1 00:01:45,235 --> 00:01:46,955 Hello guys. Welcome back to another new 2 00:01:46,955 --> 00:01:48,685 episode of the Tech Lead Journal podcast. 3 00:01:48,685 --> 00:01:52,093 Today I'm really excited to have one of my favorite authors in 4 00:01:52,093 --> 00:01:55,135 the show today. His name is Gene Kim. 5 00:01:55,225 --> 00:01:59,374 So he's the founder of IT Revolution and also the author, 6 00:01:59,374 --> 00:02:02,440 co-authors of few, several popular DevOps books, starting 7 00:02:02,440 --> 00:02:06,246 from The Phoenix Project which I really love, by the way when the 8 00:02:06,246 --> 00:02:09,549 book came out. So since then he wrote Unicorn 9 00:02:09,549 --> 00:02:12,270 Project, The DevOps Handbook, Accelerate, and so many other 10 00:02:12,270 --> 00:02:14,961 books. And recently he just published a 11 00:02:14,961 --> 00:02:17,459 book titled Vibe Coding, with Steve Yegge. 12 00:02:17,839 --> 00:02:21,102 So I know that we all have heard about the term vibe coding. 13 00:02:21,642 --> 00:02:24,606 Let's try to, you know, peel a little bit more from Gene's 14 00:02:24,606 --> 00:02:26,112 point of view what vibe coding is. 15 00:02:26,232 --> 00:02:28,392 So Gene, thank you so much for this opportunity. 16 00:02:28,422 --> 00:02:30,122 Looking forward for our conversation. 17 00:02:31,122 --> 00:02:33,106 Oh, Henry, thank you so much for having me. 18 00:02:33,106 --> 00:02:35,460 And I was mentioning how much fun I had listening to your 19 00:02:35,460 --> 00:02:38,392 episodes with Simon Wardley, John Willis and so many of the 20 00:02:38,392 --> 00:02:40,426 IT Rev authors. And, yeah. 21 00:02:40,426 --> 00:02:43,235 And the common passion that so many of us have around this 22 00:02:43,235 --> 00:02:45,977 kinda the miracles that vibe coding allows us to do 23 00:02:45,977 --> 00:02:49,424 especially for people who maybe stepped away from coding as part 24 00:02:49,424 --> 00:02:52,299 of their daily work for, you know, years or maybe even over a 25 00:02:52,299 --> 00:02:54,555 decade So, yeah, I'm so delighted to be here. 26 00:02:55,590 --> 00:02:57,642 Right. So Gene, in the beginning, I 27 00:02:57,642 --> 00:03:00,194 always love to ask my guests maybe sharing a little bit more 28 00:03:00,194 --> 00:03:01,965 from your career, right? The turning points that you 29 00:03:01,965 --> 00:03:04,882 think we all can learn from you. Yeah, for sure. 30 00:03:05,032 --> 00:03:09,821 Yeah I've spent the last 26 years studying high performing 31 00:03:09,821 --> 00:03:11,995 technology organizations. And that was a journey that 32 00:03:11,995 --> 00:03:14,905 started back in '99 when I was a CTO and technical founder of a 33 00:03:14,905 --> 00:03:16,943 company called Tripwire in the information security and 34 00:03:16,943 --> 00:03:19,455 compliance space. And so the journey started when 35 00:03:19,455 --> 00:03:21,957 we wanted to study these high performing organizations that 36 00:03:21,957 --> 00:03:24,367 simultaneously had the best project due date performance and 37 00:03:24,367 --> 00:03:26,347 development, the best operational reliability and 38 00:03:26,347 --> 00:03:29,615 stability in ops, and as well as the best posture security and 39 00:03:29,615 --> 00:03:32,055 compliance. So the reason was that we wanted 40 00:03:32,055 --> 00:03:35,045 to understand how those amazing organizations made their good to 41 00:03:35,045 --> 00:03:37,029 great transformation so that other organizations could 42 00:03:37,029 --> 00:03:38,499 replicate those amazing outcomes. 43 00:03:39,159 --> 00:03:42,175 And so I left Tripwire in 2010 and I spent three years working 44 00:03:42,175 --> 00:03:44,513 on the Phoenix Project that came on 2013. 45 00:03:44,513 --> 00:03:48,363 And it was just such a surprise and I was so grateful that it 46 00:03:48,363 --> 00:03:51,111 led me to the middle of the DevOps movement which I think, 47 00:03:51,111 --> 00:03:53,663 you know, kind of upended how technology organizations work, 48 00:03:53,663 --> 00:03:55,652 right? It was, you know, Dev versus QA 49 00:03:55,652 --> 00:03:57,611 versus Ops, right? And DevOps is really about how 50 00:03:57,611 --> 00:03:59,837 do you get those organizations to work together towards a 51 00:03:59,837 --> 00:04:03,605 common objectives. And, you know, I had so much fun 52 00:04:03,605 --> 00:04:06,791 working on with Jez Humble and Dr. Nicole Forsgren on 53 00:04:06,791 --> 00:04:09,647 understanding what are the technical practices and 54 00:04:09,647 --> 00:04:12,268 architectural practices and cultural norms that allow us to, 55 00:04:12,268 --> 00:04:13,553 you know, create great performance. 56 00:04:13,583 --> 00:04:16,651 And what's kind of funny is that for the DevOps movement, you 57 00:04:16,651 --> 00:04:19,053 know, as part of that journey I ran to Adrian Cockroft. 58 00:04:19,202 --> 00:04:21,519 He led the Netflix cloud transformation. 59 00:04:22,079 --> 00:04:25,165 You know, there he mentioned a term called NoOps, you know, 60 00:04:25,165 --> 00:04:28,187 back 12 years ago. And, he recently made a post, 61 00:04:28,187 --> 00:04:31,137 you know, on LinkedIn that said basically I'm the director of 62 00:04:31,137 --> 00:04:34,665 running a bunch of cloud agents. And, you know, 12 years ago, we 63 00:04:34,665 --> 00:04:37,954 talked about NoOps and now we should be talking about NoDev, 64 00:04:37,954 --> 00:04:41,286 you know. So NoOps is not true back then 65 00:04:41,286 --> 00:04:44,839 nor is NoDev true now. But I mean it just shows how 66 00:04:44,839 --> 00:04:46,987 vibe coding is actually bringing, you know, coding 67 00:04:46,987 --> 00:04:49,742 within the reach of so many people who haven't coded in 68 00:04:49,742 --> 00:04:51,362 years or, you know, even decades. 69 00:04:51,662 --> 00:04:53,910 It's been amazing to see how much excitement it generates and 70 00:04:53,910 --> 00:04:55,818 how many people are building so many cool things. 71 00:04:55,818 --> 00:04:57,228 And, I'm sorry, I skipped a step. 72 00:04:57,228 --> 00:05:00,549 So the reason I got sucked into this incredible adventure, the 73 00:05:00,549 --> 00:05:04,287 adventure of a lifetime, was meeting Steve Yegge last June. 74 00:05:04,617 --> 00:05:06,747 And so I've been quoting him for over 11 years. 75 00:05:06,747 --> 00:05:08,517 I've had so much admiration for his work. 76 00:05:08,959 --> 00:05:12,399 And when we met, you know, it turns out like not only, you 77 00:05:12,399 --> 00:05:14,901 know, did we have a common passion around engineering 78 00:05:14,901 --> 00:05:17,672 practices and architecture as embodied by the famous, you 79 00:05:17,672 --> 00:05:20,952 know, his depiction of the Jeff Bezos memo in the early two 80 00:05:20,952 --> 00:05:23,141 thousands. But we had this common love of 81 00:05:23,141 --> 00:05:25,669 using AI for coding. And the vibe coding book came 82 00:05:25,669 --> 00:05:29,349 out a month ago and I love the book because it tells the story 83 00:05:29,349 --> 00:05:32,637 of two people who both thought that the days of coding, the 84 00:05:32,637 --> 00:05:34,925 best days of coding were behind them, but it's now actually 85 00:05:34,925 --> 00:05:36,963 ahead of them. And this is despite Steve having 86 00:05:36,963 --> 00:05:39,969 written over a million lines of production code at Amazon for 87 00:05:39,969 --> 00:05:42,945 eight years, at Google for 13 years, as part of having written 88 00:05:42,945 --> 00:05:46,165 clouds, Code Search, you know, the most popular tool inside of 89 00:05:46,165 --> 00:05:49,124 Google. Anyway, so fun to be able to 90 00:05:49,124 --> 00:05:52,508 chronicle how vibe coding has changed our lives, has made 91 00:05:52,508 --> 00:05:55,811 things faster, more ambitious, we can do things alone and 92 00:05:55,811 --> 00:05:58,730 honestly fun and optionality. I know we'll talk about more 93 00:05:58,730 --> 00:06:01,874 about that later but, holy cow, that this has been the funnest 94 00:06:01,874 --> 00:06:04,859 things I've ever gotten to work on in my entire career. 95 00:06:05,769 --> 00:06:07,855 Yeah, thanks for sharing your story, right? 96 00:06:07,855 --> 00:06:09,865 So I'm sure that we will learn a lot. 97 00:06:10,075 --> 00:06:13,496 So maybe the first place, I saw that Phoenix Project was like 98 00:06:13,496 --> 00:06:16,691 your first big, you know, popular books, I would say. 99 00:06:17,291 --> 00:06:20,905 And it's written when the DevOps era is, you know, was kind of 100 00:06:20,905 --> 00:06:23,346 like apparent. And I'm sure like vibe coding 101 00:06:23,346 --> 00:06:26,856 might be the era as well when people are, you know, realizing 102 00:06:26,856 --> 00:06:30,269 something big is going to, you know, revolutionize the software 103 00:06:30,269 --> 00:06:33,292 engineering industry, right? You have written these so many 104 00:06:33,292 --> 00:06:35,416 books that has sold over like million copies. 105 00:06:35,446 --> 00:06:39,310 So anything that you learn like from people, you know, how did 106 00:06:39,310 --> 00:06:42,256 you come up with those books in the first place? 107 00:06:42,891 --> 00:06:44,451 Oh yeah. Well, I mean the Phoenix 108 00:06:44,451 --> 00:06:47,661 Project, I had written actually a book called, in 2004 called 109 00:06:47,661 --> 00:06:50,631 The Visible Ops Handbook. You know, how to create world 110 00:06:50,631 --> 00:06:53,091 class operational processes using ITIL, of all things. 111 00:06:53,091 --> 00:06:55,582 But like, you know, people laughed at ITIL, you know, maybe 112 00:06:55,582 --> 00:06:57,308 these days. But it's like, you know, it was 113 00:06:57,308 --> 00:06:59,264 the best we had to sort of describe operational processes, 114 00:06:59,264 --> 00:07:01,764 and I thought that was very important back in 2004. 115 00:07:02,244 --> 00:07:04,276 And, you know, a lot of those learnings are brought into the 116 00:07:04,276 --> 00:07:07,118 Phoenix Project journey which was written by the same, you 117 00:07:07,118 --> 00:07:09,024 know, author team as Visible Ops Handbook. 118 00:07:09,694 --> 00:07:13,210 But in The Phoenix Project is really based on one of our 119 00:07:13,210 --> 00:07:16,449 favorite books which was The Goal by Dr. Eliyahu Goldratt. 120 00:07:16,669 --> 00:07:18,319 And so that book is somewhere behind me. 121 00:07:18,649 --> 00:07:21,179 And it's a novel about a manufacturing plant manager who 122 00:07:21,179 --> 00:07:24,239 has to fix his cost and due date issues in 90 days, otherwise it 123 00:07:24,239 --> 00:07:27,229 shut the plant down. And we love that book so much. 124 00:07:27,229 --> 00:07:30,037 And, we, Kevin Behr and I, we talked about it with Dr. 125 00:07:30,037 --> 00:07:32,265 Goldratt. I mean he was so generous with 126 00:07:32,265 --> 00:07:34,801 the time and, you know, he changed our careers. 127 00:07:35,521 --> 00:07:38,047 And, you know, our goal was to see if we can write that book 128 00:07:38,047 --> 00:07:39,727 but for the IT context, you know. 129 00:07:39,727 --> 00:07:42,950 For hopeless situation that IT operations is in. 130 00:07:43,010 --> 00:07:45,572 You know, if you can't create good working relationships with 131 00:07:45,572 --> 00:07:48,272 QA and development. And one of the things that we 132 00:07:48,272 --> 00:07:51,983 wanted to show was that IT operations looks so tactical, 133 00:07:51,983 --> 00:07:54,365 right? I mean it looks like it's just a 134 00:07:54,365 --> 00:07:57,023 very tactical activity. And yet if it is what's in 135 00:07:57,023 --> 00:07:59,302 between, you know, your most important development projects 136 00:07:59,302 --> 00:08:02,769 and your customers, well, then it's actually as strategic as 137 00:08:02,769 --> 00:08:04,505 development. And so our goal was to really 138 00:08:04,505 --> 00:08:07,167 try to elevate that cause. And I just, I cannot tell you 139 00:08:07,167 --> 00:08:09,607 how delighted and honored I was that, you know, that that book 140 00:08:09,607 --> 00:08:12,432 really became sort of like the flag of which, you know, so much 141 00:08:12,432 --> 00:08:13,932 of the DevOps community rallied around. 142 00:08:13,992 --> 00:08:16,104 And so we wrote a book called The DevOps Handbook after that, 143 00:08:16,104 --> 00:08:18,942 you know, to really be the nonfiction guide, right? 144 00:08:19,062 --> 00:08:21,210 The prescriptive guide to say, all right, how does one 145 00:08:21,210 --> 00:08:23,370 actually, you know, go from deploying once a year which 146 00:08:23,370 --> 00:08:25,372 everyone did back then, you know, to deploying multiple 147 00:08:25,372 --> 00:08:27,432 times a day, right? Or even hundreds of thousands of 148 00:08:27,432 --> 00:08:29,825 times a day. You know, certainly not unheard 149 00:08:29,825 --> 00:08:31,818 of now, right? Many have, most great 150 00:08:31,818 --> 00:08:33,619 organizations deployed multiple times a day. 151 00:08:34,580 --> 00:08:36,803 Yeah. And I saw recently you also have 152 00:08:36,803 --> 00:08:39,956 a series, like a graphic novels, a series of the Phoenix Project. 153 00:08:39,956 --> 00:08:42,686 So tell us a little bit more about that project or that book. 154 00:08:43,260 --> 00:08:45,780 Yeah, that, a couple people just had a tremendous amount of 155 00:08:45,780 --> 00:08:49,260 passion for doing that And it has been really fun to work on. 156 00:08:49,710 --> 00:08:52,678 In fact, it's been really fun to kind of go through, go back to 157 00:08:52,678 --> 00:08:55,833 the Phoenix Project universe and it's been really invigorating 158 00:08:55,833 --> 00:08:58,295 and fun. And I, you know, I guess 159 00:08:58,295 --> 00:09:01,919 honestly, I have I'm a little bit in awe of the book just 160 00:09:01,919 --> 00:09:05,251 because, I'm not sure I could have written that today. 161 00:09:05,611 --> 00:09:10,291 I don't think I was as angry now as back then where I just felt 162 00:09:10,291 --> 00:09:13,210 like, you know, the book really represented like a decade long 163 00:09:13,210 --> 00:09:15,866 plus moral crusade of like, you know, getting people to see 164 00:09:15,866 --> 00:09:18,664 something, right? And I'm just, I'll be honest I'm 165 00:09:18,664 --> 00:09:23,321 a bit in awe of like the way that, you know, the story shows 166 00:09:23,321 --> 00:09:25,965 that how, you know, critical operations is, right? 167 00:09:25,965 --> 00:09:30,116 I mean there was one, scene that I remember where, it was so 168 00:09:30,116 --> 00:09:32,332 good. The inventory management systems 169 00:09:32,332 --> 00:09:34,809 went down, the order entry systems went down, and basically 170 00:09:34,809 --> 00:09:37,518 it was the whole scene was designed to make sure that 171 00:09:37,518 --> 00:09:40,537 everything that was reported on to the board in terms of the 172 00:09:40,537 --> 00:09:43,938 quarterly financial report, they couldn't verify account balances 173 00:09:43,938 --> 00:09:46,096 or values. And so we're like, all right, 174 00:09:46,096 --> 00:09:48,616 what would need to go wrong to basically make sure that, you 175 00:09:48,616 --> 00:09:51,186 know, they can't count what's in the inventory, what's sold, 176 00:09:51,186 --> 00:09:53,935 what's not sold. And I just remember reading that 177 00:09:53,935 --> 00:09:56,756 I was like man, that is a, that is a bad day. 178 00:09:57,896 --> 00:10:01,392 And a lot of that was informed by working with the audit 179 00:10:01,392 --> 00:10:04,110 community for so long. Anyway it was, the question was 180 00:10:04,110 --> 00:10:06,394 like how did it feel to go through it. 181 00:10:06,394 --> 00:10:08,884 And I was just like it was really really fun and to really 182 00:10:08,884 --> 00:10:12,522 work with the graphic artist. He was a DC comics guy. 183 00:10:13,082 --> 00:10:16,498 And so he does a lot of work with storyboarding for movies in 184 00:10:16,498 --> 00:10:18,398 the UK. And so it's just really cool to 185 00:10:18,398 --> 00:10:22,170 see kind of how critical it is, you know, to as storyboarding is 186 00:10:22,170 --> 00:10:25,834 to, you know for any movie. If you can't have a good 187 00:10:25,834 --> 00:10:28,843 storyboard, it is very unlikely that you're gonna have a good 188 00:10:28,843 --> 00:10:31,741 movie just because it's often the one of the most primary 189 00:10:31,741 --> 00:10:34,501 communication vehicles between the director's vision in their 190 00:10:34,501 --> 00:10:37,339 heads, right? And, you know, all of the 191 00:10:37,339 --> 00:10:40,201 costuming and the actors or production, lighting, right? 192 00:10:40,201 --> 00:10:43,457 It's basically that's what is a primary communication vehicle. 193 00:10:44,972 --> 00:10:47,827 Wow, sounds really cool working with the DC comic artist, right? 194 00:10:47,827 --> 00:10:50,347 So. I'm sure when, if people are 195 00:10:50,347 --> 00:10:53,429 interested in reading The Phoenix Project, you know, just 196 00:10:53,429 --> 00:10:56,621 text, and you know, now that we have the graphics novel, I'm 197 00:10:56,621 --> 00:10:58,957 sure it's gonna be more interesting and fun. 198 00:11:00,757 --> 00:11:03,979 So let's move to, you know, the vibe coding thing that we wanna 199 00:11:03,979 --> 00:11:06,151 talk about today. So maybe in the first place, 200 00:11:06,151 --> 00:11:09,265 let's share your a-ha moments. Because in your book you 201 00:11:09,265 --> 00:11:12,835 mentioned, that you actually got introduced to vibe coding and 202 00:11:12,835 --> 00:11:15,921 loving it so much through, you know, your experience coding, 203 00:11:15,921 --> 00:11:19,057 you know, a project that you did, that you didn't even 204 00:11:19,057 --> 00:11:20,737 foresee you would work on, right? 205 00:11:20,977 --> 00:11:23,047 So tell us a little bit more your a-ha moments. 206 00:11:23,797 --> 00:11:27,291 You know, it's funny. So, you know, for me, this 207 00:11:27,291 --> 00:11:31,543 problem I've been wanting to solve for almost 12, 13 years 208 00:11:31,543 --> 00:11:34,849 was I listened to a lot of podcasts and videos. 209 00:11:34,969 --> 00:11:36,739 And so every time I hear something interesting, I take a 210 00:11:36,739 --> 00:11:38,962 screenshot of it. And, you know, in the hopes that 211 00:11:38,962 --> 00:11:41,557 I would go back to it and maybe write about it or listen to it 212 00:11:41,557 --> 00:11:43,081 again or write a citation, look something up. 213 00:11:43,471 --> 00:11:47,221 And so it turns out I have like 3,500 photos in my camera roll 214 00:11:47,221 --> 00:11:49,850 of just screenshots. And how many times have I 215 00:11:49,850 --> 00:11:52,494 actually gone back and looked at them and tried to do something 216 00:11:52,494 --> 00:11:54,869 with it? It's like five, because it's so 217 00:11:54,869 --> 00:11:58,022 hard. And so I, um, I remember when 218 00:11:58,022 --> 00:12:02,340 ChatGPT came out and, you know, I was marveling at the way that 219 00:12:02,340 --> 00:12:05,757 you could just give it a YouTube screenshot and, you know, it 220 00:12:05,757 --> 00:12:07,787 could actually detect where what the play time was right. 221 00:12:07,787 --> 00:12:09,518 And once you have the play time you can go into the transcript 222 00:12:09,518 --> 00:12:12,059 and you can auto extract a whole bunch of stuff. 223 00:12:12,449 --> 00:12:16,095 But for YouTube often they don't show the current playtime, 224 00:12:16,095 --> 00:12:17,585 right? They only show the red progress 225 00:12:17,585 --> 00:12:18,519 bar, right? And I was like, ugh. 226 00:12:18,979 --> 00:12:22,649 So I just as my first kind of non-trivial problem, I remember 227 00:12:22,649 --> 00:12:27,350 going into ChatGPT, this is January 2024, and typing in I'm 228 00:12:27,350 --> 00:12:30,737 gonna give you a, write a Clojure program, right? 229 00:12:30,737 --> 00:12:34,084 A program that runs on the JVM, that takes a YouTube screenshot 230 00:12:34,084 --> 00:12:38,477 and I want you to march up the left hand side of the image, 231 00:12:38,477 --> 00:12:40,738 find the red, you know, RGB color. 232 00:12:40,888 --> 00:12:43,514 And once you find it, march to the right, and, you know, so you 233 00:12:43,514 --> 00:12:45,798 can compute the percentage complete of the video. 234 00:12:46,448 --> 00:12:50,980 And it worked the first time. And it used Java Libraries I've 235 00:12:50,980 --> 00:12:54,202 never used. I haven't done 2D programming in 236 00:12:54,202 --> 00:12:57,104 25+ years. I mean last time I did it was in 237 00:12:57,104 --> 00:13:00,596 C++ on Microsoft Windows. And so I, you know, I could and 238 00:13:00,596 --> 00:13:02,648 it worked, right? And that would've taken me days, 239 00:13:02,648 --> 00:13:05,487 right, to understand like, all right, how does one use the Java 240 00:13:05,487 --> 00:13:06,981 AWT library, blah, blah, blah, right? 241 00:13:06,981 --> 00:13:10,091 I mean it's just not worth it. And the fact that it worked, I 242 00:13:10,091 --> 00:13:11,785 mean my jaw hit the floor, right? 243 00:13:11,785 --> 00:13:15,690 And so, I mean, I just started reinvigorating my love of 244 00:13:15,690 --> 00:13:18,581 programming and just increased the ambition of things that I 245 00:13:18,581 --> 00:13:21,824 could actually do reasonably. And then I meet Steve Yegge. 246 00:13:21,990 --> 00:13:24,090 Right after we talked, he published the Death of the 247 00:13:24,090 --> 00:13:26,222 Junior Developer post. And one of the things that and 248 00:13:26,222 --> 00:13:28,344 he talked to the conference early on called the Enterprise 249 00:13:28,344 --> 00:13:31,200 Technology Leadership Summit about his learning since then. 250 00:13:31,781 --> 00:13:32,861 You know, he had made these offers. 251 00:13:32,861 --> 00:13:34,691 Like, hey, how about we pair program together? 252 00:13:35,411 --> 00:13:38,485 So we did that in September and that just changed my life, 253 00:13:38,485 --> 00:13:40,091 right? If someone like Steve Yegge, you 254 00:13:40,091 --> 00:13:42,441 know, says let's pair together, you know, you say yes. 255 00:13:43,001 --> 00:13:44,891 And he said, look, what problem would you like to work on? 256 00:13:45,041 --> 00:13:47,171 And it was easy. It was like, all right. 257 00:13:47,231 --> 00:13:50,640 I have a pile of these videos and transcripts and YouTube 258 00:13:50,640 --> 00:13:53,236 videos. How about we make a something, 259 00:13:53,236 --> 00:13:56,527 right, that could transform those into like little video 260 00:13:56,527 --> 00:13:59,389 excerpts, you know, with captions overlaid that I could 261 00:13:59,389 --> 00:14:02,940 post on social media. And we finished in 47 minutes. 262 00:14:03,060 --> 00:14:05,478 I mean, it was just then, you know, the whole time he's giving 263 00:14:05,478 --> 00:14:06,942 me advice. He's like, man, Gene, you're 264 00:14:06,942 --> 00:14:08,970 doing an awful lot of typing. Stop typing. 265 00:14:09,030 --> 00:14:12,599 Just lean on the AI more. And it was, I posted the 266 00:14:12,599 --> 00:14:15,740 entirety of the video but that changed my life. 267 00:14:15,740 --> 00:14:18,765 That too changed my life. And, you know, we started 268 00:14:18,765 --> 00:14:22,080 talking almost a couple times a week after that. 269 00:14:22,080 --> 00:14:25,313 And then, you know, the topic, you know, he said, Steve said 270 00:14:25,313 --> 00:14:28,392 that he was interested in writing a book on what became 271 00:14:28,392 --> 00:14:31,237 known as vibe coding back. We think it was gonna be called 272 00:14:31,237 --> 00:14:35,327 chat oriented programming. And I just found, you know, like 273 00:14:35,327 --> 00:14:38,657 whenever you finish a book, it takes a little time before you 274 00:14:38,657 --> 00:14:40,932 ever want to get involved in another book project. 275 00:14:40,932 --> 00:14:44,305 But when Steve mentioned that, I heard the words leaving my mouth 276 00:14:44,305 --> 00:14:49,358 like I would love to help. And the best if you can, it was 277 00:14:49,358 --> 00:14:52,208 the most fun I've ever had on a book project. 278 00:14:52,306 --> 00:14:55,110 I learned so much and the fact it was happening in this 279 00:14:55,110 --> 00:14:59,126 whirlwind of technology change, right, was incredible. 280 00:14:59,126 --> 00:15:02,372 But I felt, I feel like... and I love some of the people who read 281 00:15:02,372 --> 00:15:05,483 it who said, yeah, this is a somewhat, this is a book that 282 00:15:05,483 --> 00:15:07,226 will not change so much with technology. 283 00:15:07,226 --> 00:15:09,236 It has a shelf life of at least a couple years, right? 284 00:15:09,236 --> 00:15:12,662 And so that's ideally the book we want to write is that 285 00:15:12,662 --> 00:15:16,342 something that just can give the direction, stick to principles, 286 00:15:16,342 --> 00:15:19,331 and practices, right, that is independent of the technology 287 00:15:19,331 --> 00:15:21,983 used. But I will make this admission 288 00:15:21,983 --> 00:15:25,719 that when Claude Code came out in January, we rewrote the book 289 00:15:25,719 --> 00:15:27,585 from scratch. Or was it March? 290 00:15:27,805 --> 00:15:30,449 Yeah I mean, you know, basically a lot of the book we had to 291 00:15:30,449 --> 00:15:33,560 rewrite just because, you know, that was a future where the AI's 292 00:15:33,560 --> 00:15:35,882 not telling you what to do and telling you what to type. 293 00:15:36,182 --> 00:15:38,902 You are telling it to do the typing and it to do the running. 294 00:15:39,182 --> 00:15:40,952 And that just changed everything for us. 295 00:15:41,767 --> 00:15:43,300 Wow. In fact, can I tell you a quick 296 00:15:43,300 --> 00:15:45,220 story on that? Yeah, definitely. 297 00:15:45,685 --> 00:15:47,938 Yeah. I mean, I remember for the book 298 00:15:47,938 --> 00:15:51,608 to manage all the research, for most books for the last 12 299 00:15:51,608 --> 00:15:54,097 years, I've been using Trello to sort of like keep track of my 300 00:15:54,097 --> 00:15:56,467 tasks and research files and so forth. 301 00:15:57,077 --> 00:15:59,197 You know, I just wanted to have. Everything that I starred inside 302 00:15:59,197 --> 00:16:01,067 of Twitter to show up as a Trello card. 303 00:16:01,067 --> 00:16:02,675 That worked. But then I actually wanted to 304 00:16:02,675 --> 00:16:03,857 summarize the articles and so forth. 305 00:16:04,353 --> 00:16:06,645 And I was getting all these I was trying to create attachments 306 00:16:06,645 --> 00:16:09,325 for the first time and I kept on getting these API errors. 307 00:16:09,782 --> 00:16:12,264 And, you know, for 45 minutes I'm trying this, I'm trying 308 00:16:12,264 --> 00:16:14,474 that, you know. I'm asking Claude Code, I'm 309 00:16:14,474 --> 00:16:17,299 asking ChatGPT and it's like type this, type that, try this, 310 00:16:17,299 --> 00:16:19,319 try that. And I was so frustrated. 311 00:16:19,349 --> 00:16:22,479 I said screw you. You do it. 312 00:16:22,479 --> 00:16:26,429 You've got a repo open. And 45 seconds later it had 313 00:16:26,429 --> 00:16:29,729 tried five, six things. And it said oh I got it working. 314 00:16:30,529 --> 00:16:33,794 And that also changes your life, right? 315 00:16:33,794 --> 00:16:37,408 It's like it makes you recognize like your job is not to be 316 00:16:37,408 --> 00:16:39,404 bossed around by the AI and type things for it. 317 00:16:39,734 --> 00:16:43,106 Your job is to create this situation where it can see the 318 00:16:43,106 --> 00:16:45,591 results of his work and it can loop around, right? 319 00:16:45,591 --> 00:16:48,021 And it can iterate until you get to the goal. 320 00:16:48,551 --> 00:16:51,681 And, you know, Steve had a very similar reaction when he was 321 00:16:51,681 --> 00:16:57,059 using Puppeteer to, you know, do a TypeScript, React front-end 322 00:16:57,059 --> 00:16:59,594 thing. And, you know, he described a 323 00:16:59,594 --> 00:17:02,540 situation where suddenly it could see what it's doing and it 324 00:17:02,540 --> 00:17:04,955 was like stop motion video where like all the elements were 325 00:17:04,955 --> 00:17:07,406 starting to be fixed by itself with no human intervention. 326 00:17:07,886 --> 00:17:10,856 And I think we both had a sense of awe right at that moment. 327 00:17:10,856 --> 00:17:13,497 It's like I can't believe what is now possible. 328 00:17:14,015 --> 00:17:16,279 Yeah, sorry for rambling but it's like, yeah, just so 329 00:17:16,279 --> 00:17:19,323 exciting. Yeah, very exciting to hear, you 330 00:17:19,323 --> 00:17:22,240 know, your sharing, right? Because I'm sure many developers 331 00:17:22,240 --> 00:17:25,910 out there, including the, you know, senior ones who probably 332 00:17:25,910 --> 00:17:30,551 have loved coding in the past, but they don't have time now to 333 00:17:30,551 --> 00:17:33,650 actually start coding again. I'm sure we realize that the 334 00:17:33,650 --> 00:17:37,090 power of AI can actually make programming fun again and you 335 00:17:37,090 --> 00:17:39,750 kind of like reinvigorate, you know, your passion for coding. 336 00:17:39,770 --> 00:17:42,914 Similar to me as well. So I think your story definitely 337 00:17:42,914 --> 00:17:45,266 is quite fascinating, right? Especially the impact, right? 338 00:17:45,266 --> 00:17:48,052 How you could write such a program that you didn't, you 339 00:17:48,052 --> 00:17:50,851 know, think you would do before in just minutes, right? 340 00:17:51,241 --> 00:17:53,695 in one day. So I think that's always very, 341 00:17:53,695 --> 00:17:56,046 very exciting and eye opener. Yeah. 342 00:17:56,046 --> 00:17:57,486 In fact, I mean you talk about eye-opening. 343 00:17:57,516 --> 00:18:00,410 Yeah I think what's interesting to me, I love the first part of 344 00:18:00,410 --> 00:18:02,514 the book because I think lovingly written and it 345 00:18:02,514 --> 00:18:05,636 describes - you know, for me I got my graduate degree in 346 00:18:05,636 --> 00:18:09,836 computer science in 1995. I was studying compilers and 347 00:18:09,836 --> 00:18:12,312 high-speed networking. And I wrote my last line of 348 00:18:12,312 --> 00:18:14,865 production code in 2001 after, you know, creating, co-founding 349 00:18:14,865 --> 00:18:17,561 Tripwire, right? And as someone said, that we 350 00:18:17,561 --> 00:18:21,761 quoted in the book, I was now in the Pew Stack – PowerPoint, 351 00:18:21,761 --> 00:18:24,461 Word, and Excel, right? I mean it's yeah. 352 00:18:24,461 --> 00:18:26,429 And like many technology leaders, right, that's their 353 00:18:26,429 --> 00:18:28,409 path, right? And so although I fell in love 354 00:18:28,409 --> 00:18:31,173 with programming again in 2016, you know, it was a tough slog, 355 00:18:31,173 --> 00:18:32,661 right? And you know, everything was 356 00:18:32,661 --> 00:18:35,361 hard. And so just to be able to write 357 00:18:35,361 --> 00:18:38,643 all these tools that I wanna write like a PowerPoint viewer, 358 00:18:38,643 --> 00:18:41,496 a PDF viewer, tools to help the writing process. 359 00:18:41,676 --> 00:18:45,243 What did I just write today of? I wrote a mail merge tool, 360 00:18:45,243 --> 00:18:48,653 right, these, all these tools it takes me less than 10 minutes to 361 00:18:48,653 --> 00:18:50,669 build. And so Steve Yegge had the same 362 00:18:50,669 --> 00:18:52,004 thing. He actually gave up code. 363 00:18:52,004 --> 00:18:54,694 He told his doctor last January that he was done with coding. 364 00:18:55,184 --> 00:18:58,234 His hands hurt, right? It was just, took too much time, 365 00:18:58,234 --> 00:19:00,074 right, and the juice wasn't worth the squeeze. 366 00:19:00,074 --> 00:19:01,964 He's been working on his game for 35 years. 367 00:19:02,414 --> 00:19:05,806 And he looked at the, you know, person-centuries of work and he 368 00:19:05,806 --> 00:19:08,241 said it is just not worth it. You got, there's this phrase 369 00:19:08,241 --> 00:19:10,166 that he said that's really stuck with me. 370 00:19:10,166 --> 00:19:13,360 He said, you know, when you get to be our age, right, in our 371 00:19:13,360 --> 00:19:16,084 fifties, you gotta be picky about what you work on because 372 00:19:16,084 --> 00:19:18,514 there's only so many five year projects that you have left in 373 00:19:18,514 --> 00:19:21,204 you. But after Claude Code came out, 374 00:19:21,204 --> 00:19:23,416 you know, suddenly everything became possible, right? 375 00:19:23,416 --> 00:19:26,595 Like something that would've taken a year maybe might take a 376 00:19:26,595 --> 00:19:28,847 week, right? You know, things that would've 377 00:19:28,847 --> 00:19:30,631 taken months might take, you know, days, right? 378 00:19:30,631 --> 00:19:33,594 And so that change as he wrote, that changed everything. 379 00:19:34,168 --> 00:19:38,663 Yeah it's just difficult to overstate just how much changed 380 00:19:38,663 --> 00:19:42,450 our perspective of like we're not coding again to now we're 381 00:19:42,450 --> 00:19:45,546 coding more than ever. Yeah. 382 00:19:46,116 --> 00:19:49,026 So let's go to, you know, the definition of vibe coding. 383 00:19:49,026 --> 00:19:51,798 You mentioned that, you know, you had to change the title of 384 00:19:51,798 --> 00:19:54,082 the book. Vibe coding was the term 385 00:19:54,082 --> 00:19:56,644 introduced by Andrej Karpathy. And since then people are 386 00:19:56,644 --> 00:19:59,706 talking, raving about it. I think your book has some 387 00:19:59,706 --> 00:20:02,574 definition that probably not aligned to, fully to what 388 00:20:02,574 --> 00:20:04,674 Andrej's, you know, citations, right? 389 00:20:04,944 --> 00:20:07,542 So maybe tell us a little bit more from your perspective, what 390 00:20:07,542 --> 00:20:09,923 do you mean by vibe coding? Yeah. 391 00:20:09,923 --> 00:20:13,766 Our definition of vibe coding is anytime you're coding and not 392 00:20:13,766 --> 00:20:18,935 typing by hand. And I think so much of this is 393 00:20:18,935 --> 00:20:21,643 motivated by and inspired by Dr. Erik Meijer. 394 00:20:21,663 --> 00:20:25,403 Oh, so you know that video that I did in 47 minutes. 395 00:20:25,403 --> 00:20:28,869 The video I was using was Dr. Erik Meijer, a talk that he gave 396 00:20:28,869 --> 00:20:31,391 last year. So Dr. Erik Meijer, he's 397 00:20:31,391 --> 00:20:33,509 probably considered one of the greatest programming language 398 00:20:33,509 --> 00:20:37,057 designers of all time. He was part of Visual Basic, C#, 399 00:20:37,057 --> 00:20:39,313 very much contributed to Haskell, LINQ. 400 00:20:39,335 --> 00:20:42,125 He was behind the Hack programming language at Meta, 401 00:20:42,125 --> 00:20:45,165 you know, where it is a type-safe version of PHP. 402 00:20:45,755 --> 00:20:48,191 And he said the days of writing code by hand are coming to an 403 00:20:48,191 --> 00:20:51,355 end. And yeah I was just like, oh, it 404 00:20:51,355 --> 00:20:57,295 it just made us it like it gave us confidence, right? 405 00:20:57,320 --> 00:21:01,012 That it's like, yeah, if he can say it, someone who spent his 406 00:21:01,012 --> 00:21:03,430 entire career designing programming languages to be used 407 00:21:03,430 --> 00:21:06,272 by humans, what would make him say that the days of writing 408 00:21:06,272 --> 00:21:07,760 code by hand are coming to an end? 409 00:21:07,910 --> 00:21:10,046 And I think it's because, you know, he believes very much the 410 00:21:10,046 --> 00:21:12,586 same things that we believed and he was very much like the totem 411 00:21:12,586 --> 00:21:15,334 in our book. And so then when Dr. Karpathy 412 00:21:15,334 --> 00:21:19,298 introduced the word vibe coding, brought to the universe in late 413 00:21:19,298 --> 00:21:22,591 December, you know, we definitely, we thought that that 414 00:21:22,591 --> 00:21:23,821 gave voice to what we're trying to do. 415 00:21:24,353 --> 00:21:28,724 And so Dario Amodei, the CEO and co-founder of Anthropic, he 416 00:21:28,724 --> 00:21:31,067 wrote the forward to the book and he gives this wonderful 417 00:21:31,067 --> 00:21:33,379 definition. He said we don't write code by 418 00:21:33,379 --> 00:21:35,021 hand. Instead it comes from an 419 00:21:35,021 --> 00:21:37,835 iterative conversation with the AI and the AI is writing the 420 00:21:37,835 --> 00:21:40,542 code for you. And I thought it was just a, and 421 00:21:40,542 --> 00:21:43,896 he said it's, on one hand, it's a very wonderful term because it 422 00:21:43,896 --> 00:21:46,496 evokes how different it is in like hunching over a computer 423 00:21:46,496 --> 00:21:49,218 and typing things out. But he said, on the other hand, 424 00:21:49,218 --> 00:21:51,732 it's kind of a terrible term because it sounds kind of jokey 425 00:21:51,732 --> 00:21:54,612 like not a serious endeavor. But he said, you know, Anthropic 426 00:21:54,612 --> 00:21:58,376 is like the only game in town. And I just thought it was such a 427 00:21:58,376 --> 00:22:00,715 great definition. And so yeah, our definition is 428 00:22:00,715 --> 00:22:04,187 like anytime when you are not typing out code by hand which is 429 00:22:04,187 --> 00:22:08,092 like developing photos in a dark room, right, where you actually 430 00:22:08,092 --> 00:22:10,539 go into a dark room that you're in and you have to dip into the 431 00:22:10,539 --> 00:22:12,602 chemicals. So no one does that anymore. 432 00:22:12,622 --> 00:22:15,199 And I think even since the book has come out, you know, Steve 433 00:22:15,199 --> 00:22:18,156 and I have talked a lot to each other about like how we're 434 00:22:18,156 --> 00:22:20,483 writing so much more, creating so many more things. 435 00:22:20,483 --> 00:22:25,330 And we have such confidence, we've gained such confidence in, 436 00:22:25,330 --> 00:22:27,044 you know, what the AI can generate. 437 00:22:27,074 --> 00:22:29,764 And we're very cognizant of what it's bad at, right? 438 00:22:29,764 --> 00:22:32,468 But, you know, with the right engineering mentality with the 439 00:22:32,468 --> 00:22:34,944 right feedback loops with the right architecture, you know, it 440 00:22:34,944 --> 00:22:38,158 is possible to do things safely. Right. 441 00:22:38,437 --> 00:22:41,301 Interestingly some people even, don't even type the prompt these 442 00:22:41,301 --> 00:22:43,003 days. They will just, you know, 443 00:22:43,003 --> 00:22:44,984 transcribe, use a transcript, yeah, transcription tool. 444 00:22:45,334 --> 00:22:48,876 And, you know, it's like, writing code through, you know, 445 00:22:48,876 --> 00:22:51,675 the conversation with a computer, with the AI itself. 446 00:22:52,419 --> 00:22:54,891 Oh my gosh, yeah. I was just got off a call with 447 00:22:54,891 --> 00:22:56,826 John Rauser. He's a senior director of 448 00:22:56,826 --> 00:22:59,808 security at Cisco. Senior director of engineering 449 00:22:59,808 --> 00:23:03,413 at Cisco Security. And so he convinced his SVP to 450 00:23:03,413 --> 00:23:06,679 basically mandate a hundred of the top leaders vibe code 451 00:23:06,679 --> 00:23:09,695 something and put into production within a quarter. 452 00:23:09,695 --> 00:23:13,057 So that ended 11 days ago. And so trying to generate a 453 00:23:13,057 --> 00:23:16,176 survey like we did in the early days of the State of DevOps 454 00:23:16,176 --> 00:23:18,604 research to understand, all right, how did it feel like? 455 00:23:18,670 --> 00:23:22,300 What did you accomplish? I wanna know what the vector is. 456 00:23:22,300 --> 00:23:23,890 What was their - Did they have an a-ha moment? 457 00:23:23,890 --> 00:23:26,590 So vector is a direction and a magnitude. 458 00:23:26,590 --> 00:23:29,420 So I wanna understand like to what degree are they excited and 459 00:23:29,420 --> 00:23:31,850 what are they gonna do, right, the direction? 460 00:23:32,508 --> 00:23:34,712 I started in ChatGPT saying all right, here's kinda the survey 461 00:23:34,712 --> 00:23:36,362 instrument I want to design, right? 462 00:23:36,392 --> 00:23:39,257 Here's kind of the, you know, vague idea of the instruments. 463 00:23:39,497 --> 00:23:41,087 And it, you know, it was interesting. 464 00:23:41,087 --> 00:23:43,785 But then I put it into the Claude Code and I said read all 465 00:23:43,785 --> 00:23:46,358 this, right? You know, create a one pager of 466 00:23:46,358 --> 00:23:48,513 the survey design and it showed me something. 467 00:23:48,513 --> 00:23:50,737 I'm like, eh, not quite, you know, like... 468 00:23:50,737 --> 00:23:52,693 And it just through conversational iteration, after 469 00:23:52,693 --> 00:23:55,713 five rounds I had something I was ready to show John. 470 00:23:55,803 --> 00:23:58,437 And we were so excited. And then I was like, okay, take 471 00:23:58,437 --> 00:24:01,223 a first cut of the questions. And, you know, and it's just so 472 00:24:01,223 --> 00:24:05,010 different than the way we did it 12 years ago, right? 473 00:24:05,010 --> 00:24:07,392 And obviously I'm gonna go in there and I'm gonna, I'm... you 474 00:24:07,392 --> 00:24:10,820 know, this is a we're gonna have a hundred people spend time, you 475 00:24:10,820 --> 00:24:12,450 know, maybe 15 minutes going through the survey. 476 00:24:12,450 --> 00:24:16,020 So I wanna make sure that it's right, that it counts, that it's 477 00:24:16,170 --> 00:24:18,218 gonna take us and hopefully generate some interesting 478 00:24:18,218 --> 00:24:20,370 research findings. But I'm responsible. 479 00:24:21,160 --> 00:24:23,250 But am I actually writing every question by hand? 480 00:24:23,250 --> 00:24:25,362 Am I making every Likert question on a scale of one to 481 00:24:25,362 --> 00:24:27,240 seven, to what degree do you blah blah blah blah? 482 00:24:27,690 --> 00:24:29,790 No. And it's just amazing to see, 483 00:24:29,790 --> 00:24:31,350 like it's not just coding, right? 484 00:24:31,350 --> 00:24:34,080 It's just even survey design and survey analysis. 485 00:24:34,080 --> 00:24:35,790 It's just amazing how much AI can help you. 486 00:24:36,833 --> 00:24:39,012 Yeah. In your book, you mentioned this 487 00:24:39,012 --> 00:24:40,986 term, FAAFO, right, as the benefits of vibe coding. 488 00:24:41,006 --> 00:24:44,376 So you've mentioned a lot of things, you know, making 489 00:24:44,376 --> 00:24:47,335 programming fun, obviously faster generation of code, 490 00:24:47,335 --> 00:24:50,043 right? So FAAFO kind of like summarized 491 00:24:50,043 --> 00:24:53,391 the things that people could benefit by using, you know, AI 492 00:24:53,391 --> 00:24:55,577 and vibe coding. So tell us a little bit more 493 00:24:55,577 --> 00:24:58,467 about this FAAFO, because I'm sure some people might not, you 494 00:24:58,467 --> 00:25:01,840 know, realizing the other elements of FAAFO apart from the 495 00:25:01,840 --> 00:25:04,445 Fast and the Fun part, yeah. Oh, exactly. 496 00:25:04,445 --> 00:25:09,675 I just, Steve and I, we did a lot of thinking around December 497 00:25:09,675 --> 00:25:12,786 about, you know, how do you articulate the value of it. 498 00:25:12,786 --> 00:25:16,225 And yeah we came up with the FAAFO construct which we just we 499 00:25:16,225 --> 00:25:18,885 get endless delight out of. And interestingly and 500 00:25:18,885 --> 00:25:21,691 conveniently, the first dimension is Faster. 501 00:25:21,851 --> 00:25:25,327 And our claim is that that is actually the least interesting 502 00:25:25,327 --> 00:25:28,046 dimension of value. But there's no doubt that like, 503 00:25:28,046 --> 00:25:30,354 you know, whether it's the YouTube percentage complete 504 00:25:30,354 --> 00:25:33,406 generator, whatever, like AI let you do things faster for sure. 505 00:25:34,036 --> 00:25:37,457 But I think far more powerful is the notion of Ambitious is that 506 00:25:37,547 --> 00:25:40,760 you can now do things that you could not have done before, 507 00:25:40,760 --> 00:25:43,195 right? And I think there's actually two 508 00:25:43,195 --> 00:25:45,679 extreme two parts of the spectrum for this. 509 00:25:46,087 --> 00:25:47,677 Take that video excerpt generator. 510 00:25:48,398 --> 00:25:51,428 It might have taken me three, four days if I'm feeling kind of 511 00:25:51,428 --> 00:25:55,523 generous, right, and confident. But, you know, like anytime you 512 00:25:55,523 --> 00:25:58,328 work with FFmpeg, you know, things take longer, right? 513 00:25:58,328 --> 00:26:01,348 I mean it's like simple things can be very very difficult. 514 00:26:01,762 --> 00:26:06,241 And so I think the... if you ask why have I never generated the 515 00:26:06,241 --> 00:26:09,263 video generator, you know, despite the fact that it only 516 00:26:09,263 --> 00:26:13,825 took 47 minutes using AI, I think economist would say that I 517 00:26:13,825 --> 00:26:16,842 was making a conscious unconscious decision that the 518 00:26:16,842 --> 00:26:18,462 juice wasn't worth the squeeze, right? 519 00:26:18,462 --> 00:26:20,780 The effort wasn't worth the return. 520 00:26:21,605 --> 00:26:23,471 It's not worth three days, right, this other thing they 521 00:26:23,471 --> 00:26:26,189 should be working on. But when it takes 47 minutes, 522 00:26:26,189 --> 00:26:27,605 right, suddenly that changes everything. 523 00:26:28,183 --> 00:26:30,743 Like suddenly the impossible becomes possible. 524 00:26:30,963 --> 00:26:33,053 I think that's the, really the point of Ambitious. 525 00:26:33,303 --> 00:26:36,889 The other thing that's really interesting is that the small 526 00:26:36,889 --> 00:26:39,608 things become free. The Claude Code team talked 527 00:26:39,608 --> 00:26:42,443 about like there's one thing they're counting is how many 528 00:26:42,443 --> 00:26:45,654 times does a customer issue come up and they just fix it on the 529 00:26:45,654 --> 00:26:47,497 spot. They don't create a Jira ticket 530 00:26:47,497 --> 00:26:50,257 and then debate about it and like, you know, prioritize and 531 00:26:50,257 --> 00:26:51,853 triage it. They just do it. 532 00:26:52,303 --> 00:26:54,509 And I think that is also very interesting, right? 533 00:26:54,509 --> 00:26:57,529 So like impossible becomes possible, and the small things 534 00:26:57,529 --> 00:26:59,544 become free. Especially we can launch a whole 535 00:26:59,544 --> 00:27:02,739 bunch of agents in parallel. And so that just fundamentally 536 00:27:02,739 --> 00:27:04,439 changes the economics of software. 537 00:27:05,122 --> 00:27:08,113 So you got Faster you got more Ambitious and you got, oh, 538 00:27:08,113 --> 00:27:11,075 Autonomous and Alone. One of the things that I find so 539 00:27:11,075 --> 00:27:14,979 exciting doing these vibe coding workshops, you know, is that 540 00:27:14,979 --> 00:27:19,288 everyone, everyone has things they want developers to do but 541 00:27:19,288 --> 00:27:23,180 it's just not above the line. And so like, you know, so you 542 00:27:23,180 --> 00:27:26,710 got software leaders, you got designers, you got UX people 543 00:27:26,710 --> 00:27:30,004 like they're just pestering developers to try, please, 544 00:27:30,024 --> 00:27:31,524 please solve this problem for me, right? 545 00:27:31,704 --> 00:27:33,534 Or the Tableau people. Or somebody, right? 546 00:27:33,534 --> 00:27:36,581 Someone, you know, is, they want something done but they, you 547 00:27:36,581 --> 00:27:39,159 have to sort of cajole and politic and escalate, you know, 548 00:27:39,159 --> 00:27:41,927 to get these things done. And suddenly you can do them by 549 00:27:41,927 --> 00:27:44,058 yourself. Or, you know, maybe it's a team 550 00:27:44,058 --> 00:27:46,656 and you just you've been trying to get the other functional 551 00:27:46,656 --> 00:27:49,110 specialty to help. And, you know, suddenly you 552 00:27:49,110 --> 00:27:52,526 don't need the, you know, the CI/CD expert or the Kubernetes 553 00:27:52,526 --> 00:27:54,713 expert. You can just do it yourself. 554 00:27:54,993 --> 00:27:57,926 And, you know, that significantly changes, again, 555 00:27:57,926 --> 00:28:01,264 the economics of software, because there's two costs that 556 00:28:01,264 --> 00:28:03,694 are being lessened. One is the coordination cost. 557 00:28:04,114 --> 00:28:07,374 In other words, you don't have to prioritize and synchronize 558 00:28:07,374 --> 00:28:10,929 and agree on priorities. And then, you know, code 559 00:28:10,929 --> 00:28:12,477 together, deploy together, test together, right? 560 00:28:12,477 --> 00:28:15,793 You can do it yourself. The other one is like even if 561 00:28:15,793 --> 00:28:18,333 you get someone else to believe that a problem is as important 562 00:28:18,333 --> 00:28:20,817 as you think it is, they can't read your mind. 563 00:28:20,817 --> 00:28:24,273 And it turns out these LLMs are so good at intermediating, you 564 00:28:24,273 --> 00:28:27,724 know, between multiple parties, you know, to work together. 565 00:28:27,934 --> 00:28:30,787 And then so it's like that's, the both of those I think allow 566 00:28:30,787 --> 00:28:35,083 Autonomous and Alone. And so that's FAAF. 567 00:28:35,253 --> 00:28:37,993 The second F is for Fun. Like, oh my gosh. 568 00:28:38,203 --> 00:28:40,813 There's something very addictive about vibe coding. 569 00:28:41,203 --> 00:28:43,858 In fact, have you found that like once you're vibe coding, 570 00:28:43,858 --> 00:28:45,539 creating things it's very difficult to stop? 571 00:28:45,980 --> 00:28:48,580 Yes, yes it is. Especially once you can see, I 572 00:28:48,580 --> 00:28:50,500 mean, coming back to the ambitious thing, right? 573 00:28:50,500 --> 00:28:53,178 Once you can see the possibility and the amount of things that 574 00:28:53,178 --> 00:28:56,020 you can do in such a short time, it's very addictive. 575 00:28:56,020 --> 00:28:57,734 Definitely. Oh my gosh, yeah. 576 00:28:57,734 --> 00:28:59,419 And, yeah. So like creating something that 577 00:28:59,419 --> 00:29:00,914 you had in your head and it's working. 578 00:29:01,274 --> 00:29:03,296 I mean, it's just incredible. And like even the unfun things 579 00:29:03,296 --> 00:29:06,078 become sort of fun like, you know, reducing test times, like 580 00:29:06,078 --> 00:29:09,239 injecting test data into a function so that it's not 581 00:29:09,239 --> 00:29:11,084 hitting a database. Like even that's fun, right, you 582 00:29:11,084 --> 00:29:12,914 know? It's just, yeah. 583 00:29:13,284 --> 00:29:18,559 And it just shows us a dopamine release that is, I think, very 584 00:29:18,559 --> 00:29:20,987 addictive. And, you know, good or bad that 585 00:29:20,987 --> 00:29:23,294 changes the dynamics of how we create software. 586 00:29:23,294 --> 00:29:26,498 And I'll take fun any day over boring or tedious or like 587 00:29:26,498 --> 00:29:28,114 maddening. So that's awesome. 588 00:29:28,264 --> 00:29:29,404 And then the last one's Option value. 589 00:29:29,464 --> 00:29:32,841 You know, you can explore so much more of the decision space. 590 00:29:32,841 --> 00:29:35,529 You know, you can try, you know, three four ideas where, you 591 00:29:35,529 --> 00:29:38,301 know, you'd be, before you'd be lucky if you can get one. 592 00:29:38,481 --> 00:29:40,381 And it's so expensive and unhappy making that you're just 593 00:29:40,381 --> 00:29:43,125 kind of stuck with it. So just recently I tried 594 00:29:43,125 --> 00:29:44,721 something three ways. What did I do? 595 00:29:44,721 --> 00:29:47,572 It was like oh I need to increase the time to first 596 00:29:47,572 --> 00:29:49,858 render. Like let's try one time by 597 00:29:49,858 --> 00:29:51,601 delaying the generation of the vagal light. 598 00:29:51,601 --> 00:29:53,401 Let's try another one by something, something, right? 599 00:29:54,181 --> 00:29:58,235 And I just sent them all off in parallel and, you know, after 600 00:29:58,235 --> 00:30:00,687 some exploration I decided to choose one of them. 601 00:30:00,707 --> 00:30:02,217 And I don't even remember what it is. 602 00:30:02,487 --> 00:30:06,032 Anyway so like Option value is actually what allows, they say 603 00:30:06,032 --> 00:30:08,828 it is actually a huge creator of economic value. 604 00:30:08,888 --> 00:30:12,436 In fact, Option value is what allows modularity to be so 605 00:30:12,436 --> 00:30:14,961 powerful. And so there's just no doubt 606 00:30:14,961 --> 00:30:17,828 that, you know, the more variations that you can try, 607 00:30:17,828 --> 00:30:20,891 ideally in parallel, right, that is just one of the biggest 608 00:30:20,891 --> 00:30:22,374 levers for economic value creation. 609 00:30:23,426 --> 00:30:25,499 Yeah. And I would say also optionality 610 00:30:25,499 --> 00:30:27,883 doesn't mean that all the options come from your head, 611 00:30:27,883 --> 00:30:30,419 right? But AI itself can actually gives 612 00:30:30,419 --> 00:30:34,265 you more options such that you can think which one is best for 613 00:30:34,265 --> 00:30:36,196 your situations, right? Yeah. 614 00:30:36,196 --> 00:30:38,651 Totally, right? It is on the one hand like you 615 00:30:38,651 --> 00:30:41,691 got an infinite army of summer interns but it also is your 616 00:30:41,691 --> 00:30:44,251 world class expert in almost every category, right? 617 00:30:44,251 --> 00:30:47,447 And that is a what an amazing tool to have. 618 00:30:48,514 --> 00:30:51,437 Yeah, you mentioned that a lot of people now can create, you 619 00:30:51,437 --> 00:30:54,014 know, ambitious projects. Even you saying that small 620 00:30:54,014 --> 00:30:57,755 things now could be free or people, non-coders can actually 621 00:30:57,755 --> 00:31:00,432 solve issues. And obviously some people have 622 00:31:00,432 --> 00:31:03,409 these concerns, right? So when a lot of people can do 623 00:31:03,409 --> 00:31:06,329 vibe coding now to generate software and create, you know, 624 00:31:06,329 --> 00:31:09,818 code, do you think that software developer's role will kind of 625 00:31:09,818 --> 00:31:12,958 like be decreasing? Or do you think also like more 626 00:31:12,958 --> 00:31:16,073 people would become a coder so to speak, right, vibe coder? 627 00:31:16,523 --> 00:31:18,585 So tell us a little bit more on this because some, so many 628 00:31:18,585 --> 00:31:20,234 people have different perspectives on this, and I 629 00:31:20,234 --> 00:31:21,978 would love to hear from you what your thoughts are. 630 00:31:23,097 --> 00:31:25,327 Yeah I was just earlier talking to Dr. Jeffrey Liker. 631 00:31:25,377 --> 00:31:29,004 So he wrote the Toyota Way, I mean just very famous in the 632 00:31:29,004 --> 00:31:31,201 Lean circles. And he actually said, Gene, 633 00:31:31,201 --> 00:31:32,821 you're very, you're confusing me. 634 00:31:32,821 --> 00:31:35,438 On the one hand, you say that it's just a typewriter, right? 635 00:31:35,438 --> 00:31:37,238 I mean it's helping you do all these things, right? 636 00:31:37,478 --> 00:31:40,365 And you also say that it is capable of, you know, helping 637 00:31:40,365 --> 00:31:43,305 you be creative and so forth. Like they seem contradictory. 638 00:31:43,305 --> 00:31:45,595 And I was like, no. Dr. Liker, it was like, no, no. 639 00:31:45,595 --> 00:31:47,545 I don't see them as contradictory at all. 640 00:31:48,225 --> 00:31:51,076 And, because to him, coming from the manufacturing space, I mean 641 00:31:51,076 --> 00:31:54,419 I think when he hears automation, you know, the magic 642 00:31:54,419 --> 00:31:57,495 typewriter can type, you know, 10,000 words a minute, you know, 643 00:31:57,495 --> 00:32:01,015 that's reduction of jobs. And I think what I was trying to 644 00:32:01,015 --> 00:32:04,456 convey to him was that, uh, you know, the notion of Jevon's 645 00:32:04,456 --> 00:32:06,277 Paradox. The notion that if you reduce 646 00:32:06,277 --> 00:32:09,187 the cost of software production, do you end up with 1/10th of 647 00:32:09,187 --> 00:32:11,780 number of developers or do you end up with 10 times more 648 00:32:11,780 --> 00:32:14,568 software? And I just, in my head, I just 649 00:32:14,568 --> 00:32:17,836 cannot conceive of a world where we're like, oh, we're done 650 00:32:17,836 --> 00:32:20,608 writing software. We just, we've had our, we've 651 00:32:20,608 --> 00:32:23,408 had our appetite filled and we don't need any more software. 652 00:32:23,408 --> 00:32:26,681 I just, I cannot conceive of that possibility at, like at 653 00:32:26,681 --> 00:32:29,561 all. And so to me what that means is 654 00:32:29,561 --> 00:32:32,702 that we're gonna end up with more developers and solving 655 00:32:32,702 --> 00:32:35,923 bigger problems. And the people who will be doing 656 00:32:35,923 --> 00:32:38,641 coding will be actually much broader than the, what, 28 657 00:32:38,641 --> 00:32:40,405 million developers we have on the planet right now. 658 00:32:40,495 --> 00:32:43,386 I mean it'll be anything adjacent to developers like, you 659 00:32:43,386 --> 00:32:46,555 know, Kubernetes and operations and UX and product and design 660 00:32:46,555 --> 00:32:49,699 and so forth. Yeah I just... 661 00:32:49,699 --> 00:32:54,086 Am I a hundred percent certain that, you know, it's a good 662 00:32:54,086 --> 00:32:57,551 times ahead? Uh, not entirely but I just I 663 00:32:57,551 --> 00:33:01,352 feel like the notion that somehow we're gonna shed, we're 664 00:33:01,352 --> 00:33:04,884 not gonna need developers anymore, I just find logically 665 00:33:04,884 --> 00:33:06,909 impossible. Does this say that some 666 00:33:06,909 --> 00:33:09,163 wrong-headed leaders are gonna say we're gonna cut 20% of 667 00:33:09,163 --> 00:33:10,759 developers because we don't need them anymore? 668 00:33:10,909 --> 00:33:13,806 I'm sure that's gonna come. Like, you know, that's a, you 669 00:33:13,806 --> 00:33:17,007 know, assholes like that, you know, have always been around 670 00:33:17,007 --> 00:33:19,065 and they're gonna suffer the consequences. 671 00:33:19,617 --> 00:33:22,473 But yeah I think the right-headed leaders will 672 00:33:22,473 --> 00:33:24,928 realize, you know, we need developers more than ever. 673 00:33:25,820 --> 00:33:27,240 Yeah. In fact, so many organizations 674 00:33:27,240 --> 00:33:29,768 are thinking, you know, by introducing AI, we will be able 675 00:33:29,768 --> 00:33:32,180 to kind of like cut a number of people that we have. 676 00:33:32,420 --> 00:33:34,520 And obviously the layoffs that are happening in some 677 00:33:34,520 --> 00:33:35,990 organizations probably due to that. 678 00:33:36,320 --> 00:33:39,920 And I think soon or maybe later they will realize that, oh, 679 00:33:39,920 --> 00:33:43,007 probably we cannot just cut developers because we still need 680 00:33:43,007 --> 00:33:44,798 developers. And like what you mentioned, the 681 00:33:44,798 --> 00:33:47,480 amount of software, if it gets written by a lot, right? 682 00:33:47,690 --> 00:33:50,760 Obviously sometimes when there are issues or you wanna change, 683 00:33:50,760 --> 00:33:53,540 evolve the software, you'll need more developers to help you. 684 00:33:53,810 --> 00:33:56,829 But this time probably a better developer that has a good skill, 685 00:33:56,829 --> 00:33:58,790 right? Rather than a developer that 686 00:33:58,790 --> 00:34:02,048 just do the mundane tasks. So I think that's a very 687 00:34:02,048 --> 00:34:03,380 important thing to realize as well. 688 00:34:04,129 --> 00:34:06,817 Yeah, in fact I was talking to Rodo at John Deere. 689 00:34:06,817 --> 00:34:10,072 Apparently he is the number one token consumer inside John Deere 690 00:34:10,072 --> 00:34:12,917 and I think the GitHub team told him that he might be in the 691 00:34:12,917 --> 00:34:16,568 single digits of top token users on the planet which is amazing. 692 00:34:16,828 --> 00:34:20,239 But yeah, he said, yeah developers were craftsmen, were 693 00:34:20,239 --> 00:34:23,860 craftspeople, right? That it is a tool and the tool 694 00:34:23,860 --> 00:34:25,984 is only as good as the person using it. 695 00:34:26,074 --> 00:34:30,643 And so AI is great for juniors. But like what seniors need to do 696 00:34:30,643 --> 00:34:33,232 is like every mistake that they're making, somehow we need 697 00:34:33,232 --> 00:34:35,417 to make it impossible for juniors to make the same 698 00:34:35,417 --> 00:34:38,728 mistake. Because like to really get great 699 00:34:38,728 --> 00:34:43,129 leverage out of it, like not everyone can be a senior, right? 700 00:34:43,389 --> 00:34:46,018 If everyone has to be a senior developer to get value out of 701 00:34:46,018 --> 00:34:47,592 it, then, you know, like we're not there yet. 702 00:34:49,268 --> 00:34:53,357 Yeah, so maybe in terms of the downsides, because we just talk 703 00:34:53,357 --> 00:34:55,853 about the benefits, right? So I'm sure you have seen in the 704 00:34:55,853 --> 00:34:58,368 news as well, you know, the downsides of vibe coding, you 705 00:34:58,368 --> 00:35:01,524 know, production database gets deleted, you know, things that 706 00:35:01,524 --> 00:35:03,382 should be private becomes public. 707 00:35:03,472 --> 00:35:05,062 And so many other things, issues. 708 00:35:05,482 --> 00:35:08,194 So tell us a little bit more some downsides, because I'm sure 709 00:35:08,194 --> 00:35:11,144 there's a risk of vibe coding too much, especially if you 710 00:35:11,144 --> 00:35:12,922 don't check the code that gets generated. 711 00:35:13,192 --> 00:35:16,238 So tell us some of the, these things so that we can kind of 712 00:35:16,238 --> 00:35:17,812 like manage our expectations better as well. 713 00:35:18,662 --> 00:35:21,057 Oh yeah, yeah. So there are definite downsides 714 00:35:21,057 --> 00:35:24,908 for using AI for coding. I mean it is well known that 715 00:35:24,908 --> 00:35:26,148 there's like significant limitations. 716 00:35:26,148 --> 00:35:29,220 They are, one of them is like, you know, context window 717 00:35:29,220 --> 00:35:30,554 limitations, right? It's like, you know. 718 00:35:30,554 --> 00:35:32,896 And unfortunately, and it is true that the more you put into 719 00:35:32,896 --> 00:35:35,731 the context window, not only if you get to a certain point it 720 00:35:35,731 --> 00:35:37,507 won't accept it but the dumber they get. 721 00:35:37,567 --> 00:35:39,700 There's actually this benchmark that shows, it's called Fiction 722 00:35:39,700 --> 00:35:42,793 Bench, that you give it a story and then you ask questions like 723 00:35:42,793 --> 00:35:45,033 what happened to the character, you know, between chapter two 724 00:35:45,033 --> 00:35:47,697 and chapter seven. And it turns out like, this 725 00:35:47,697 --> 00:35:52,080 paper was from a year and a half ago, even above 3,000 tokens, 726 00:35:52,080 --> 00:35:53,940 like performance degrades significantly. 727 00:35:54,055 --> 00:35:57,614 So it's just really interesting to know that if you can't even 728 00:35:57,614 --> 00:36:00,753 follow a character in a, you know, a moderate like novella, 729 00:36:00,753 --> 00:36:03,800 if you can't do that then you should have, you know, don't 730 00:36:03,800 --> 00:36:05,194 trust it. That's a limitation that's gonna 731 00:36:05,194 --> 00:36:06,814 hit you when you're trying to code as well. 732 00:36:07,684 --> 00:36:10,217 Another one is like AIs are actually bad at instruction 733 00:36:10,217 --> 00:36:12,280 following. There are so many times when you 734 00:36:12,280 --> 00:36:15,132 tell it to do something and it'll forget or it will, you 735 00:36:15,132 --> 00:36:17,801 know, it will change the meaning of the instructions. 736 00:36:17,801 --> 00:36:19,489 And then there's actually something called, like it's 737 00:36:19,489 --> 00:36:21,321 actually prone to hacking its reward function. 738 00:36:21,381 --> 00:36:24,199 In other words, aI are trained to be helpful, right, and 739 00:36:24,199 --> 00:36:26,475 completing tasks. And often like as a context 740 00:36:26,475 --> 00:36:29,490 window fills up, it will almost start to panic and just say, 741 00:36:29,490 --> 00:36:32,011 okay, I'm done. I've made all your tests pass 742 00:36:32,011 --> 00:36:34,090 but it doesn't tell you that it's actually disabled, the way 743 00:36:34,090 --> 00:36:36,211 it did it was disabling a whole bunch of tests. 744 00:36:36,803 --> 00:36:39,667 Like we highlight a whole bunch of disasters that happened to 745 00:36:39,667 --> 00:36:42,523 us. One is Steve had a 80% of his 746 00:36:42,523 --> 00:36:46,027 unit tests deleted one time and only found out like a week later 747 00:36:46,027 --> 00:36:49,613 when his colleagues told him. I've created haunted code bases 748 00:36:49,613 --> 00:36:52,283 that were impossible to understand, impossible to test 749 00:36:52,283 --> 00:36:53,833 and almost impossible to rewrite. 750 00:36:54,450 --> 00:36:58,217 Steve said, in fact, I've almost had a entire repos disappear, 751 00:36:58,217 --> 00:37:01,079 right? In fact, Steve tells the story 752 00:37:01,079 --> 00:37:05,216 about like the only surviving link to a repo of, you know, to 753 00:37:05,216 --> 00:37:08,207 contain weeks of work, thousand volumes of tokens of work, was 754 00:37:08,207 --> 00:37:11,267 in one window. And had he close that window the 755 00:37:11,267 --> 00:37:13,484 inode would've been disappeared in the garbage collected away 756 00:37:13,484 --> 00:37:15,059 and like it would've been all gone. 757 00:37:15,869 --> 00:37:17,770 I've had it corrupt production databases. 758 00:37:17,889 --> 00:37:20,974 You know, you might be asking why would you give it production 759 00:37:20,974 --> 00:37:22,623 access. Because like it's so helpful, 760 00:37:22,623 --> 00:37:24,931 right? It's like, oh my gosh. 761 00:37:24,931 --> 00:37:27,729 I once asked it to the Google Photos API. 762 00:37:28,701 --> 00:37:31,681 They deprecated something. In other words, you couldn't 763 00:37:31,681 --> 00:37:34,425 retrieve photos, screenshots that it didn't upload. 764 00:37:34,889 --> 00:37:36,559 And so they basically, it was useless to me. 765 00:37:36,559 --> 00:37:38,129 I was relying on that for over a decade. 766 00:37:38,499 --> 00:37:42,096 And so I said, hey go into my Apple Photos SQLite database and 767 00:37:42,096 --> 00:37:45,608 just run it around and see if you can, you know, gimme a way 768 00:37:45,608 --> 00:37:47,496 to download, you know, the full size image. 769 00:37:47,585 --> 00:37:50,270 And that was scary. Startling. 770 00:37:51,330 --> 00:37:54,950 Awe inspiring. Amazing, but scary, you know. 771 00:37:55,210 --> 00:37:59,042 But, you know, anyway just sometimes I have to decide to 772 00:37:59,042 --> 00:38:01,909 take a risk, right, because it's just something I really really 773 00:38:01,909 --> 00:38:04,753 want. The lesson we try to give in the 774 00:38:04,753 --> 00:38:08,128 book is that as an engineer you have to be aware of these 775 00:38:08,128 --> 00:38:10,433 limitations. And as an engineer, right, it's 776 00:38:10,433 --> 00:38:14,303 all about, what engineers do is we create reliable systems out 777 00:38:14,303 --> 00:38:15,743 of unreliable technologies, right? 778 00:38:15,743 --> 00:38:18,963 That's what cloud was, that's what Amazon S3 was, right? 779 00:38:19,574 --> 00:38:21,164 Kubernetes, right? You know, same thing. 780 00:38:21,644 --> 00:38:24,224 You know, they create the illusion, you know, that you 781 00:38:24,224 --> 00:38:27,064 have a, you know, long lived, you know, container image 782 00:38:27,064 --> 00:38:30,181 running in production, you know, that you don't have to worry 783 00:38:30,181 --> 00:38:33,309 about things failing. And so I think it's all about, 784 00:38:33,309 --> 00:38:37,133 in the book, we say it's all about taking all your same 785 00:38:37,133 --> 00:38:39,277 engineering instincts but then turning up to 11. 786 00:38:39,407 --> 00:38:44,008 There's this thing called the Nygard Stability Criterion that 787 00:38:44,008 --> 00:38:47,572 says the control mechanism must be faster than what it's 788 00:38:47,572 --> 00:38:49,374 controlling. And so when your code generation 789 00:38:49,374 --> 00:38:52,298 time goes up by a 100x, your control systems have to go up by 790 00:38:52,298 --> 00:38:54,843 at least a 100x. And so suddenly you're coming 791 00:38:54,843 --> 00:38:58,932 into version control more, you are relying on tests more, you 792 00:38:58,932 --> 00:39:02,812 have to have more modular systems because you can't let 793 00:39:02,812 --> 00:39:04,372 small problems become huge problems. 794 00:39:04,372 --> 00:39:05,942 You wanna be able to contain and isolate them. 795 00:39:06,683 --> 00:39:10,581 So all, you know, these are not new but as we put in the book 796 00:39:10,581 --> 00:39:14,088 like if the fastest you've ever gone is like 15 miles an hour on 797 00:39:14,088 --> 00:39:17,245 horseback and someone gives you a keys to a car and says go 15 798 00:39:17,245 --> 00:39:19,687 miles an hour or 500 miles an hour, you'll wreck the car. 799 00:39:19,717 --> 00:39:22,273 And we write in explicit excruciating details how many 800 00:39:22,273 --> 00:39:25,399 times have you wrecked the car, like how our instincts failed us 801 00:39:25,399 --> 00:39:28,207 and like the new instincts that, you know, we need to replace it 802 00:39:28,207 --> 00:39:29,910 with. So yes, there's definite 803 00:39:29,910 --> 00:39:33,269 downsides but we believe that the same engineering principles 804 00:39:33,269 --> 00:39:35,982 that, you know, we've learned over the last couple of decades, 805 00:39:35,982 --> 00:39:39,768 we just need to speed them up even more, to create faster 806 00:39:39,768 --> 00:39:41,852 feedback earlier, more frequently, and stronger because 807 00:39:41,852 --> 00:39:45,168 you need to know when things are about to go off the rails. 808 00:39:45,378 --> 00:39:47,879 Does that resonate with you? Yeah, definitely. 809 00:39:47,879 --> 00:39:50,459 I mean like especially if you don't check the output, right? 810 00:39:50,489 --> 00:39:54,353 And if you give it a big prompt such that, you know, it can 811 00:39:54,353 --> 00:39:57,315 generate tons of code, obviously there's a lot of dangers in 812 00:39:57,315 --> 00:39:58,919 doing that. And especially just accepting 813 00:39:58,919 --> 00:40:01,265 and move on. And people these days also have 814 00:40:01,265 --> 00:40:04,184 this, you know, auto accept. They didn't, they don't even 815 00:40:04,184 --> 00:40:06,899 check, you know, the iteration that the AI agents are doing. 816 00:40:07,049 --> 00:40:08,609 So definitely there are a lot of dangers. 817 00:40:08,849 --> 00:40:12,099 And I love when I read about this hijacking reward function 818 00:40:12,099 --> 00:40:15,547 that you mentioned in the book, because there are so many things 819 00:40:15,547 --> 00:40:18,612 that you need to understand kind of like the, I don't know, like 820 00:40:18,612 --> 00:40:21,075 the behavior of AI, the philosophy of AI, right? 821 00:40:21,075 --> 00:40:24,135 You mentioned that AI tries to be helpful all the time, even 822 00:40:24,135 --> 00:40:27,037 though it can cheat. It can achieve that by cheating, 823 00:40:27,037 --> 00:40:28,705 right? So I think that's very 824 00:40:28,705 --> 00:40:32,211 fascinating when I read that. So you also mentioned that now 825 00:40:32,211 --> 00:40:35,162 we have to put a lot more guardrails, developers job 826 00:40:35,162 --> 00:40:37,052 probably is changing, shifting, right? 827 00:40:37,052 --> 00:40:40,546 It's not just the writing of the code, but now becomes like a 828 00:40:40,546 --> 00:40:42,362 head chef you mentioned in your book, right? 829 00:40:42,592 --> 00:40:45,071 So tell us some of these transition things that all 830 00:40:45,071 --> 00:40:48,108 developers must be aware of, because I'm sure the role itself 831 00:40:48,108 --> 00:40:51,652 is changing a lot with AI. And what are the important skill 832 00:40:51,652 --> 00:40:54,980 sets that we need to invest in order to become a much better 833 00:40:54,980 --> 00:40:59,158 developer equipped with AI? Yeah, the metaphor we use in the 834 00:40:59,158 --> 00:41:02,533 book is you go from line cook to head chef, right? 835 00:41:02,533 --> 00:41:05,481 Your job is not to cut every vegetable, sear every steak, 836 00:41:05,481 --> 00:41:08,920 right, your job is to help create the systems, to allow all 837 00:41:08,920 --> 00:41:11,715 these AI agents, you know, these line cooks to work for you. 838 00:41:12,225 --> 00:41:15,141 And I guess what I love about the metaphor is that we're 839 00:41:15,141 --> 00:41:17,995 saying that, you know, as it's your kitchen, your Michelin 840 00:41:17,995 --> 00:41:21,122 stars, right, you are ultimately responsible for whatever comes 841 00:41:21,122 --> 00:41:22,895 outta the kitchen. So if someone gets food 842 00:41:22,895 --> 00:41:25,445 poisoning, right, if you get a health code violation, right, 843 00:41:25,445 --> 00:41:28,115 it's not the AI's fault. It's your fault. 844 00:41:28,933 --> 00:41:31,651 And I just love that because it extends to whether it's a small 845 00:41:31,651 --> 00:41:34,593 kitchen with a, you know, one help, one or two helpers to, you 846 00:41:34,593 --> 00:41:36,178 know, a large kitchen with 20 helpers. 847 00:41:36,178 --> 00:41:38,158 Or maybe it's a chain of restaurants, you know, with a, 848 00:41:38,158 --> 00:41:41,346 you know, hundreds of people, you know, who are all part of 849 00:41:41,346 --> 00:41:43,382 this endeavor. You know, it's interesting. 850 00:41:43,382 --> 00:41:45,506 At the conference, we had Stuart Pearce. 851 00:41:45,526 --> 00:41:48,881 He's a group CTO of Hg. So they're a private equity 852 00:41:48,881 --> 00:41:50,994 firm. He was saying that, you know, 853 00:41:50,994 --> 00:41:53,637 he's met so many senior leaders, engineering, senior engineers 854 00:41:53,637 --> 00:41:57,396 who he could just sort of tell if they're gonna be great vibe 855 00:41:57,396 --> 00:42:00,517 coders or not. And he said it was like two 856 00:42:00,517 --> 00:42:02,828 things. One was like are they good at 857 00:42:02,828 --> 00:42:04,818 specification? Are they good at delegation? 858 00:42:05,338 --> 00:42:08,307 And are they willing to jump in to finish the last 5% if the AI 859 00:42:08,307 --> 00:42:10,528 can't get right? And I was like, yeah, that makes 860 00:42:10,528 --> 00:42:12,328 a lot of, just I found that familiar. 861 00:42:12,768 --> 00:42:14,818 And I was talking to another friend of mine Jason Cox that 862 00:42:14,818 --> 00:42:18,502 he's a executive director at Disney. 863 00:42:18,502 --> 00:42:20,602 And I told that to him and his eyes lit up. 864 00:42:20,602 --> 00:42:23,620 He said, yeah, you know what? A whole bunch of engineers who 865 00:42:23,620 --> 00:42:25,132 were very resistant to vibe coding. 866 00:42:25,462 --> 00:42:28,472 And so he sort of he pulled their, the performance reviews 867 00:42:28,472 --> 00:42:32,251 and he said the single theme that kept on coming up over and 868 00:42:32,251 --> 00:42:35,212 over again was we want to give you more responsibility but you 869 00:42:35,212 --> 00:42:37,837 need to delegate more. And he was just like... 870 00:42:37,837 --> 00:42:39,678 And so delegation is a teachable skill. 871 00:42:39,678 --> 00:42:43,804 And so I think, again, I just I love these anecdotes that I 872 00:42:43,804 --> 00:42:47,081 really do believe are, you know, there's a basis for it. 873 00:42:47,081 --> 00:42:49,391 And I think kind of through benchmarking, I think we're 874 00:42:49,391 --> 00:42:52,246 gonna find these kind of predictors of whether people are 875 00:42:52,246 --> 00:42:54,822 gonna be good at it or not, right? 876 00:42:54,822 --> 00:42:57,720 And what are the skills that need to be taught to take full 877 00:42:57,720 --> 00:42:59,412 advantage of this amazing new technology? 878 00:43:00,277 --> 00:43:02,287 Yeah. So, head chef. 879 00:43:02,962 --> 00:43:06,156 I think what it means is that sensibilities that I think 880 00:43:06,156 --> 00:43:10,256 become more important than ever is are you creating a system 881 00:43:10,256 --> 00:43:13,521 that has very effective feedback loops, where when something goes 882 00:43:13,521 --> 00:43:15,496 wrong you get fast feedback right away. 883 00:43:15,676 --> 00:43:18,769 It's a strong signal, you can stop the line, and you can fix 884 00:43:18,769 --> 00:43:20,757 it. Or are you creating a system 885 00:43:20,757 --> 00:43:22,832 that has very weak signals and you're not paying attention 886 00:43:22,832 --> 00:43:25,307 anyway and, you know, you're just letting accept all you 887 00:43:25,307 --> 00:43:27,850 know, dangerously, you know, allow anything, you know, you 888 00:43:27,850 --> 00:43:31,779 are gonna have a bad time. I think another sensibility is 889 00:43:31,779 --> 00:43:32,952 like architectural sensibilities. 890 00:43:32,952 --> 00:43:36,020 I mean are you creating a system that agents can work on in 891 00:43:36,020 --> 00:43:37,972 parallel without interfering with each other, right? 892 00:43:38,602 --> 00:43:41,619 Or they all snarl together and every time you try to commit, 893 00:43:41,619 --> 00:43:44,012 you end up with these horrendous merge conflicts which, you know, 894 00:43:44,012 --> 00:43:46,378 has happened to me more often than I'd like to admit. 895 00:43:46,812 --> 00:43:48,801 And, you know, what's interesting is that now you're 896 00:43:48,801 --> 00:43:50,681 spending all your time not doing cool things. 897 00:43:50,681 --> 00:43:53,040 All you're doing is like un-snarling, un-coordinating, 898 00:43:53,040 --> 00:43:55,651 you know, coordinating, re-coordinating, and that's not 899 00:43:55,651 --> 00:43:57,745 good. And so just those same 900 00:43:57,745 --> 00:44:00,488 sensibilities that work for people, you know, also apply to 901 00:44:00,488 --> 00:44:03,286 AI. Then I think another thing is 902 00:44:03,286 --> 00:44:06,456 like you have to love learning. I think some people get 903 00:44:06,456 --> 00:44:09,469 energized by, you know, the rate of change in what's happening 904 00:44:09,469 --> 00:44:11,563 with coding agents right now and so forth. 905 00:44:11,983 --> 00:44:14,503 Other people get frightened and sort of dig the heels in. 906 00:44:14,563 --> 00:44:18,042 And I think it's fun to be part of the first group. 907 00:44:18,402 --> 00:44:20,652 I think we have to have sympathy for the second group, because 908 00:44:20,712 --> 00:44:24,162 eventually, right, we want them to be a part of the ride, right? 909 00:44:24,162 --> 00:44:25,646 As Steve said, this is the big one, right? 910 00:44:25,803 --> 00:44:27,945 You know, like imagine you're in Hawaii and the big wave is 911 00:44:27,945 --> 00:44:29,643 coming. You gotta get ready, get the 912 00:44:29,643 --> 00:44:32,661 surfboards out, practice so you can not just ride the big wave 913 00:44:32,661 --> 00:44:36,191 but survive the big wave, right? And then the last one that we 914 00:44:36,191 --> 00:44:39,337 say that you need to do is you have to love to cook, you know. 915 00:44:39,427 --> 00:44:41,257 You love, have to love creating things, right? 916 00:44:41,257 --> 00:44:43,889 If, you know, coding is just a job and you're just sort of 917 00:44:43,889 --> 00:44:47,050 punching a ticket, maybe vibe coding is not gonna be the most 918 00:44:47,050 --> 00:44:49,086 fun for you. Yeah. 919 00:44:49,086 --> 00:44:51,946 So I think those are very important skills indeed, right? 920 00:44:51,966 --> 00:44:55,005 I'll just try to summarize, like faster feedback loop, right? 921 00:44:55,755 --> 00:44:58,507 And I think in fact it becomes much, much more important simply 922 00:44:58,507 --> 00:45:00,645 for the guardrails reason that you mentioned. 923 00:45:00,945 --> 00:45:03,489 And the modularity, the architecture thing, definitely 924 00:45:03,489 --> 00:45:06,846 makes sense, because you still wanna come up with the code that 925 00:45:06,846 --> 00:45:09,084 is designed well. Because the code will evolve 926 00:45:09,084 --> 00:45:11,497 unless you're building like one time project where, you know, 927 00:45:11,497 --> 00:45:13,738 once you use it, that's it, that's done, right? 928 00:45:13,888 --> 00:45:17,242 And then, embrace learning. Definitely these days it can be 929 00:45:17,242 --> 00:45:19,558 very overwhelming. Sometimes I feel overwhelmed as 930 00:45:19,558 --> 00:45:21,778 well, like so many rapid advancements happening. 931 00:45:22,169 --> 00:45:24,962 And sometimes you need to be patient as well about learning 932 00:45:24,962 --> 00:45:27,436 those new things. Lastly is mastering the craft, 933 00:45:27,436 --> 00:45:30,225 right? Like you need to learn like how 934 00:45:30,225 --> 00:45:33,424 to enjoy your role, right? Like become a problem solver, 935 00:45:33,424 --> 00:45:35,035 writing software and things like that. 936 00:45:35,605 --> 00:45:38,461 So I wanna dive a little bit on the faster feedback loop, 937 00:45:38,461 --> 00:45:41,501 because I feel feedback loop is such an important thing. 938 00:45:41,501 --> 00:45:45,005 In the past, when in the DevOps world, especially like with 939 00:45:45,005 --> 00:45:47,429 CI/CD, automated tests, and all that, feedback loop is very 940 00:45:47,429 --> 00:45:50,312 important to give you like signals where you are heading 941 00:45:50,312 --> 00:45:53,258 to, either are you heading to the right direction or wrong 942 00:45:53,258 --> 00:45:55,052 direction? Now with AI, you mentioned that 943 00:45:55,052 --> 00:45:56,732 you are given like a very fast car. 944 00:45:57,041 --> 00:45:58,738 Things are churned out very, very fast. 945 00:45:58,978 --> 00:46:01,519 Without fast feedback loop, I think it's much more dangerous. 946 00:46:01,519 --> 00:46:04,459 So tell us about the importance of this feedback loop. 947 00:46:05,059 --> 00:46:08,370 And you have this new cycle that gets introduced, like prevent, 948 00:46:08,370 --> 00:46:10,110 correct, and detect those kind of things. 949 00:46:10,290 --> 00:46:12,727 So maybe tell us a little bit more about these techniques such 950 00:46:12,727 --> 00:46:14,550 that developer can benefit from that. 951 00:46:15,950 --> 00:46:18,045 Yeah. Oh, you know, so I, you know, I 952 00:46:18,045 --> 00:46:20,901 feel like all my entire career I gotten me ready to write this 953 00:46:20,901 --> 00:46:23,296 book. And so I got to work on a book 954 00:46:23,296 --> 00:46:25,468 called Wiring the Winning Organization with Dr. Steven 955 00:46:25,468 --> 00:46:27,949 Spear. And so he is famous for many 956 00:46:27,949 --> 00:46:31,538 things but among the thing he's famous for is writing the most 957 00:46:31,538 --> 00:46:34,078 widely downloaded Harvard Business Review article of all 958 00:46:34,078 --> 00:46:37,350 time called Decoding the DNA of the Toyota Production System. 959 00:46:37,620 --> 00:46:40,870 And I remember reading it in 2001 and thinking like this is 960 00:46:40,870 --> 00:46:44,097 amazing that, you know, he talked about how in any Toyota 961 00:46:44,097 --> 00:46:46,927 manufacturing plant there are thousands of people working in 962 00:46:46,927 --> 00:46:49,825 the community of scientists doing experiments every day and 963 00:46:49,825 --> 00:46:52,229 continuously learning. And that's enabled by feedback. 964 00:46:52,249 --> 00:46:56,395 And so feedback whether you're making cars or launching a 965 00:46:56,395 --> 00:46:59,849 rocket or performing a surgery, you have to see what you're 966 00:46:59,849 --> 00:47:03,473 doing, you know. And it's not like a one 967 00:47:03,473 --> 00:47:05,864 screenshot, you know, per operation, right? 968 00:47:05,864 --> 00:47:08,029 It's like you wanna see what you're doing at any given point 969 00:47:08,029 --> 00:47:10,371 in time. And so it turns out like 970 00:47:10,371 --> 00:47:13,025 feedback is like one of the most critical things in any 971 00:47:13,025 --> 00:47:16,509 management system but also like in any engineering endeavor. 972 00:47:17,291 --> 00:47:19,919 You know, to go into that Nygard, there's something called 973 00:47:19,919 --> 00:47:23,391 the Nygard the Shannon Nyquist Theorem. 974 00:47:23,421 --> 00:47:27,939 This, the provenance of this is not quite clear to me but 975 00:47:27,939 --> 00:47:30,191 there's somewhere there's something that says the 976 00:47:30,191 --> 00:47:33,341 controller must operate at at least twice the rate of what 977 00:47:33,341 --> 00:47:35,986 it's controlling. And I think the Nyquist Shannon 978 00:47:35,986 --> 00:47:38,553 Theorem actually says, you know, for measuring something, you 979 00:47:38,553 --> 00:47:40,891 have to sample at twice the frequency, right? 980 00:47:40,921 --> 00:47:44,496 Which I think is like, you know, you can correlate you can then, 981 00:47:44,496 --> 00:47:47,378 the corollary is to control it, you have to be able to, you 982 00:47:47,378 --> 00:47:50,640 know, measure it, et cetera. And so like what we're trying 983 00:47:50,640 --> 00:47:54,124 to... what we it did in the book is said, alright, how do we have 984 00:47:54,124 --> 00:47:56,886 to modify our inner, middle, and outer development loops, you 985 00:47:56,886 --> 00:47:59,016 know, to use this amazing new technology. 986 00:47:59,016 --> 00:48:02,284 And, you know, we had some reviewers say, oh, you're 987 00:48:02,284 --> 00:48:05,109 actually breaking how the use of a term, right? 988 00:48:05,109 --> 00:48:08,320 Inner is kinda like what you do in your laptop, outer is what 989 00:48:08,320 --> 00:48:10,786 you do in CI/CD. It's like, yeah, you know what? 990 00:48:10,786 --> 00:48:13,319 It's like, it is it is, yeah, we'll go there. 991 00:48:13,389 --> 00:48:15,679 I mean it's just you have to change the way you work. 992 00:48:15,769 --> 00:48:18,024 And the way we define it was inner is like what do you have 993 00:48:18,024 --> 00:48:20,678 to do kinda like every second, every minute to prevent, detect, 994 00:48:20,678 --> 00:48:22,526 and correct bad things from happening. 995 00:48:23,098 --> 00:48:26,483 And then middle is, you know, what do you have to do every 996 00:48:26,483 --> 00:48:28,803 hour or days. And outer loop is maybe days or 997 00:48:28,803 --> 00:48:30,342 weeks. There's certain practices, you 998 00:48:30,342 --> 00:48:33,299 know, are like, alright, you gotta check your stuff into more 999 00:48:33,299 --> 00:48:35,084 versions more frequently. Whatever you're doing, you're 1000 00:48:35,084 --> 00:48:38,349 probably gonna be at least two to five to 10x more frequently. 1001 00:48:39,485 --> 00:48:41,255 It's funny. Steve Yegge, he allows his 1002 00:48:41,255 --> 00:48:43,955 coding agents to check in code all the time. 1003 00:48:44,015 --> 00:48:47,487 And I just thought it was like for me, I just, I let it do it 1004 00:48:47,487 --> 00:48:50,571 sometimes but I really wanna be the arbiter and decider about 1005 00:48:50,571 --> 00:48:53,974 what gets in and what isn't. There's sometimes emergencies 1006 00:48:53,974 --> 00:48:58,476 where I do let it do it but, for me, version control is my 1007 00:48:58,476 --> 00:49:01,919 ultimate fallback and I just, I don't want to, I just need to 1008 00:49:01,919 --> 00:49:05,076 keep that clear in my head. Another one is, yeah, like you 1009 00:49:05,076 --> 00:49:09,036 have to be aware of tests. Like you, the idea that you have 1010 00:49:09,036 --> 00:49:11,531 to have situational awareness. You have to know what branch 1011 00:49:11,531 --> 00:49:13,686 you're in, what window you're in, what project you're in. 1012 00:49:14,150 --> 00:49:17,710 You know, those are critical signals you need because how 1013 00:49:17,710 --> 00:49:21,562 many times - having a recall moment here - where you type in 1014 00:49:21,562 --> 00:49:23,630 the wrong prompt in the wrong window, right? 1015 00:49:23,630 --> 00:49:27,460 And so anyone who's done like server admin knows what it feels 1016 00:49:27,460 --> 00:49:31,975 like to drop a table when you thought you were in staging but 1017 00:49:31,975 --> 00:49:33,435 you're actually in production, right? 1018 00:49:33,435 --> 00:49:37,375 I mean I have similar moments inside of Claude Code. 1019 00:49:37,705 --> 00:49:39,905 Yeah, so there's certain things you gotta do like always, front 1020 00:49:39,905 --> 00:49:42,428 and center, like literally every second, and there are things 1021 00:49:42,428 --> 00:49:45,330 that you need to be doing kind of less frequently but, you 1022 00:49:45,330 --> 00:49:48,067 know, just as important. And where did that term come 1023 00:49:48,067 --> 00:49:49,653 from? It actually came from the audit 1024 00:49:49,653 --> 00:49:50,956 community. Like what are all the things, 1025 00:49:50,956 --> 00:49:53,066 bad things can happen? And we know from a personal 1026 00:49:53,066 --> 00:49:54,898 experience, you know, what are those bad things? 1027 00:49:54,928 --> 00:49:57,840 So now you need controls to either prevent that bad thing 1028 00:49:57,840 --> 00:50:00,862 from happening, and if it does happen you need to have 1029 00:50:00,862 --> 00:50:04,978 detection correction mechanisms. And so we have a whole section 1030 00:50:04,978 --> 00:50:07,902 on like, you know, the catalog of practices that we built up, 1031 00:50:07,902 --> 00:50:11,167 you know, all the disasters that we just talked about, you know, 1032 00:50:11,167 --> 00:50:13,051 how do you prevent, detect, and correct? 1033 00:50:14,147 --> 00:50:16,426 Yeah. So I think one key insights here 1034 00:50:16,426 --> 00:50:18,532 is like your controls must be in place, right? 1035 00:50:18,532 --> 00:50:21,232 And in fact, it must be greater than in the past. 1036 00:50:21,262 --> 00:50:25,250 Because now when code is very cheap to produce, you need more 1037 00:50:25,250 --> 00:50:28,869 controls to kind of like detect, right, if there's anything going 1038 00:50:28,869 --> 00:50:30,667 wrong, missing, or whatever that is, right? 1039 00:50:30,757 --> 00:50:33,748 And how to prevent that and how you correct that after that, 1040 00:50:33,748 --> 00:50:35,165 right? Sometimes the developers need to 1041 00:50:35,165 --> 00:50:37,237 be hands-on in order to be able to correct that. 1042 00:50:37,616 --> 00:50:38,506 Oh very. Yeah, yeah. 1043 00:50:38,527 --> 00:50:41,287 Yeah. So the last part of this 1044 00:50:41,287 --> 00:50:43,711 conversation, I wanna talk a little bit more because vibe 1045 00:50:43,711 --> 00:50:46,867 coding, a lot of discussions these days is more, you know, on 1046 00:50:46,867 --> 00:50:50,257 developers, individuals, right? Like everyone now can, you know, 1047 00:50:50,257 --> 00:50:53,167 vibe code. But I think the big potential is 1048 00:50:53,167 --> 00:50:56,737 actually to make vibe coding more kind of like a cultural 1049 00:50:56,737 --> 00:50:59,662 thing in an organization. And in your book, in the last 1050 00:50:59,662 --> 00:51:02,209 few chapters, you also mentioned like how can organization adopt 1051 00:51:02,209 --> 00:51:04,555 vibe coding as part of their culture? 1052 00:51:04,975 --> 00:51:07,927 So I know this probably a little bit more advanced for some 1053 00:51:07,927 --> 00:51:10,722 organizations who are still trying to adopt AI, like the 1054 00:51:10,722 --> 00:51:13,308 chat AI or some kind of other AI tools. 1055 00:51:13,578 --> 00:51:16,625 But vibe coding itself can be a powerful thing for organization 1056 00:51:16,625 --> 00:51:19,519 to actually innovate a lot more and produce something, you know, 1057 00:51:19,519 --> 00:51:22,534 that is more innovative. So tell us how can we start 1058 00:51:22,534 --> 00:51:23,769 creating this vibe coding culture? 1059 00:51:23,769 --> 00:51:25,149 What are the important things to set? 1060 00:51:26,269 --> 00:51:29,031 I don't know. But I mean, I think the way we 1061 00:51:29,031 --> 00:51:31,835 learn is by people sharing experience reports, which is the 1062 00:51:31,835 --> 00:51:34,599 reason why I've been running this conference for last 12 1063 00:51:34,599 --> 00:51:36,651 years. Yeah, and it was, I'll just 1064 00:51:36,651 --> 00:51:38,595 share some of the highlights that kind of really blew me 1065 00:51:38,595 --> 00:51:40,902 away. One was a guy named Sree 1066 00:51:40,902 --> 00:51:42,840 Balakrishnan. He's a head of technology and 1067 00:51:42,840 --> 00:51:46,350 product at Travelopia. So it's a $1.5 billion a year 1068 00:51:46,350 --> 00:51:49,054 travel company. And yeah, he was talking about 1069 00:51:49,054 --> 00:51:51,846 how they were able to rewrite a Drupal application in six weeks, 1070 00:51:51,846 --> 00:51:54,772 right, with a very small team of people using AI tools. 1071 00:51:55,264 --> 00:51:58,580 They only had the code, no documentation, but they did have 1072 00:51:58,580 --> 00:52:01,352 the users and they were able to reconstruct all the screens 1073 00:52:01,352 --> 00:52:03,606 needed to, you know, basically do everything outside of the 1074 00:52:03,606 --> 00:52:04,924 mainframe. Amazing! 1075 00:52:04,924 --> 00:52:07,116 You know, they rewrote the payment platform and there it 1076 00:52:07,116 --> 00:52:10,998 was like a race between kind of a typical dev team and a dev+AI 1077 00:52:10,998 --> 00:52:13,581 team. And, you know, the way he 1078 00:52:13,581 --> 00:52:18,400 described it was the dev team no AI was kind of this linear, one 1079 00:52:18,400 --> 00:52:21,776 dev+AI was more flat. But then it kind of hockey 1080 00:52:21,776 --> 00:52:25,984 sticked at the six week mark. They matched the progress of the 1081 00:52:25,984 --> 00:52:28,853 manual dev team. But he said the most important 1082 00:52:28,853 --> 00:52:31,519 thing is that the only team that was willing to go into 1083 00:52:31,519 --> 00:52:34,960 production was the dev+AI team. And then they basically shut 1084 00:52:34,960 --> 00:52:38,874 down the race after that. A company called Exabeam, they, 1085 00:52:38,874 --> 00:52:42,389 the head of product, he described how among many other 1086 00:52:42,389 --> 00:52:44,846 things, you know, they're a private equity owned firm. 1087 00:52:44,996 --> 00:52:47,499 And they said, you know, the classic playbook is that you 1088 00:52:47,499 --> 00:52:50,476 find an aging competitor or competitor adjacent and then you 1089 00:52:50,476 --> 00:52:52,136 acquire them for a hundred million dollars. 1090 00:52:52,196 --> 00:52:54,502 And then, you know, you grow the company that way. 1091 00:52:54,942 --> 00:52:56,752 And he said we did an experiment. 1092 00:52:56,752 --> 00:53:00,818 We got eight people together, we decided to instead of buying the 1093 00:53:00,818 --> 00:53:03,874 company which, you know, has balance sheet implications, 1094 00:53:03,874 --> 00:53:07,538 we're just gonna find an aging competitor and see if we can 1095 00:53:07,538 --> 00:53:09,646 create an MVP of exactly what they do. 1096 00:53:10,066 --> 00:53:13,237 And so the verdict isn't out yet but they said we learned so 1097 00:53:13,237 --> 00:53:14,927 much, we got so much further than we thought. 1098 00:53:15,377 --> 00:53:18,986 And the notion that product owners, designers can contribute 1099 00:53:18,986 --> 00:53:22,015 to a coding project was just so much higher than they ever 1100 00:53:22,015 --> 00:53:24,587 thought. The big challenge was how does 1101 00:53:24,587 --> 00:53:27,417 this product manager get someone to review their code when they 1102 00:53:27,417 --> 00:53:29,591 don't have those relationships, when you're not in the dev team, 1103 00:53:29,591 --> 00:53:33,286 you know, which is illuminating. And one thing that Shree talked 1104 00:53:33,286 --> 00:53:36,397 about was like he has a real sense that the team size is like 1105 00:53:36,397 --> 00:53:38,021 gonna change. It's not gonna be eight people 1106 00:53:38,021 --> 00:53:39,394 anymore. You don't need six developers 1107 00:53:39,394 --> 00:53:42,070 plus UX plus product. You just need a person with a 1108 00:53:42,070 --> 00:53:43,789 problem and a developer who can fix it, right? 1109 00:53:43,789 --> 00:53:47,500 And maybe a pair of those too can actually go further and 1110 00:53:47,500 --> 00:53:49,039 faster. That's super exciting. 1111 00:53:49,189 --> 00:53:53,104 But like one of my favorite presentations came from Dr. Topo 1112 00:53:53,104 --> 00:53:54,889 Pal, who I've known for over a decade. 1113 00:53:54,889 --> 00:53:56,755 He's a VP of Architecture at Fidelity. 1114 00:53:57,775 --> 00:54:00,466 And among one of the applications he's responsible 1115 00:54:00,466 --> 00:54:04,178 for is this application where you ask, you know, which of our 1116 00:54:04,178 --> 00:54:06,736 20,000 plus applications have Log4j in it, who's the 1117 00:54:06,736 --> 00:54:08,856 responsible manager, which business unit, and so forth. 1118 00:54:09,436 --> 00:54:12,751 And he's had this vision of what this application should be in 1119 00:54:12,751 --> 00:54:15,084 his head. But when he ever asked his own 1120 00:54:15,084 --> 00:54:17,413 dev team like what would it take to build it? 1121 00:54:17,413 --> 00:54:21,565 They would say eight months, and plus we need a front-end React 1122 00:54:21,565 --> 00:54:25,225 person, right? And he got so frustrated that he 1123 00:54:25,225 --> 00:54:28,735 decided to write it himself. He spent a whole weekend vibe 1124 00:54:28,735 --> 00:54:32,099 coding. 10 days later, right, he had something that followed all 1125 00:54:32,099 --> 00:54:34,538 of the Fidelity coding standards of which, you know, he's 1126 00:54:34,538 --> 00:54:36,869 partially responsible for. And it went into production last 1127 00:54:36,869 --> 00:54:38,585 month. The question is, all right, 1128 00:54:38,585 --> 00:54:40,587 who's gonna maintain it? You know, who's gonna help Topo 1129 00:54:40,587 --> 00:54:41,795 build the next set of functionality? 1130 00:54:41,945 --> 00:54:44,555 And her name is Swathi, the most junior engineer. 1131 00:54:44,555 --> 00:54:46,865 Because all the senior engineers are like no, no thank you. 1132 00:54:47,855 --> 00:54:50,417 And the surprising thing is he's getting more headcount because 1133 00:54:50,417 --> 00:54:53,948 the number of people using this tool is so much broader than 1134 00:54:53,948 --> 00:54:57,955 what it could do than before. And so like who saw that coming? 1135 00:54:58,755 --> 00:55:02,952 So... oh and he's saying the teams who are vibe coding, some 1136 00:55:02,952 --> 00:55:07,551 of them are saying for the, you know, for 20 plus years or what 1137 00:55:07,551 --> 00:55:10,491 however that long they've been using kind of Scrum-like 1138 00:55:10,491 --> 00:55:14,911 patterns, we've always been able to demo everything we need to 1139 00:55:14,911 --> 00:55:18,427 demo every two weeks. He's now consistently, there are 1140 00:55:18,427 --> 00:55:21,657 some teams who don't have time to demo everything they built 1141 00:55:21,657 --> 00:55:24,825 and now they're asking do we need to have these demos more 1142 00:55:24,825 --> 00:55:27,697 frequently? It just, it says, it just opens 1143 00:55:27,697 --> 00:55:30,991 up so many questions and I think this all very exciting. 1144 00:55:30,991 --> 00:55:33,121 I mean, it's not boring. So, yeah. 1145 00:55:33,215 --> 00:55:34,695 And then plus John Rouser, right? 1146 00:55:34,915 --> 00:55:39,244 He's presenting tomorrow about what happened when a hundred 1147 00:55:39,244 --> 00:55:42,395 leaders within Cisco Security, all have to vibe code. 1148 00:55:42,555 --> 00:55:45,099 And so, you know, the survey instrument that, you know, we 1149 00:55:45,099 --> 00:55:48,240 were working on is really designed to say, all right, who 1150 00:55:48,240 --> 00:55:52,056 is the most excited, right? Who is like we called it super 1151 00:55:52,056 --> 00:55:54,920 forward or something else. Like the early adopters 1152 00:55:54,920 --> 00:55:57,647 enthusiasts, the partial completers did not complete. 1153 00:55:58,452 --> 00:56:02,292 What are the critical enablers that allowed the high performers 1154 00:56:02,292 --> 00:56:04,422 to be high performers and what are they gonna do about it? 1155 00:56:04,482 --> 00:56:08,920 Like given all these a-ha moments, how do they wanna 1156 00:56:08,920 --> 00:56:12,792 change the way, you know, their engineering teams work. 1157 00:56:13,812 --> 00:56:18,119 I can't wait to learn. Yeah, and I think as you 1158 00:56:18,119 --> 00:56:22,196 mentioned, like the juniors now is being able to kind of like be 1159 00:56:22,196 --> 00:56:25,106 part of an active problem solver maybe in the organization. 1160 00:56:25,196 --> 00:56:28,316 And this I think also comes back to how you hire, right? 1161 00:56:28,496 --> 00:56:31,436 What are important traits these days that you think we should 1162 00:56:31,436 --> 00:56:34,574 look when hiring developers? Because now that everyone 1163 00:56:34,574 --> 00:56:38,126 potentially can write the same, you know, amount of code, maybe 1164 00:56:38,126 --> 00:56:41,006 the same quality, because AI is kind of like the leveler here. 1165 00:56:41,336 --> 00:56:44,006 So what are the important skills to look at when you hire? 1166 00:56:44,946 --> 00:56:48,625 Yeah, I don't know. But I mean I was at the Grace 1167 00:56:48,625 --> 00:56:51,155 Hopper Celebration last week so that was 14,000 women in 1168 00:56:51,155 --> 00:56:52,695 technology. And it was, I've never been to 1169 00:56:52,695 --> 00:56:55,990 anything like it in my life. It was people across all parts 1170 00:56:55,990 --> 00:57:00,304 of the, you know, career stages. You know, college students, 1171 00:57:00,304 --> 00:57:03,917 researchers, senior leaders. And like I don't think anyone 1172 00:57:03,917 --> 00:57:07,463 knows but I'll tell you what really caught my attention. 1173 00:57:07,463 --> 00:57:11,363 A couple students, they asked is what I'm learning here in 1174 00:57:11,363 --> 00:57:13,483 university still important? And I'm like I don't know. 1175 00:57:14,113 --> 00:57:17,559 But I do feel like and we put in the book some positions and this 1176 00:57:17,559 --> 00:57:19,800 mostly came from Steve, but I actually agree wholeheartedly. 1177 00:57:19,800 --> 00:57:22,680 It's like do we really need to be teaching data structures this 1178 00:57:22,680 --> 00:57:24,250 much? I mean is it really a semester's 1179 00:57:24,250 --> 00:57:26,340 worth of work or is it like a couple weeks, right? 1180 00:57:26,340 --> 00:57:29,437 I mean, I don't think very many people these days are gonna be 1181 00:57:29,437 --> 00:57:31,533 writing certainly not linked list or complex b-trees, right? 1182 00:57:31,533 --> 00:57:34,425 It's like that's, I don't think that's as relevant but I think 1183 00:57:34,425 --> 00:57:37,308 we need to be teaching more, right, things like software 1184 00:57:37,308 --> 00:57:40,444 architecture and how to create dynamic feedback loops and, you 1185 00:57:40,444 --> 00:57:43,782 know, tools like how do we use tools like vibe coding. 1186 00:57:44,142 --> 00:57:47,658 Yeah, it's super interesting and I just, you know, one of the 1187 00:57:47,658 --> 00:57:51,633 things that we put in the book was if we expect developers to 1188 00:57:51,633 --> 00:57:55,202 be using AI to solve problems, well then shouldn't we want 1189 00:57:55,202 --> 00:57:59,030 developers to be using AI as part of the interview process? 1190 00:57:59,535 --> 00:58:01,251 And so I, you know, there's a whole bunch of reasons that, you 1191 00:58:01,251 --> 00:58:03,801 know, if you have to be air gapped and, you know, working on 1192 00:58:03,801 --> 00:58:06,015 something sensitive and classified well then, you gotta 1193 00:58:06,015 --> 00:58:08,415 learn how to work on stuff without it. 1194 00:58:08,505 --> 00:58:12,530 However, right, I mean if the majority of problems will be AI 1195 00:58:12,530 --> 00:58:14,620 assisted, well then maybe that's the way we should be 1196 00:58:14,620 --> 00:58:16,605 interviewing as well. But then I totally understand 1197 00:58:16,605 --> 00:58:18,685 the risks of like all right you're not actually hiring. 1198 00:58:19,428 --> 00:58:23,640 You might be accidentally hiring an AI, not a person so, which is 1199 00:58:23,640 --> 00:58:27,335 a thing. Yeah, hiring an AI, paying for a 1200 00:58:27,335 --> 00:58:29,946 human salary. So I think that's a kind of 1201 00:58:29,946 --> 00:58:32,600 hilarious. So Gene, it's been a pleasant 1202 00:58:32,600 --> 00:58:34,087 conversation. Unfortunately, we have to wrap 1203 00:58:34,087 --> 00:58:36,115 up pretty soon. I have one last question that 1204 00:58:36,115 --> 00:58:38,802 I'd always love to ask my guests, right, to learn from 1205 00:58:38,802 --> 00:58:40,806 you. So I call this the three 1206 00:58:40,806 --> 00:58:43,193 technical leadership wisdom. Just think of it like an advice 1207 00:58:43,193 --> 00:58:44,736 you wanna give to the listeners here. 1208 00:58:44,976 --> 00:58:47,312 So maybe if you can share your version today that would be 1209 00:58:47,312 --> 00:58:49,999 great. Yeah I'll give one that's AI 1210 00:58:49,999 --> 00:58:51,791 specific and one that's not AI specific. 1211 00:58:52,571 --> 00:58:56,589 So I think AI specific, something that Dr. Ron Westrum 1212 00:58:56,589 --> 00:59:00,629 who so much influenced our work in the Accelerate book, DevOps 1213 00:59:00,629 --> 00:59:03,758 Handbook, around culture norms. You know, he introduced me to 1214 00:59:03,758 --> 00:59:05,178 the term the sociotechnical maestro. 1215 00:59:05,772 --> 00:59:09,254 And basically he says really five things: high energy, high 1216 00:59:09,254 --> 00:59:12,282 standards, great in the large, great in the small, and they 1217 00:59:12,282 --> 00:59:16,071 love walking the floor. And so, yeah, high energy, high 1218 00:59:16,071 --> 00:59:18,379 standards, that's obvious. Great in the large, that means, 1219 00:59:18,379 --> 00:59:20,383 you know, to me systems thinking, you know, being able 1220 00:59:20,383 --> 00:59:24,225 to see the larger picture. Great in the small means like, 1221 00:59:24,225 --> 00:59:27,013 yeah you're detail oriented, right, you, you know, care 1222 00:59:27,013 --> 00:59:30,034 about, you know, you don't just believe anything that people 1223 00:59:30,034 --> 00:59:32,322 tell you, right? You're willing to sort of dig on 1224 00:59:32,322 --> 00:59:34,939 the covers and really find out and then love walking the floor, 1225 00:59:34,939 --> 00:59:36,619 right? You love, you know, when you're 1226 00:59:36,619 --> 00:59:39,919 responsible for the work of others, you go to where the work 1227 00:59:39,919 --> 00:59:42,949 is being formed, you actually see how what's in their way. 1228 00:59:43,099 --> 00:59:45,547 You know, are you really creating the conditions, you 1229 00:59:45,547 --> 00:59:47,773 know, to enable people to do their work easy and well? 1230 00:59:48,043 --> 00:59:53,315 I think people engineers leaders who are that are, we all need 1231 00:59:53,315 --> 00:59:55,567 leaders, right? Any endeavor that takes more 1232 00:59:55,567 --> 00:59:58,033 than one person, you know, we need leaders for. 1233 00:59:58,871 --> 01:00:01,989 I think the second thing that I would add that's not AI specific 1234 01:00:01,989 --> 01:00:05,262 is like what is one piece of advice I would give to people 1235 01:00:05,262 --> 01:00:07,918 early in their careers. And I would say find someone 1236 01:00:07,918 --> 01:00:10,614 with lots of problems, right? Someone's working on important 1237 01:00:10,614 --> 01:00:12,978 problems and they can't solve all of them. 1238 01:00:13,208 --> 01:00:17,804 And I think what I found in my career, in fact where Tripwire 1239 01:00:17,804 --> 01:00:21,173 came from was me going to Professor Jean Stafford. 1240 01:00:21,658 --> 01:00:24,850 And he had a lot of research and funding and he had lots of 1241 01:00:24,850 --> 01:00:26,968 problems and projects he wanted to complete. 1242 01:00:27,268 --> 01:00:29,803 And I just, I was free labor as an undergraduate student and I 1243 01:00:29,803 --> 01:00:32,588 got to work on under him on graduate level problems. 1244 01:00:32,608 --> 01:00:35,710 And one of them became Tripwire which became the foundation of a 1245 01:00:35,710 --> 01:00:37,468 company that I was there for 13 years. 1246 01:00:37,468 --> 01:00:40,643 And we did our filing to go public and I'm just incredibly 1247 01:00:40,643 --> 01:00:43,277 grateful for that experience. And I think so often, right, 1248 01:00:43,277 --> 01:00:46,108 that's what it comes down to is like especially early in your 1249 01:00:46,108 --> 01:00:48,368 career, yeah, you don't know really what you're good at. 1250 01:00:48,398 --> 01:00:52,367 And the only way that you can tell is by asking someone with 1251 01:00:52,367 --> 01:00:55,624 more experience, you know, give me a problem to work on, right? 1252 01:00:55,624 --> 01:00:57,994 And you will learn a lot and you get a lot of feedback. 1253 01:00:58,084 --> 01:01:01,020 And I think that actually is applicable, you know, to all of 1254 01:01:01,020 --> 01:01:03,829 us. Wow, I think that's really two 1255 01:01:03,829 --> 01:01:05,817 great advice for people who listen, right? 1256 01:01:05,817 --> 01:01:08,667 So I love the first one as well like kind of summarize the kind 1257 01:01:08,667 --> 01:01:10,077 of leaders they should be, right? 1258 01:01:10,077 --> 01:01:13,110 The important attributes and especially like the walking 1259 01:01:13,110 --> 01:01:15,898 thing, right? Because some leaders are kind of 1260 01:01:15,898 --> 01:01:19,066 like detached from the day-to-day workers, employees, 1261 01:01:19,066 --> 01:01:21,748 right? And they kind of like think it's 1262 01:01:21,748 --> 01:01:23,556 easy to do a lot of stuff, right? 1263 01:01:23,556 --> 01:01:27,189 Because, yeah, think about the layoffs that happening, because 1264 01:01:27,189 --> 01:01:29,859 they think by introducing AI they can just replace people. 1265 01:01:29,919 --> 01:01:31,509 But sometimes it's not as easy as that. 1266 01:01:33,069 --> 01:01:35,898 So Gene, if people love this conversation, they would love to 1267 01:01:35,898 --> 01:01:38,541 connect with you or ask you more questions, is there a place 1268 01:01:38,541 --> 01:01:40,963 where they can find you online? Absolutely. 1269 01:01:41,383 --> 01:01:45,567 I'm on Twitter at @realgenekim. And the other place that I love 1270 01:01:45,567 --> 01:01:47,833 being on is LinkedIn. So just Gene Kim. 1271 01:01:48,283 --> 01:01:51,288 Just Google Gene Kim, vibe coding and you'll find me. 1272 01:01:52,359 --> 01:01:54,894 Right. And I would recommend people to 1273 01:01:54,894 --> 01:01:57,189 check out Gene's book, right? Vibe Coding. 1274 01:01:57,189 --> 01:02:01,094 I think it gives a very good overview of what is important, 1275 01:02:01,094 --> 01:02:04,111 especially these days when people can, you know, vibe code, 1276 01:02:04,111 --> 01:02:07,323 generate code a lot. And I think, there are important 1277 01:02:07,323 --> 01:02:09,380 lessons, especially the engineering practices that I 1278 01:02:09,380 --> 01:02:11,944 find very, very important for people to learn from that book. 1279 01:02:12,484 --> 01:02:14,104 So thank you again for your time today. 1280 01:02:14,104 --> 01:02:15,634 So it's been a pleasant conversation. 1281 01:02:15,964 --> 01:02:17,464 Thank you so much for writing the book as well. 1282 01:02:18,218 --> 01:02:20,558 Oh, Henry, thank you so much for allowing me to be a part of this 1283 01:02:20,558 --> 01:02:21,938 and keep up all your amazing work.