00:27:48,063 people into this vibe coding. So, you know, you just type your 513 00:27:48,063 --> 00:27:51,784 thing and, uh, AI create the code for you without even 514 00:27:51,784 --> 00:27:55,109 looking at the code itself. So what do you think about vibe 515 00:27:55,109 --> 00:27:56,532 coding? Have you tried vibe coding? 516 00:27:56,532 --> 00:27:58,972 Is it a feasible way of building software, do you think? 517 00:28:00,495 --> 00:28:04,049 I mean some of my insights, it might be shallow one or deeper 518 00:28:04,049 --> 00:28:06,245 one, but some of my insights on vibe coding. 519 00:28:06,270 --> 00:28:10,187 So I recently wrote an article line which says I no longer look 520 00:28:10,187 --> 00:28:14,088 at code, at source code. But there are a lot of things 521 00:28:14,088 --> 00:28:16,611 that enables me, I think enables me to do this because the 522 00:28:16,611 --> 00:28:19,846 projects are small. I'm doing on one hand, but I'm 523 00:28:19,846 --> 00:28:24,381 also like 45 years developers. So I know probably how to prompt 524 00:28:24,381 --> 00:28:28,629 things because I know how where Claude Code could derail or 525 00:28:28,629 --> 00:28:31,562 where there are challenges around caching, caching, cache 526 00:28:31,562 --> 00:28:33,862 consistency, where to store data. 527 00:28:34,162 --> 00:28:38,706 And so a lot of stuff that I know can go wrong, I put in my 528 00:28:38,706 --> 00:28:41,252 prompting. So I think I'm a good vibe coder 529 00:28:41,252 --> 00:28:45,044 on one hand, so it works. But I need to be a good coder to 530 00:28:45,044 --> 00:28:48,116 be a good vibe coder because your prompting is very 531 00:28:48,116 --> 00:28:50,216 different. And I'm happy with the outcome. 532 00:28:50,246 --> 00:28:54,139 It's like the application that I'm currently writing is 35,000 533 00:28:54,139 --> 00:28:57,434 lines of code. And I'm not looking at the code 534 00:28:57,434 --> 00:28:59,021 and I'm happy. It's, um, when, so. 535 00:28:59,947 --> 00:29:01,657 Works great. Lot of tests and stuff. 536 00:29:02,300 --> 00:29:04,758 So that's one thing. The second thing about vibe 537 00:29:04,758 --> 00:29:07,988 coding, I think if you start from scratch, there is a danger 538 00:29:07,988 --> 00:29:10,890 if you start from scratch. I think people do not understand 539 00:29:10,890 --> 00:29:16,396 what an AI is or what an LLM is. AI is this big thing, LLM and 540 00:29:16,396 --> 00:29:19,530 neural networks. LLM is a part of that one. 541 00:29:19,800 --> 00:29:23,985 And I think LLMs, they don't understand what LLM is that, 542 00:29:23,985 --> 00:29:27,693 it's just a non-deterministic probability machine. 543 00:29:28,173 --> 00:29:33,033 And a very, very important input to this is your existing code. 544 00:29:33,993 --> 00:29:37,566 And LLM is not a senior developer who looks at bad code 545 00:29:37,566 --> 00:29:42,168 and says, oh, that code is bad. You want that feature? 546 00:29:42,168 --> 00:29:45,493 I need to first, A, refactor this one, and B, build the 547 00:29:45,493 --> 00:29:50,184 feature in a great way. But people think like they, all 548 00:29:50,184 --> 00:29:54,762 the talking about how an AI is trained on GitHub code, it's bad 549 00:29:54,762 --> 00:29:57,986 because it's trained on bad code and all of these things. 550 00:29:58,416 --> 00:30:02,569 I think that's a mis, a gross misunderstanding on how LLMs 551 00:30:02,569 --> 00:30:05,150 work. I think it's much more useful to 552 00:30:05,150 --> 00:30:08,391 think of LLMs like they take your code as input, they take 553 00:30:08,391 --> 00:30:11,100 your prompt as input, and based on their training and 554 00:30:11,100 --> 00:30:14,027 probabilities, they create the most likely code that fulfills 555 00:30:14,027 --> 00:30:16,410 the prompt and your existing code base. 556 00:30:16,650 --> 00:30:20,880 So if your code base is bad, the added code would also be bad. 557 00:30:20,880 --> 00:30:24,570 It's not like, it's not, oh, I read some real great code on 558 00:30:24,570 --> 00:30:26,640 GitHub so I apply this here. No. 559 00:30:26,640 --> 00:30:30,540 If your code base is bad, it will create bad code. 560 00:30:30,900 --> 00:30:33,498 So I think it's very, very important to start from a great 561 00:30:33,498 --> 00:30:35,610 code base with a great template on one hand. 562 00:30:36,090 --> 00:30:39,132 And I think it's also important that at some point you need to 563 00:30:39,132 --> 00:30:41,903 refactor it. From getting it from here to 564 00:30:41,903 --> 00:30:45,088 here, and then you're happy. But if you stay here, it gets 565 00:30:45,088 --> 00:30:48,715 worse and worse and worse. And I think this is something, 566 00:30:48,715 --> 00:30:54,517 these two things are reasons why vibe coding goes bad, you know? 567 00:30:54,517 --> 00:30:58,082 So, um, if you are doing it right, I think it can be great 568 00:30:58,082 --> 00:31:01,700 for small right architecture that I mentioned before, uh, it 569 00:31:01,700 --> 00:31:03,944 can work. But it's easy to shoot yourself 570 00:31:03,944 --> 00:31:06,334 in the foot if you don't know what you're doing. 571 00:31:07,024 --> 00:31:10,289 It's really dangerous, I think. Yeah. 572 00:31:10,289 --> 00:31:13,825 So that's why I think we hear and see a lot of different 573 00:31:13,825 --> 00:31:15,957 outcomes when people vibe coding, right? 574 00:31:15,957 --> 00:31:18,387 So I think definitely the few key things that you mentioned. 575 00:31:18,537 --> 00:31:20,007 First, you need to be a good coder, right? 576 00:31:20,007 --> 00:31:24,074 You really need to know how a good design, good coding looks 577 00:31:24,074 --> 00:31:27,696 like because, otherwise when AI suggests you a lot of code 578 00:31:27,696 --> 00:31:30,958 suddenly in one go, right? Uh, you would be able to 579 00:31:30,958 --> 00:31:33,508 understand whether it's going in the right way or in the wrong 580 00:31:33,508 --> 00:31:35,464 direction, and you kind of like tweak along the way, right? 581 00:31:35,833 --> 00:31:39,106 And the danger here is like, let's say many, many people sell 582 00:31:39,106 --> 00:31:41,529 this promise that now non-technical people can 583 00:31:41,529 --> 00:31:44,182 actually write code simply by vibe coding. 584 00:31:44,332 --> 00:31:47,182 And a lot of such tools are built, you know, like Lovable, 585 00:31:47,182 --> 00:31:49,907 Bolt and things like that. Uh, but I think for prototype, 586 00:31:49,907 --> 00:31:52,120 simple things that you mentioned, uh, maybe it would 587 00:31:52,120 --> 00:31:54,707 work. But to make it something that is 588 00:31:54,707 --> 00:31:57,349 more robust, enterprise ready, secure, and all that, probably 589 00:31:57,349 --> 00:31:59,281 needs a lot more engineering fundamentals, right? 590 00:31:59,452 --> 00:32:02,056 Yeah. So definitely, I agree with, uh, 591 00:32:02,056 --> 00:32:05,190 your approach, right? So speaking about the approach 592 00:32:05,190 --> 00:32:08,938 here, I think you mentioned, that you have been around for 45 593 00:32:08,938 --> 00:32:11,443 years, right? So a lot of people don't have 594 00:32:11,443 --> 00:32:15,146 this luxury in their career. So what do you think will be a 595 00:32:15,146 --> 00:32:18,274 good advice, thinking about your head as a CTO coach, what would 596 00:32:18,274 --> 00:32:21,950 be a good advice for maybe, you know, five years, 10 years of 597 00:32:21,950 --> 00:32:24,761 experienced coders, right? How should they approach AI? 598 00:32:24,821 --> 00:32:28,857 What kind of things that should maybe learn more or upskill more 599 00:32:28,857 --> 00:32:31,149 in terms of leveraging AI the best way? 600 00:32:32,248 --> 00:32:35,420 I think it's important. I think the industry for various 601 00:32:35,420 --> 00:32:38,280 reasons will move towards to this product engineering role. 602 00:32:38,310 --> 00:32:41,556 I think that's, it just is a sweet spot for, it solves a lot 603 00:32:41,556 --> 00:32:45,386 of problems or several problems. Uh, so I would upskill myself 604 00:32:45,386 --> 00:32:48,426 more on product and product thinking and business thinking. 605 00:32:48,486 --> 00:32:52,711 Uh, then I would upskill on algorithms, you know, so that's 606 00:32:52,711 --> 00:32:56,619 what I would look into. And the second thing is you need 607 00:32:56,619 --> 00:33:00,030 to work with AI and see its deficits and see its benefits 608 00:33:00,030 --> 00:33:02,879 and learn about it. If you want to be a great 609 00:33:02,879 --> 00:33:04,535 prompter, you need to do a lot of prompting. 610 00:33:04,921 --> 00:33:07,381 If you want to be a great coder, you need to write a lot of code. 611 00:33:07,651 --> 00:33:10,460 So that's how I see it and what people should do. 612 00:33:10,957 --> 00:33:14,764 Just experiment and see what's happening and what's not 613 00:33:14,764 --> 00:33:18,020 happening. It's kind of a transitional 614 00:33:18,020 --> 00:33:22,415 problem because I really believe that today, as I said, I, for 615 00:33:22,415 --> 00:33:26,176 vibe coding, you need to have a good developer background to 616 00:33:26,176 --> 00:33:29,842 create a positive outcome. I'm not so sure about what 617 00:33:29,842 --> 00:33:33,371 happens in five years. I'm giving talks to companies, 618 00:33:33,371 --> 00:33:37,347 universities, which is one of them is called Beyond Software 619 00:33:37,347 --> 00:33:40,845 or AI is Not Software. That's one of the talks I'm 620 00:33:40,845 --> 00:33:44,932 giving. And some months ago, one example 621 00:33:44,932 --> 00:33:49,630 is playing tic-tac-toe. Because I believe that in the 622 00:33:49,630 --> 00:33:53,420 future, code generation is a transitional technology. 623 00:33:53,780 --> 00:33:56,705 You know, there will be no code, there will be no software, no 624 00:33:56,705 --> 00:33:59,350 code. There will just be AI in some 625 00:33:59,350 --> 00:34:02,420 years, I dunno when, but in some years, you know. 626 00:34:02,420 --> 00:34:06,860 And some months ago, I wanted to play tic-tac-toe, uh, with 627 00:34:06,860 --> 00:34:11,579 ChatGPT as an example to, there is no tic-tac-toe game in 628 00:34:11,579 --> 00:34:14,208 ChatGPT. What is ChatGPT able to do? 629 00:34:14,358 --> 00:34:17,109 And I played tic-tac-toe and it didn't work. 630 00:34:17,379 --> 00:34:20,777 You know, it made mistakes and there were very, very, a lot of 631 00:34:20,777 --> 00:34:23,438 handholding. Because I have a conference talk 632 00:34:23,438 --> 00:34:26,427 coming up this week. I tried the same tic-tac-toe 633 00:34:26,427 --> 00:34:29,023 example with Claude and there was no problem. 634 00:34:29,632 --> 00:34:33,273 So I was just saying, let's play a game of tic-tac-toe and Claude 635 00:34:33,273 --> 00:34:34,803 played a game of tic-tac-toe with me. 636 00:34:35,193 --> 00:34:38,944 Um, made no mistakes. And so we had, I've seen some 637 00:34:38,944 --> 00:34:42,934 progress on that. Very, very low difficulty level 638 00:34:42,934 --> 00:34:45,054 obviously. Tic-tac-toe is a very simple 639 00:34:45,054 --> 00:34:48,103 thing but I've seen some progress on can't play to I 640 00:34:48,103 --> 00:34:50,744 don't need to write a tic-tac-toe game, I can just go 641 00:34:50,744 --> 00:34:53,714 to the AI and say, let's play tic-tac-toe. 642 00:34:54,344 --> 00:34:57,921 And because I do a lot of these experiments, I did a minesweeper 643 00:34:57,921 --> 00:35:02,211 experiment some time ago where, um, I vibe coded from one 644 00:35:02,211 --> 00:35:05,442 prompt, prompt shotting Minesweeper, worked, and now I 645 00:35:05,442 --> 00:35:09,480 thought, okay, let's play Minesweeper with Claude Code or 646 00:35:09,480 --> 00:35:11,630 ChatGPT. That did not work yet. 647 00:35:11,660 --> 00:35:14,210 So as of yesterday, that's not there. 648 00:35:14,321 --> 00:35:19,837 But I believe we are moving to a no-source-code set up in the 649 00:35:19,837 --> 00:35:23,328 future. So for a transition, very long 650 00:35:23,328 --> 00:35:26,350 talk, sorry. But for a transition, I think 651 00:35:26,350 --> 00:35:29,159 moving, taking on more product understanding is something I 652 00:35:29,159 --> 00:35:32,149 would go. In the long term, I think source 653 00:35:32,149 --> 00:35:33,817 code is not that important anymore. 654 00:35:34,672 --> 00:35:38,512 So don't worry as a junior. It's just, it's a transition. 655 00:35:39,357 --> 00:35:41,607 Yeah. So definitely there could be a 656 00:35:41,607 --> 00:35:44,673 possibility where we can kind of like, uh, write software by 657 00:35:44,673 --> 00:35:48,094 leveraging on a lot of natural language, just like vibe coding 658 00:35:48,094 --> 00:35:51,619 in this case, right? So maybe in the future, the AI 659 00:35:51,619 --> 00:35:54,481 could be much smarter, right? In terms of writing better 660 00:35:54,481 --> 00:35:56,794 design as to what we want it to be. 661 00:35:57,218 --> 00:35:59,168 So definitely looking forward for that future. 662 00:35:59,635 --> 00:36:02,472 And I think a lot of organizations these days 663 00:36:02,472 --> 00:36:04,302 definitely want to roll out AI, right? 664 00:36:04,302 --> 00:36:07,421 So maybe as part of your experience as well, how do you 665 00:36:07,421 --> 00:36:10,683 think would be a good strategy for organizations to start 666 00:36:10,683 --> 00:36:14,418 rolling out AI or making sure that they get the most effective 667 00:36:14,418 --> 00:36:18,588 benefits from using AI? First, I think you mentioned it. 668 00:36:18,785 --> 00:36:21,320 First, I think you need strategy, you know. 669 00:36:21,480 --> 00:36:25,260 And the strategy and... Do we want to lay off 20% of 670 00:36:25,260 --> 00:36:29,265 people uh, so we make more profit and we use AI for this? 671 00:36:29,325 --> 00:36:31,725 That's not a strategy. That's not an AI strategy. 672 00:36:32,085 --> 00:36:34,661 I think you need to have a vision where you want to be and 673 00:36:34,661 --> 00:36:39,080 how AI helps you in that vision or how AI is part of that 674 00:36:39,080 --> 00:36:41,799 vision. And then from that drive a 675 00:36:41,799 --> 00:36:44,573 strategy and then say, okay, we need to do this and this and 676 00:36:44,573 --> 00:36:45,749 this, and this needs to be in place. 677 00:36:46,409 --> 00:36:50,052 Like my example for strategy is, which shows I have no clue about 678 00:36:50,052 --> 00:36:52,532 Mount Everest, but my example is about Mount Everest. 679 00:36:53,063 --> 00:36:57,776 For a strategy, you need to have two things: things you want to 680 00:36:57,776 --> 00:37:00,750 achieve and things that you have as capabilities. 681 00:37:00,750 --> 00:37:03,644 For example, if you want to climb Mount Everest, there is 682 00:37:03,644 --> 00:37:07,283 Base Camp One and Base Camp Two and the North Col, Ridge, and 683 00:37:07,283 --> 00:37:10,170 then The Summit. That's the things you need to 684 00:37:10,170 --> 00:37:12,074 achieve. And then there are things you 685 00:37:12,074 --> 00:37:15,053 need to have like an ice pick and a tent and an oxygen mask 686 00:37:15,053 --> 00:37:17,870 and check it and stuff that you need. 687 00:37:18,490 --> 00:37:21,968 And that's what for me is a strategy and that maps to AI. 688 00:37:21,998 --> 00:37:24,910 What are the things that I need to have in place and what are 689 00:37:24,910 --> 00:37:26,498 the things that I need to achieve? 690 00:37:26,933 --> 00:37:30,763 And that would be part, for example, you need to 691 00:37:30,763 --> 00:37:34,169 re-architecture probably your code base, you need to change 692 00:37:34,169 --> 00:37:37,544 your processes, you need to add the deep prototyping. 693 00:37:37,904 --> 00:37:41,054 So you need to have a lot of changes in place and say, I need 694 00:37:41,054 --> 00:37:42,554 to do this, this, this, this, this, this. 695 00:37:42,644 --> 00:37:44,354 And that's part of my strategy, you know. 696 00:37:44,991 --> 00:37:47,934 Whatever you want to climb, if it's Mount Everest or do 697 00:37:47,934 --> 00:37:49,671 something else, cross the world or you know. 698 00:37:50,621 --> 00:37:53,605 Whatever you want to do with AI, but you need to have a strategy 699 00:37:53,605 --> 00:37:57,031 that shows that you can leverage AI to get there. 700 00:37:57,111 --> 00:37:58,531 That's what I think is important. 701 00:37:58,811 --> 00:38:02,569 Not just being reactive and say, okay, let's lay off 20% of 702 00:38:02,569 --> 00:38:06,026 people so we increase profits and that's not a strategy. 703 00:38:07,141 --> 00:38:09,061 Right. I still remember back then when 704 00:38:09,061 --> 00:38:12,341 we discussed part of the role of CTO is to come up with a 705 00:38:12,341 --> 00:38:14,421 strategy, right? And you mentioned that a good 706 00:38:14,421 --> 00:38:17,515 strategy enables people to make better and easier decision, 707 00:38:17,515 --> 00:38:19,699 right? Um, so I think the same thing 708 00:38:19,699 --> 00:38:21,991 here, right? If you can lay out a good AI 709 00:38:21,991 --> 00:38:24,224 strategy, definitely you can help people make good decision 710 00:38:24,224 --> 00:38:26,716 on how to leverage AI or benefit with AI. 711 00:38:27,392 --> 00:38:29,642 Maybe if you can give some examples, I don't know whether 712 00:38:29,642 --> 00:38:30,963 you have something in mind, right? 713 00:38:30,963 --> 00:38:34,315 What are the typical good AI strategy that people can adopt, 714 00:38:34,315 --> 00:38:36,598 you know, especially for software development team? 715 00:38:37,607 --> 00:38:38,847 A lot of the stuff that I mentioned. 716 00:38:38,965 --> 00:38:41,864 I think you need to have uh Move to prototype first. 717 00:38:42,194 --> 00:38:46,738 Um, have a product funnel of prototypes, MVPs, PMF, and use 718 00:38:46,738 --> 00:38:50,292 AI to drive all of that to a certain point before you take 719 00:38:50,292 --> 00:38:53,449 over the source code. I think that's very important 720 00:38:53,449 --> 00:38:58,079 part of an AI strategy today. Second, for the whole company, 721 00:38:58,079 --> 00:39:00,544 that's not CTO, but mostly CIO stuff. 722 00:39:00,544 --> 00:39:04,008 But you also should have a strategy on how you want to 723 00:39:04,008 --> 00:39:06,434 automate, what you want to automate, who owns this? 724 00:39:07,114 --> 00:39:10,462 It's also part of the strategy, who owns AI is like everyone 725 00:39:10,462 --> 00:39:14,315 owning their own AI? Is there an AI officer who makes 726 00:39:14,315 --> 00:39:17,679 sure efforts are coordinated and compliant and secure? 727 00:39:18,062 --> 00:39:21,719 So that's part of the strategy. That's a decision to be made, I 728 00:39:21,719 --> 00:39:23,823 think. I think, re-architecture, I 729 00:39:23,823 --> 00:39:26,732 mentioned several times. That's a part to to of your 730 00:39:26,732 --> 00:39:29,196 strategy. Also part of the strategy, how 731 00:39:29,196 --> 00:39:31,824 do you get developers, all developers to adopt this? 732 00:39:32,355 --> 00:39:36,471 Two of my clients just made, uh, the decision to re... to have a 733 00:39:36,471 --> 00:39:38,625 new position, which is product engineer. 734 00:39:39,148 --> 00:39:41,637 And if you want to become product engineer, you need to do 735 00:39:41,637 --> 00:39:45,612 A, B, C, which is part of that is creating prototypes, doing 736 00:39:45,612 --> 00:39:47,952 AI. So that creates a natural push 737 00:39:47,952 --> 00:39:51,754 for everyone to adopt AI and to think about how to use it so 738 00:39:51,754 --> 00:39:54,360 they become product engineers. It's just not only a relabeling, 739 00:39:54,360 --> 00:39:57,596 which is something I would've done, I would relabel the titles 740 00:39:57,596 --> 00:39:59,731 which would not be the best thing. 741 00:39:59,731 --> 00:40:02,441 I think this is something to earn is greater. 742 00:40:02,502 --> 00:40:06,448 But the strategy needs to be explained how you get these 743 00:40:06,448 --> 00:40:11,028 people on AI. But also like how do you do AI 744 00:40:11,028 --> 00:40:15,064 training and security that ties into do I have an AI officer or 745 00:40:15,064 --> 00:40:17,210 not? But nevertheless, what's the 746 00:40:17,210 --> 00:40:20,472 strategy about data? I think part of the strategy 747 00:40:20,472 --> 00:40:23,167 could also be do you want to build your own models? 748 00:40:23,527 --> 00:40:26,803 Is there a benefit in owning your own models, training them 749 00:40:26,803 --> 00:40:29,159 or not? Are you a prompting company? 750 00:40:29,709 --> 00:40:34,009 Are you using prompts for 25, 26, and then move to your own 751 00:40:34,009 --> 00:40:35,471 model? So what are your... 752 00:40:35,471 --> 00:40:39,456 I think a lot of these things need to be decided and put in 753 00:40:39,456 --> 00:40:43,291 place so you, so that you can believe you make it to your 754 00:40:43,291 --> 00:40:44,947 vision. Yeah. 755 00:40:45,217 --> 00:40:47,390 So definitely a few things are very interesting. 756 00:40:47,570 --> 00:40:48,980 I wanna pick up first thing first, right? 757 00:40:48,980 --> 00:40:52,100 Because these days there are so many AI tools available, and 758 00:40:52,100 --> 00:40:55,190 even, for example, model, right? We know there's a Claude Code, 759 00:40:55,190 --> 00:40:59,378 there's a Codex, OpenAI, right? There's a Gemini, open model as 760 00:40:59,378 --> 00:41:02,670 well, Chinese model. And I remember back then, we 761 00:41:02,670 --> 00:41:06,186 talked about tech zoo, you know, having a lot of tech, in your, 762 00:41:06,186 --> 00:41:08,642 uh, stack, right? I think is it the case? 763 00:41:08,711 --> 00:41:11,491 Same thing that will happen, AI tech zoo? 764 00:41:12,121 --> 00:41:15,171 And what will be your strategy? Because these tools are just 765 00:41:15,171 --> 00:41:16,871 gonna be, you know, exploding, right? 766 00:41:17,636 --> 00:41:19,381 Yeah, yeah, yeah, yeah, yeah. Totally, totally with you. 767 00:41:19,411 --> 00:41:20,851 Didn't, I didn't make that connection. 768 00:41:20,851 --> 00:41:22,591 It's, I will use that in the future. 769 00:41:23,055 --> 00:41:24,256 Thank you for coming up with that. 770 00:41:24,754 --> 00:41:26,114 Yeah, I think we have a model zoo. 771 00:41:26,120 --> 00:41:30,793 There is a slew of models and tools, uh, like if I talked to 772 00:41:30,793 --> 00:41:34,858 my clients, they have a license for Copilot because that comes 773 00:41:34,858 --> 00:41:38,741 with their Microsoft stuff. Some developer prefer IntelliJ 774 00:41:38,741 --> 00:41:43,648 so they have June, uh, Junie, then some have Claude Code. 775 00:41:43,668 --> 00:41:47,199 So, yeah, that's a challenge. And also model drift is a 776 00:41:47,199 --> 00:41:48,660 challenge. Like models change. 777 00:41:48,660 --> 00:41:51,900 So it's something that works now, does not work with the next 778 00:41:51,900 --> 00:41:55,323 model. Like stuff that works in a 779 00:41:55,323 --> 00:41:59,607 certain way with Opus 3.x then suddenly stops working with 780 00:41:59,607 --> 00:42:02,098 four. So there is a new dimension on, 781 00:42:02,098 --> 00:42:04,756 um, on this. So managing models I think is 782 00:42:04,756 --> 00:42:07,256 important and everything that part of your strategy, yeah. 783 00:42:07,706 --> 00:42:12,458 What I would do is because I personally, I'm using ChatGPT, 784 00:42:12,458 --> 00:42:16,469 Claude, and Gemini. And I feel they are different. 785 00:42:17,076 --> 00:42:20,469 So I think, still think Claude personally, personal opinion, 786 00:42:20,469 --> 00:42:24,213 uh, not a scientist, Claude is the best coding agent, I feel. 787 00:42:24,303 --> 00:42:26,523 Uh, Sonnet 4.5 is really great, I think. 788 00:42:27,193 --> 00:42:31,089 But I use, also use Gemini because it's part of my Pixel 789 00:42:31,089 --> 00:42:32,998 phone, integrates great with Pixel. 790 00:42:33,486 --> 00:42:38,163 But I think they are different. And what I would do is really 791 00:42:38,163 --> 00:42:42,102 pay someone in my company to experiment with different models 792 00:42:42,102 --> 00:42:46,038 on coding tasks and on other tasks, and to find out what's 793 00:42:46,038 --> 00:42:48,778 the best model for us. Because I feel like a little 794 00:42:48,778 --> 00:42:52,438 bit, when I talk to my clients, they think there is no 795 00:42:52,438 --> 00:42:53,918 difference. You know, if I talk to my 796 00:42:53,918 --> 00:42:54,982 clients, they think there is no difference. 797 00:42:54,982 --> 00:42:59,317 It's no difference if we use Copilot or Claude or Climax or 798 00:42:59,317 --> 00:43:02,668 something, you know. But I feel there is a huge 799 00:43:02,668 --> 00:43:05,002 difference between models. And, um, a very minor thing 800 00:43:05,002 --> 00:43:08,136 which has nothing to do with software development, but if I 801 00:43:08,136 --> 00:43:12,924 argue with a model, I fair found ChatGPT is I found, yeah, 802 00:43:12,924 --> 00:43:17,563 ChatGPT and Claude are very good at arguing and taking my 803 00:43:17,563 --> 00:43:21,483 arguments and going back to sources and arguing with me. 804 00:43:22,093 --> 00:43:24,683 Whereas I feel Gemini is very, very bad at arguing. 805 00:43:24,683 --> 00:43:26,933 It's not arguing, it just says, no, no, you're wrong. 806 00:43:26,933 --> 00:43:28,814 I'm right. Because this is this. 807 00:43:28,814 --> 00:43:31,184 And I say, I bring another argument and says, no, no. 808 00:43:31,184 --> 00:43:34,319 Yeah, I know why you think this, but you're wrong because, so 809 00:43:34,319 --> 00:43:38,436 it's not taking any input. Gemini, I feel that's how I use 810 00:43:38,436 --> 00:43:40,433 it. Non-native speaker, doing a lot 811 00:43:40,433 --> 00:43:44,019 of English in Gemini. I feel Gemini is bad at arguing 812 00:43:44,019 --> 00:43:46,484 and thinking, whereas Claude and ChatGPT are much better. 813 00:43:46,484 --> 00:43:50,494 So I would pay someone to have a great understanding about tools, 814 00:43:50,554 --> 00:43:55,198 what's there, how they differ. If I think this is a competitive 815 00:43:55,198 --> 00:43:58,564 advantage, I would want to know what to use. 816 00:43:58,564 --> 00:44:02,635 And quite frankly, the belief of my clients that all the tools 817 00:44:02,635 --> 00:44:06,408 are the same, all the models are the same, I think that's not 818 00:44:06,408 --> 00:44:09,051 true. Yeah, so I personally think it's 819 00:44:09,051 --> 00:44:11,880 not true as well. Uh, I think we can tell the 820 00:44:11,880 --> 00:44:13,743 difference. And especially if you layer it 821 00:44:13,743 --> 00:44:16,846 on top with agentic AI, right? That becomes much more 822 00:44:16,846 --> 00:44:18,946 different, right? So maybe if you use Cursor 823 00:44:18,946 --> 00:44:20,986 versus either Windsurf or Junie and all that. 824 00:44:21,461 --> 00:44:24,031 I think the difference could be really, really huge because some 825 00:44:24,031 --> 00:44:27,402 could do more plan-based thing. They can reverify the task 826 00:44:27,402 --> 00:44:30,710 before they hand it over to you. I think, yeah, definitely there 827 00:44:30,710 --> 00:44:33,737 are a lot of things that could change and even the upgrade of 828 00:44:33,737 --> 00:44:36,112 the model itself could introduce a new behavior. 829 00:44:36,142 --> 00:44:39,325 Maybe some are more credit, like more credit consumption. 830 00:44:39,325 --> 00:44:42,468 Uh, you know, they will do and some will be lesser. 831 00:44:42,773 --> 00:44:45,243 So I think these are, I think, forever changing. 832 00:44:45,273 --> 00:44:47,733 Uh, and don't forget to always keep researching on that. 833 00:44:48,333 --> 00:44:52,623 So speaking about data privacy, security, I think this is one of 834 00:44:52,623 --> 00:44:56,538 the major concerns, especially for, you know, CISO, the InfoSec 835 00:44:56,538 --> 00:44:59,373 people. As the CTO coach, uh, what would 836 00:44:59,373 --> 00:45:02,898 you advise people in order not to neglect this part of data 837 00:45:02,898 --> 00:45:07,554 privacy and security? I have a mental, mental model, 838 00:45:07,554 --> 00:45:11,184 which might be right, might be correct, might be incorrect, but 839 00:45:11,184 --> 00:45:12,906 at least, I think it makes sense. 840 00:45:12,906 --> 00:45:17,020 It makes sense to me. AIs have a very, are very, very 841 00:45:17,020 --> 00:45:22,338 good at broad knowledge. So they know about zip codes in 842 00:45:22,338 --> 00:45:23,714 China. I don't. 843 00:45:24,372 --> 00:45:26,826 Probably Chinese developers don't know about zip codes in 844 00:45:26,826 --> 00:45:29,213 Germany. So that's something, this kind 845 00:45:29,213 --> 00:45:32,648 of broad knowledge is something that AIs are very, very good at. 846 00:45:33,108 --> 00:45:36,490 It's like these on Hacker News. I'm a vivid Hacker News reader. 847 00:45:36,851 --> 00:45:40,398 There's always these threads about, uh, I think 848 00:45:40,398 --> 00:45:43,494 misconceptions developers have about money, about addresses, 849 00:45:43,494 --> 00:45:47,908 about, like this is where the people are very bad, broad 850 00:45:47,908 --> 00:45:50,368 knowledge. This is where AIs are very good. 851 00:45:50,984 --> 00:45:56,144 Where you as a person are very, very good in a company is a very 852 00:45:56,144 --> 00:46:00,222 deep knowledge about your problem domain of your company. 853 00:46:01,372 --> 00:46:06,778 And AI companies, the most precious thing you have is this 854 00:46:06,778 --> 00:46:09,770 deep knowledge that they don't have, you know. 855 00:46:09,800 --> 00:46:15,005 About perhaps, TSMC might have very special knowledge about how 856 00:46:15,005 --> 00:46:18,671 to produce chips with a, with the newest node. 857 00:46:19,183 --> 00:46:21,747 Other companies are struggling. Something is behind that 858 00:46:21,747 --> 00:46:24,163 interest, behind that. So other fabs are behind that. 859 00:46:24,163 --> 00:46:26,833 That's very, very deep knowledge only they have. 860 00:46:27,417 --> 00:46:30,337 But also you as a startup might have, because you're doing a lot 861 00:46:30,337 --> 00:46:33,737 of research, you have some very, very deep knowledge on 862 00:46:33,737 --> 00:46:37,221 something. And I would be concerned that 863 00:46:37,221 --> 00:46:41,548 beside regulations, compliance, GDPR, and all of these things, I 864 00:46:41,548 --> 00:46:45,578 would be concerned that this deep knowledge that only I as a 865 00:46:45,578 --> 00:46:49,241 company have, is getting out. Because that's what the AI 866 00:46:49,241 --> 00:46:52,926 companies want their very broad knowledge, but they want to have 867 00:46:52,926 --> 00:46:56,195 also very deep knowledge. And um, yeah, that is something 868 00:46:56,195 --> 00:46:59,602 I would be concerned about. Yeah, I think that's a valid 869 00:46:59,602 --> 00:47:00,793 mental model, I would say, right? 870 00:47:00,793 --> 00:47:03,745 So, um, but one of the challenge, for people, right? 871 00:47:03,985 --> 00:47:06,835 'Cause these tools are so easy to access, right? 872 00:47:06,835 --> 00:47:10,195 And sometimes we kind of like inadvertently, uh, leak out 873 00:47:10,195 --> 00:47:11,905 something that is not supposed to be. 874 00:47:12,238 --> 00:47:14,650 And do you think just by subscribing maybe more like an 875 00:47:14,650 --> 00:47:17,722 enterprise or business plan, where they say we have zero 876 00:47:17,722 --> 00:47:20,885 retention of your data, should you still be worried with such 877 00:47:20,885 --> 00:47:22,907 a, I dunno, like an agreement or clause? 878 00:47:23,306 --> 00:47:27,182 Or if not, should everyone now start thinking about running 879 00:47:27,182 --> 00:47:30,530 model in-house? It depends on what you believe. 880 00:47:30,530 --> 00:47:34,751 I think if it depends what you believe as a CTO, how paranoid 881 00:47:34,751 --> 00:47:38,099 you are. Yeah, I think it's more on a 882 00:47:38,099 --> 00:47:40,379 paranoid your scale than on science. 883 00:47:40,709 --> 00:47:43,559 But if they say, what I would not do if they say we use your 884 00:47:43,559 --> 00:47:46,928 data and your source code for training, then probably I would 885 00:47:46,928 --> 00:47:48,894 not. And I have some deep knowledge. 886 00:47:48,914 --> 00:47:54,018 If I'm just a, I dunno. If I have a very, if I have no 887 00:47:54,018 --> 00:47:56,298 deep knowledge, I might not care too much. 888 00:47:56,298 --> 00:47:59,845 Like this coaching operations thing, I'm right. 889 00:47:59,845 --> 00:48:03,897 I don't care if Claude Code is training on that code because 890 00:48:03,897 --> 00:48:07,317 there is no really, really deep insights that I have. 891 00:48:07,651 --> 00:48:10,381 So it depends on your business and on your paranoia level. 892 00:48:10,381 --> 00:48:14,316 But on this very, we use your data to train our model, that's 893 00:48:14,316 --> 00:48:16,718 probably what I would not do, uh, as a company. 894 00:48:17,408 --> 00:48:19,058 And then it depends again on your paranoia. 895 00:48:19,058 --> 00:48:21,728 If you think, okay, you trust the company that they don't 896 00:48:21,728 --> 00:48:25,013 train if they tell you this. Or you say, I don't trust them, 897 00:48:25,013 --> 00:48:28,232 I run my own models. Currently, I probably would not 898 00:48:28,232 --> 00:48:31,772 run my own models because in the trade off between being 899 00:48:31,772 --> 00:48:36,975 competitive and running my own models, I would rather be more 900 00:48:36,975 --> 00:48:41,236 competitive, you know, and use better models than the ones I 901 00:48:41,236 --> 00:48:45,330 can run on my own. But perhaps I'm not sure where 902 00:48:45,330 --> 00:48:48,834 this is going and how programmer... like before Sonnet 903 00:48:48,834 --> 00:48:54,234 4.5, I would say like Opus 4.1, I think. 904 00:48:54,714 --> 00:48:58,854 I thought like the models kind of plateaued. 905 00:48:59,787 --> 00:49:04,247 I think Sonnet 4.5 is a big jump forward again for coding. 906 00:49:04,897 --> 00:49:08,258 So it's not plateauing. But if it starts plateauing, I 907 00:49:08,258 --> 00:49:10,400 would think more about using my own models. 908 00:49:10,789 --> 00:49:13,350 But I'm interested in running own models. 909 00:49:13,840 --> 00:49:15,850 It seems no one else is really interested in. 910 00:49:15,850 --> 00:49:19,305 There are no benchmarks. Like if I look at, on the 911 00:49:19,305 --> 00:49:22,869 internet, there are gazillion benchmarks for games like this 912 00:49:22,869 --> 00:49:27,590 hardware is running this games at 105 FPS, you know. 913 00:49:27,590 --> 00:49:31,898 But I'm not seeing a lot of benchmarks of this hardware is 914 00:49:31,898 --> 00:49:36,613 running this model at this size with 50 token per second. 915 00:49:36,973 --> 00:49:39,725 You know, that's not what I, there are here and there are 916 00:49:39,725 --> 00:49:42,073 some benchmark, but I don't see a lot of them. 917 00:49:42,763 --> 00:49:45,643 So I think there is not a lot of interest currently to do this. 918 00:49:45,643 --> 00:49:50,437 So for now, probably I would use better model than, uh, running 919 00:49:50,437 --> 00:49:54,533 stuff on my own. If I, but I have some clients or 920 00:49:54,533 --> 00:49:58,891 talk to people at conferences and clients who work, uh, in the 921 00:49:58,891 --> 00:50:02,599 defense industry. And, uh, well, they don't, they 922 00:50:02,599 --> 00:50:06,087 don't have that option. So it also depends on your 923 00:50:06,087 --> 00:50:09,708 business case, perhaps. Yeah, so I think we have to 924 00:50:09,708 --> 00:50:11,008 understand our own trade-offs, right? 925 00:50:11,032 --> 00:50:14,510 Our own situations and contexts. So yeah, I think what you 926 00:50:14,510 --> 00:50:17,375 mentioned the benchmark probably is not a lot available, right? 927 00:50:17,375 --> 00:50:20,233 Probably it's also like running the benchmark itself could be 928 00:50:20,233 --> 00:50:23,161 expensive, right? Plus you have to run a lot of, 929 00:50:23,161 --> 00:50:26,161 you know, like context, tokens, and all that in order to produce 930 00:50:26,161 --> 00:50:28,467 a meaningful result. So we're probably looking 931 00:50:28,467 --> 00:50:30,481 forward one day when we have all this available. 932 00:50:31,269 --> 00:50:34,427 I wanna go to the next thing, um, which kind of like ties back 933 00:50:34,427 --> 00:50:36,805 to the role of CTO itself, right? 934 00:50:36,955 --> 00:50:40,362 So definitely we can see the impact of AI to the software 935 00:50:40,362 --> 00:50:42,445 development team and organizations, but I wanna 936 00:50:42,445 --> 00:50:46,225 understand from you what is the impact of AI to a CTO, right? 937 00:50:46,645 --> 00:50:49,971 And again, I remember back then when we had this conversation, 938 00:50:49,971 --> 00:50:54,625 you said that AI has the potential of making CTO feel 939 00:50:54,625 --> 00:50:56,875 more creative and innovate, right? 940 00:50:57,364 --> 00:51:00,279 And maybe one year ahead after that, you know, conversation. 941 00:51:00,279 --> 00:51:03,369 What do you think is the impact of AI to a CTO, right? 942 00:51:03,669 --> 00:51:06,009 Is it something that CTO should leverage a lot more? 943 00:51:06,154 --> 00:51:09,375 And should CTO be hands-on coding again, with something 944 00:51:09,375 --> 00:51:11,739 that you didn't actually agree back then? 945 00:51:12,722 --> 00:51:15,177 You can write code as a CTO if you have some time. 946 00:51:15,207 --> 00:51:19,193 Like if everything works, like if all the operations works and 947 00:51:19,193 --> 00:51:23,843 you have great business impact as an executive and you have 948 00:51:23,843 --> 00:51:27,967 time left, well, and you enjoy coding, do write some code. 949 00:51:27,967 --> 00:51:33,717 So I don't have something against coding by itself. 950 00:51:33,717 --> 00:51:37,737 But usually I think you need to do a lot of other stuff before 951 00:51:37,737 --> 00:51:39,057 you can come back to coding again. 952 00:51:39,726 --> 00:51:43,725 Yes, I think I've discussed this with several clients. 953 00:51:44,351 --> 00:51:49,404 One, uh, One challenge of the CTO role is that you have not 954 00:51:49,404 --> 00:51:55,068 enough opportunities to shine. So if everything is well, no one 955 00:51:55,068 --> 00:51:58,298 recognizes you. If things go bad, everyone is up 956 00:51:58,298 --> 00:52:02,113 your neck. So that's I, uh, see the CTO 957 00:52:02,113 --> 00:52:05,608 role. And that's often the case 958 00:52:05,608 --> 00:52:10,181 because sometimes CTOs confuse their role with that of the, of 959 00:52:10,181 --> 00:52:13,677 a VP of engineering. You know, they see themselves as 960 00:52:13,677 --> 00:52:17,539 an execution machine which is more like being the VP of 961 00:52:17,539 --> 00:52:19,588 engineering. Whereas the CTO should have 962 00:52:19,588 --> 00:52:22,172 business impact and have visibility in the board. 963 00:52:22,714 --> 00:52:26,552 And everyone should be able to answer the question, why do we 964 00:52:26,552 --> 00:52:29,960 have a CTO on the board? You know, that that's, or in the 965 00:52:29,960 --> 00:52:32,969 management board. And I think AI is something, if 966 00:52:32,969 --> 00:52:37,252 you go into AI, if you do prototyping, if you also as a 967 00:52:37,252 --> 00:52:41,325 CTO, if you especially prototyping or show you can do a 968 00:52:41,325 --> 00:52:45,678 lot of like also the stuff like I said before, um, tying AI to 969 00:52:45,678 --> 00:52:50,108 your data warehouse and then asking the AI why are our best 970 00:52:50,108 --> 00:52:52,721 customers our best customers or something like this? 971 00:52:53,081 --> 00:52:56,456 You know, and they showed that to the CEO or to the top 972 00:52:56,456 --> 00:52:59,561 management board because you're the CTO, because you have a 973 00:52:59,561 --> 00:53:01,377 technical understanding of what is possible. 974 00:53:01,377 --> 00:53:04,347 You could do things earlier than other people, perhaps. 975 00:53:04,678 --> 00:53:08,414 And that's a way to shine. I think AI is a great 976 00:53:08,414 --> 00:53:11,741 opportunity for CTOs to shine in the top management. 977 00:53:12,169 --> 00:53:16,269 Something that was difficult from the time when, uh, tech and 978 00:53:16,269 --> 00:53:20,643 product was split in two and the shining things were taken away 979 00:53:20,643 --> 00:53:24,703 from the CTO, from that point on I think it was difficult to 980 00:53:24,703 --> 00:53:27,705 shine for a CTO. But I think AI brings it back. 981 00:53:27,997 --> 00:53:31,307 And you can easier shine. And I think, uh, rise and shine 982 00:53:31,307 --> 00:53:34,424 is something that's important for a CTO and for your career 983 00:53:34,424 --> 00:53:36,774 and for your success and for all of these things. 984 00:53:36,774 --> 00:53:39,324 So that's where AI is really helpful. 985 00:53:40,140 --> 00:53:41,379 Yeah, I think it's a great point, right? 986 00:53:41,484 --> 00:53:44,730 Because naturally if you're a good CTO yourself, right? 987 00:53:44,730 --> 00:53:47,302 So you would have a good understanding of the business, 988 00:53:47,302 --> 00:53:50,100 right, the business impact and also the potential of tech, 989 00:53:50,100 --> 00:53:52,212 right? And by having the ability to 990 00:53:52,212 --> 00:53:54,879 kind of like, I would say a creative thinking where you 991 00:53:54,879 --> 00:53:58,355 think which area that AI can help you and tie it back to the 992 00:53:58,355 --> 00:54:00,657 business impact. Yeah, probably you have a good 993 00:54:00,657 --> 00:54:03,487 chance to actually, uh, rise and shine within your organization 994 00:54:03,487 --> 00:54:07,014 to show something that would never be possible before, I 995 00:54:07,014 --> 00:54:08,794 guess. Like, because, uh, now AI simply 996 00:54:08,794 --> 00:54:12,107 can open up a lot of rooms for innovations and potential things 997 00:54:12,107 --> 00:54:15,994 that could be easily done. Before, probably we would worry 998 00:54:15,994 --> 00:54:17,794 about how we execute that, right? 999 00:54:17,794 --> 00:54:20,173 Because you probably need to develop code, you need more 1000 00:54:20,173 --> 00:54:22,572 people to help. So I think that's a very good 1001 00:54:22,572 --> 00:54:24,874 point. Are there other things that you 1002 00:54:24,874 --> 00:54:28,102 think a CTO would be able to leverage AI, uh, maybe day to 1003 00:54:28,102 --> 00:54:30,454 day from your conversation with your clients or maybe what 1004 00:54:30,454 --> 00:54:31,844 you've seen in the industry so far? 1005 00:54:33,649 --> 00:54:35,669 Otherwise, I haven't seen too much. 1006 00:54:36,129 --> 00:54:39,579 Um, we would need to talk again in a year, I think. 1007 00:54:39,853 --> 00:54:43,249 Currently, most of my, really as I laid out in the beginning, 1008 00:54:43,249 --> 00:54:47,336 most of my clients, which is also to be sure, um, to be 1009 00:54:47,336 --> 00:54:50,976 clear, my clients are from like five to a hundred developers. 1010 00:54:50,976 --> 00:54:54,075 So that's my kind of expertise and that's where I'm working 1011 00:54:54,075 --> 00:54:56,647 with. So I don't have insights into 1012 00:54:56,647 --> 00:54:59,911 companies with 500 developers. So the dev of 5,000. 1013 00:55:00,301 --> 00:55:03,777 So these have different setups. And so there might be something 1014 00:55:03,777 --> 00:55:07,385 true that I don't know. So just to be clear, it's, 1015 00:55:07,385 --> 00:55:11,201 that's where I have expertise. But in this area, they are just 1016 00:55:11,201 --> 00:55:14,905 struggling to adopt AI in a meaningful way with a lot of 1017 00:55:14,905 --> 00:55:18,145 chaotic things going on from developers and demand from 1018 00:55:18,145 --> 00:55:22,233 developers, faster, slower business like that's where they 1019 00:55:22,233 --> 00:55:25,457 currently are, most of my clients are currently with AI. 1020 00:55:26,500 --> 00:55:29,268 Yeah, I think that sums up a very good situation for 1021 00:55:29,268 --> 00:55:31,403 everyone, right? Scrambling to adopt AI, but at 1022 00:55:31,403 --> 00:55:34,823 the same time also trying to rationalize the benefits of 1023 00:55:34,823 --> 00:55:37,103 using AI. Some people talk positive about 1024 00:55:37,103 --> 00:55:39,628 AI, some talk about negative. I think, yeah, we are in this 1025 00:55:39,628 --> 00:55:42,158 like midst of chaos, I would say, like, try to rationalize 1026 00:55:42,158 --> 00:55:44,547 AI. So we have talked a lot about, 1027 00:55:44,547 --> 00:55:47,184 you know, AI. Are there any other things that 1028 00:55:47,184 --> 00:55:50,796 you think, uh, we should also, um, discuss before we move on to 1029 00:55:50,796 --> 00:55:54,731 the last questions that I have? I really, I mean, there are all 1030 00:55:54,731 --> 00:55:57,947 these stuff on CTO things. Uh, they, a lot of the stuff 1031 00:55:57,947 --> 00:56:00,296 that's, that has been there is not going away. 1032 00:56:00,296 --> 00:56:05,258 So AI is getting on top of a lot of these other things, and makes 1033 00:56:05,258 --> 00:56:09,246 some things more difficult. Like, um, most of my clients 1034 00:56:09,246 --> 00:56:13,267 have also struggled with the proper organization and roles 1035 00:56:13,267 --> 00:56:18,017 and accountability and all of these topics that my clients 1036 00:56:18,017 --> 00:56:21,145 struggle with. And I feel like AI is just 1037 00:56:21,145 --> 00:56:24,092 making it more complicated and more, sorry, there is a sun now 1038 00:56:24,092 --> 00:56:26,721 shining in a window on the opposite house and it's shining 1039 00:56:26,721 --> 00:56:30,015 directly in the face. Um, um, so, so it makes things 1040 00:56:30,015 --> 00:56:32,595 more difficult what the role looks should look like, what the 1041 00:56:32,595 --> 00:56:35,234 organization should look like. So it's, so that's, but it's 1042 00:56:35,234 --> 00:56:38,015 really, really about a lot about AI and strategy. 1043 00:56:38,345 --> 00:56:42,594 Um, so I don't have anything more interesting than what we 1044 00:56:42,594 --> 00:56:44,540 discussed last time, I think, beside AI. 1045 00:56:45,620 --> 00:56:48,241 So yeah, definitely what you mentioned is valid, right? 1046 00:56:48,241 --> 00:56:52,580 So for people who wants to become a good CTO, I would still 1047 00:56:52,580 --> 00:56:55,794 highly suggest, uh, reading your book, right, the Amazing CTO's 1048 00:56:55,794 --> 00:56:58,896 Missing Manual, right? Because a lot of stuff I think 1049 00:56:58,896 --> 00:57:02,092 would still be relevant, if not more relevant because like AI, 1050 00:57:02,092 --> 00:57:05,409 yeah, we could amplify certain things where you can get more 1051 00:57:05,409 --> 00:57:08,009 things done probably. But there are other aspects that 1052 00:57:08,009 --> 00:57:09,626 AI currently is not able to leverage. 1053 00:57:09,626 --> 00:57:11,470 So things like what you mentioned, accountability, 1054 00:57:11,470 --> 00:57:15,249 providing good strategy, and not to be swamped into the 1055 00:57:15,249 --> 00:57:17,792 day-to-day tasks, right? So always thinking strategically 1056 00:57:17,792 --> 00:57:21,077 and helping people to grow. So I think that's still kind of 1057 00:57:21,077 --> 00:57:22,994 like part of the big responsibility of the CTO. 1058 00:57:23,874 --> 00:57:26,304 So Stephan, uh, it's been a great conversation. 1059 00:57:26,334 --> 00:57:30,095 Um, before we wrap up, I would like to ask you the same 1060 00:57:30,095 --> 00:57:32,934 question I asked last time, which I call the three technical 1061 00:57:32,934 --> 00:57:35,214 leadership wisdom. So maybe, uh, after this 1062 00:57:35,214 --> 00:57:38,274 conversation, do you have some version of, uh, wisdom that you 1063 00:57:38,274 --> 00:57:42,058 would like to convey this time? Yeah, I have three. 1064 00:57:42,830 --> 00:57:45,680 Two are about AI and one is not about AI. 1065 00:57:45,680 --> 00:57:50,200 The first one is, be a leader. Just today or yesterday, I read 1066 00:57:50,200 --> 00:57:52,983 something on LinkedIn, which is people make, sometimes make it 1067 00:57:52,983 --> 00:57:55,278 very complicated, the concept of leadership, sometimes people 1068 00:57:55,278 --> 00:57:57,710 make it very complicated. I think leadership is simple and 1069 00:57:57,710 --> 00:58:01,157 hard at the same time, but it's simple in a way that you want 1070 00:58:01,157 --> 00:58:04,982 to, you think people should move somewhere, organizations should 1071 00:58:04,982 --> 00:58:08,774 move somewhere, and then you get people to get there, you know, 1072 00:58:08,774 --> 00:58:10,761 to move there. That's I think, is a leader. 1073 00:58:10,881 --> 00:58:13,701 You decide we should go there, let's go there. 1074 00:58:13,731 --> 00:58:16,777 That's leadership. Um, and, uh, and something that 1075 00:58:16,777 --> 00:58:20,768 a lot of CTOs fall into trap, uh, CTOs fall into is this, uh, 1076 00:58:20,768 --> 00:58:23,971 servant leader. They identify as servant leader, 1077 00:58:23,971 --> 00:58:28,435 which is kind of fine if they would not concentrate on being a 1078 00:58:28,435 --> 00:58:29,845 servant instead of being a leader. 1079 00:58:29,845 --> 00:58:33,060 So that's a trap. Uh, I think servant leader is a 1080 00:58:33,060 --> 00:58:34,925 trap. Uh, so be a leader. 1081 00:58:34,925 --> 00:58:37,859 That's important. Uh, the second thing is AI is 1082 00:58:37,859 --> 00:58:39,885 not software. We're going to move towards... 1083 00:58:39,885 --> 00:58:43,947 I strongly believe we're going to move to a setup that favors 1084 00:58:43,947 --> 00:58:48,010 AI without source code. And, uh, you should, as a 1085 00:58:48,010 --> 00:58:51,250 leader, a tech leader, you should think today what that 1086 00:58:51,250 --> 00:58:55,464 means for you and what you can already, what microservices you 1087 00:58:55,464 --> 00:58:58,780 might already be able to replace from code to AI. 1088 00:58:59,676 --> 00:59:02,250 The third one is not a leadership thing per se, but 1089 00:59:02,250 --> 00:59:05,614 something I think you should push in every developer, because 1090 00:59:05,614 --> 00:59:09,018 it's, I think the biggest prompting mistake people make is 1091 00:59:09,018 --> 00:59:13,290 you should iterate on the plan and not iterate on the prompt. 1092 00:59:13,945 --> 00:59:16,513 I use Claude Code. Probably it's the same in other 1093 00:59:16,513 --> 00:59:19,116 tools. Uh, you can have plan mode and 1094 00:59:19,116 --> 00:59:21,848 then you go into plan mode and click plan mode. 1095 00:59:21,848 --> 00:59:26,486 And then you plan and plan and plan and plan, until you tell 1096 00:59:26,486 --> 00:59:29,955 the Claude also check, recheck the documentation, recheck the 1097 00:59:29,955 --> 00:59:33,823 code if your plan works. And then after some iterations 1098 00:59:33,823 --> 00:59:36,915 on the plan, you tell Claude Code to go. 1099 00:59:37,365 --> 00:59:40,245 Whereas what I see is people iterating on the prompt. 1100 00:59:40,245 --> 00:59:43,065 They say, do this, and then the outcome is not what they expect. 1101 00:59:43,065 --> 00:59:45,345 They say, no, don't, don't do this, do that. 1102 00:59:45,825 --> 00:59:48,315 And then it's still not working. And then they say, do this. 1103 00:59:48,495 --> 00:59:52,060 So they iterate on the execution, and then that's not 1104 00:59:52,060 --> 00:59:54,187 working. You know, they just get deeper 1105 00:59:54,187 --> 00:59:57,789 and deeper and deep in the woods, to a point where the 1106 00:59:57,789 --> 01:00:01,944 agent can't find out anymore and they're lost in the, lost in the 1107 01:00:01,944 --> 01:00:04,790 woods. So iterate on the plan, but 1108 01:00:04,790 --> 01:00:08,215 don't iterate on the prompt. That's the third one. 1109 01:00:09,463 --> 01:00:12,029 Nice tips, right? So for people who still fiddle 1110 01:00:12,029 --> 01:00:14,143 with the prompt, uh, including myself, right? 1111 01:00:14,143 --> 01:00:16,608 Sometimes I just try to make the prompt better. 1112 01:00:16,813 --> 01:00:20,179 Uh, but yeah, I think, uh, making the plan better sounds 1113 01:00:20,179 --> 01:00:22,566 better because you iterate on the small thing, right? 1114 01:00:22,566 --> 01:00:25,111 You break down the things that you wanna do in chunks, right? 1115 01:00:25,171 --> 01:00:28,096 And making sure that you know the plan for those, uh, 1116 01:00:28,096 --> 01:00:30,638 iteration looks good. And, you know, you let AI tries 1117 01:00:30,638 --> 01:00:34,236 to solve that one by one. So definitely pro tips for using 1118 01:00:34,236 --> 01:00:36,679 AI. So Stephan, if people want to 1119 01:00:36,679 --> 01:00:39,631 continue, uh, this conversation, they would like to ask you more 1120 01:00:39,631 --> 01:00:42,487 things or they expect an updated version of your book, where they 1121 01:00:42,487 --> 01:00:45,555 can find you online? They should search for me on 1122 01:00:45,555 --> 01:00:47,083 LinkedIn. They can find me on LinkedIn. 1123 01:00:47,173 --> 01:00:51,613 Then they can connect and ask me whatever they want or tell me 1124 01:00:51,613 --> 01:00:55,813 their opinion or whatever share. Uh, I'm very open. 1125 01:00:56,113 --> 01:00:58,663 So the primary method would be going to LinkedIn. 1126 01:00:59,023 --> 01:01:02,653 The other thing is I have a website, which is called 1127 01:01:02,653 --> 01:01:04,723 amazingcto.com. People can also go there. 1128 01:01:05,891 --> 01:01:07,576 Right. Thank you again for your time, 1129 01:01:07,576 --> 01:01:09,476 Stephan. I'm very excited, you know, 1130 01:01:09,476 --> 01:01:12,647 having this possibility to learn about possibilities of using AI, 1131 01:01:12,647 --> 01:01:15,467 and especially for those CTOs out there who still are 1132 01:01:15,467 --> 01:01:18,503 struggling how to adopt AI successfully, on how to think 1133 01:01:18,503 --> 01:01:20,351 about using AI much more effectively. 1134 01:01:20,591 --> 01:01:22,571 So I hope this conversation inspires you. 1135 01:01:22,901 --> 01:01:25,071 So thanks again, Stephan. Thanks. 1136 01:01:25,071 --> 01:01:28,145 Thanks, Henry, for having me. And I hope I can be back in a 1137 01:01:28,145 --> 01:01:30,040 year. Looking forward for that.