Programming Isn't Dying, It's Climbing the Abstraction Ladder의 전체 녹취록

extraordinary times. 00:03:30.479 --> 00:03:35.280 In a lot of ways, objectively, it is extraordinary to have machine intelligence, very sci-fi, you know, yeah, yeah. 00:03:35.439 --> 00:03:36.400 And everything like that. 00:03:36.560 --> 00:03:45.280 Um, but then you go, okay, well, you know, when programmers moved from hand rolling uh assembly code, you know, to to generating machine code or you know. 00:03:45.520 --> 00:03:51.840 I wasn't quite in that era, but I mean I started programming in the in middle eighties and then professionally in in '95. 00:03:52.319 --> 00:03:58.719 So I was close enough to talk to people that went through that era and to sort of live through certain aspects of it. 00:03:58.879 --> 00:04:00.560 So it's a little visceral for me. 00:04:00.639 --> 00:04:10.960 Like I've and I lived through the era right at the beginning of my career where Visual Basic and other you know kind of 4GLs came along, and then people were like, Yeah, you don't need programmers anymore. 00:04:11.039 --> 00:04:12.400 So, you know, it keeps happening. 00:04:12.879 --> 00:04:16.079 You tell people all that and then they nod, and then they don't get it. 00:04:16.240 --> 00:04:18.319 It's like we're moving up the abstraction ladder. 00:04:18.560 --> 00:04:26.000 There's gonna be a new set of frameworks, new set of tools, new set of ways to version the artifacts that we want to collaborate on. 00:04:26.160 --> 00:04:26.399 Right. 00:04:26.720 --> 00:04:27.120 Right. 00:04:27.920 --> 00:04:41.600 But the act of putting that all together and you know translating what the executives want, let's say capital, because it's gonna become startlingly clearer uh very, very soon, I think, what that actually means for society. 00:04:41.680 --> 00:04:45.279 But let's say the people that call the shots they don't want a program. 00:04:45.680 --> 00:04:57.279 No, they may think they want a program, but some of them, like like my great CEO Brandon, uh loves the vibe code and he puts stuff together directly based on his uh ideas, but they're they're still prototypes. 00:04:57.439 --> 00:04:57.680 Sure. 00:04:57.759 --> 00:04:58.079 Yeah. 00:04:58.399 --> 00:05:01.360 He doesn't actually want to be a programmer, he wants to be a CEO. 00:05:01.519 --> 00:05:02.160 Yeah, right? 00:05:02.240 --> 00:05:02.399 Yeah. 00:05:02.639 --> 00:05:08.800 Based on that experience, based on the perspective, based on how I see things come together, I say, okay, we're moving up the abstraction layer. 00:05:08.959 --> 00:05:11.920 Where's the next level languages and frameworks? 00:05:12.000 --> 00:05:18.399 Where, you know, what's the thing that's gonna get branded with a brand where people are like, now I'm a brand programmer, right? 00:05:18.639 --> 00:05:18.800 Yeah. 00:05:18.959 --> 00:05:27.920 So here in this conference, I've been talking to people, and I call it I you know since we're at Rails World and I'm a Rails guy, you know, I'm like, okay, well, what is Rails next? 00:05:28.480 --> 00:05:42.639 Let's imagine that trademark issues aside, we were able to create the layer, you know, the the translation layer that turned a set of artifacts, written down artifacts, obviously, you know, preferably in GitHub. 00:05:42.800 --> 00:05:43.360 Yeah, yeah. 00:05:43.600 --> 00:05:55.519 You know, so we can collaborate easily on it, you know, put that through a layer that transforms it into a properly written Rails app in such a way that's repeatable that you don't have to actually look at that code. 00:05:55.680 --> 00:05:55.920 Yeah. 00:05:56.079 --> 00:06:01.759 Because that code is now the machine code or the intermediate language or you know, pick your poison for that particular trope. 00:06:01.839 --> 00:06:06.639 And then people just I I mean, some people get it, but I think uh as a whole, people are not getting it. 00:06:06.720 --> 00:06:08.800 They're like, oh, you know, I don't code anymore. 00:06:08.959 --> 00:06:09.759 Well, it's like, guess what? 00:06:09.920 --> 00:06:12.079 The code where the code is changed. 00:06:12.399 --> 00:06:13.759 Yeah, I totally agree with you there. 00:06:13.839 --> 00:06:15.519 And we were talking about this yesterday. 00:06:15.759 --> 00:06:27.360 To me, the exciting thing is the harnesses that are being built and the knowledge graphs, all the pieces that you're putting into place around the agents to support them and get the outcomes that you want. 00:06:27.519 --> 00:06:32.160 Because that to me is creative problem solving, which is what I got into programming to do. 00:06:32.319 --> 00:06:32.480 Yeah. 00:06:32.639 --> 00:06:38.079 Um, I'm curious because I know like you have a soft spot in your heart for this, you know, for this kind of development. 00:06:38.160 --> 00:06:40.240 You've been doing it for longer than most people. 00:06:40.319 --> 00:06:46.240 So where do you see that fitting in to let's say Rails Next or into this higher level of abstraction? 00:06:46.319 --> 00:06:50.160 Like, are we are we touching it when we start working on the tools that support the agents? 00:06:50.480 --> 00:06:56.319 Yeah, you're touching it directly, you're manipulating text symbols that end up being inversion control. 00:06:56.480 --> 00:06:56.720 Yeah. 00:06:57.120 --> 00:07:00.959 That that is a paradigm that I think it's premature to call that dead. 00:07:01.199 --> 00:07:01.360 Yeah. 00:07:01.519 --> 00:07:07.439 You know, you want easy collaboration, you want repeatability, and that's that to me, that's still the way to do it. 00:07:07.600 --> 00:07:13.199 Like right now, everyone's just got you know running around with like a chicken with their head cut off and like trying all sorts of different things. 00:07:13.360 --> 00:07:17.040 There's a million different ways to do spectrum development. 00:07:17.199 --> 00:07:34.240 But we we also live in a fairly large corner of the programming world, which is like kind of web and consumer mobile kind of stuff, but the programming world

is is big and getting bigger because it's much easier to apply software to all sorts of things than it was ever before. 00:07:34.560 --> 00:07:39.759 Man, if people think the AI bubble is huge, just wait till we get into the robotics bubble. 00:07:40.480 --> 00:07:40.879 What do you mean? 00:07:41.040 --> 00:07:41.519 Say more about that. 00:07:41.839 --> 00:07:47.519 Capital wants to replace white-collar labor because now it's apparent that you can probably replace a lot of layers of it. 00:07:48.079 --> 00:07:50.000 Wait till they can replace blue collar. 00:07:50.079 --> 00:07:52.160 They at least like white-collar kinda. 00:07:52.800 --> 00:07:54.240 They don't like blue collar. 00:07:54.319 --> 00:07:56.000 I mean, pardon the spicy take partner. 00:07:56.240 --> 00:07:56.560 Yeah. 00:07:56.720 --> 00:07:57.360 No, that's right. 00:07:57.680 --> 00:07:58.240 You get where I'm coming from. 00:07:59.279 --> 00:08:01.040 You get where I you get where I'm coming from, though, right? 00:08:01.199 --> 00:08:01.519 No, I do. 00:08:01.680 --> 00:08:01.839 I do. 00:08:02.079 --> 00:08:02.240 Yeah. 00:08:02.399 --> 00:08:04.079 This is a tale as old as time, right? 00:08:04.399 --> 00:08:08.959 You know, we wanna, you know, we could make all this stuff a lot cheaper if it wasn't for all the people we had to pick. 00:08:09.199 --> 00:08:12.879 But if there's no consumers, like who's gonna empower the economy? 00:08:13.120 --> 00:08:13.519 There isn't. 00:08:13.839 --> 00:08:18.319 And I'm like, did you forget that history didn't end like 10 years ago with Obama? 00:08:18.800 --> 00:08:22.720 And history includes things like a lot of people dying. 00:08:23.040 --> 00:08:24.399 Yeah, yeah, for sure. 00:08:24.639 --> 00:08:29.600 You know, are we that arrogant to think that history, you know, that particular kind of history is not gonna repeat itself? 00:08:29.680 --> 00:08:31.920 And I think that's where people get into the doom and gloom. 00:08:32.240 --> 00:08:32.960 That's for sure. 00:08:33.200 --> 00:08:41.279 You know, and it is scary, and I got kids, and I hope to have grandkids someday, and it's you know, the sort of thing that I go, okay, well, you know, hopefully we don't go down that route. 00:08:41.360 --> 00:08:53.759 But I think one of the takeaways from DHH, you know, just bringing back here to Austin that was cool yesterday, is the idea that okay, we we face an uncertain future, we participate in creating that future. 00:08:53.919 --> 00:08:55.039 Let's be optimistic. 00:08:55.120 --> 00:08:59.759 Yeah, you know, let's manifest a cool utopian, you know, future with all this shit. 00:08:59.840 --> 00:09:02.399 I mean, I don't know that he's necessarily saying utopia. 00:09:02.480 --> 00:09:02.960 I'm saying that. 00:09:03.279 --> 00:09:12.639 The part that I respect is the um, hey, with no information about the future, or with very imperfect information about the future, we have two options. 00:09:12.799 --> 00:09:15.519 One is to be fearful and the other is to be optimistic. 00:09:15.679 --> 00:09:18.320 Yeah, and it's the rational choice is to be optimistic. 00:09:18.720 --> 00:09:22.240 The ra the rational choice, it's also you know, fear is the mind killer. 00:09:22.480 --> 00:09:25.279 And I I truly believe that I have believed that all the whole time. 00:09:25.360 --> 00:09:34.559 And people accuse me of having paranoia all the time, which is you know, another way of saying the universe is out to help me instead of out to get me. 00:09:34.720 --> 00:09:34.879 Yeah. 00:09:35.200 --> 00:09:36.960 That truly works in different ways. 00:09:37.120 --> 00:09:43.679 I mean, you know, some people call it God, some people call it uh you know, manifesting, some people you know wear crystals around their neck. 00:09:44.480 --> 00:09:47.600 That doesn't work actually, but sorry, crystal people. 00:09:47.840 --> 00:09:48.240 Is that it? 00:09:48.480 --> 00:09:49.200 Have you tried it? 00:09:49.440 --> 00:09:49.840 You don't know. 00:09:50.080 --> 00:09:51.039 I tried and it didn't work. 00:09:51.200 --> 00:09:54.000 So empirical evidence. 00:09:54.240 --> 00:09:56.159 Yeah, yeah. 00:09:56.399 --> 00:09:57.600 Yeah, I I get it. 00:09:57.759 --> 00:10:04.240 And I know people, and I just wrote I just wrote this post on on X that went viral and about meat proxies. 00:10:04.320 --> 00:10:04.639 Uh huh. 00:10:04.799 --> 00:10:06.000 And I do have this real friend. 00:10:06.080 --> 00:10:07.519 Some people are like, uh, Claude wrote this. 00:10:07.600 --> 00:10:08.480 I was like, fuck off. 00:10:08.639 --> 00:10:14.159 This is the most not Claude writing that I've done since Opus came out. 00:10:14.320 --> 00:10:15.759 So, like, you know, fuck off. 00:10:15.840 --> 00:10:16.639 That was really me. 00:10:16.720 --> 00:10:19.519 And some people were like, that friend doesn't exist, and this story didn't happen. 00:10:19.679 --> 00:10:20.559 I was like, fuck off. 00:10:20.639 --> 00:10:21.919 This friend does exist. 00:10:22.399 --> 00:10:35.279 He does work for a company that most people, at least programmers, have heard of, and he is a staff engineer, and he is afraid of for his future, but in ways that I found enlightening rather than oh yeah, that checks out. 00:10:35.360 --> 00:10:47.120 Like the enlightening part is that this may be another case where people whose heart is not in the industry end up getting pushed out because those are the people that are most afraid, right? 00:10:47.200 --> 00:10:54.480 Like if your heart's in this industry, if you're a creator specifically with programming, this is my friend's creative and he's super smart. 00:10:54.639 --> 00:11:01.440 It's just that he read the article and he he sent me a message afterwards and he says, Hey, dude, the thing is, my heart's never been in this career. 00:11:01.519 --> 00:11:02.480 And I was like, Oh. 00:11:03.360 --> 00:11:03.840 Yeah. 00:11:05.039 --> 00:11:06.480 Yeah, about that. 00:11:06.639 --> 00:11:17.279 So

yeah, that's not gonna work out very well for you in this in this phase that's coming up because like what's gonna be rewarded in this phase that's coming up is really the applied creativity towards actual programming. 00:11:17.519 --> 00:11:26.879 I agree with you, and I'm gonna add that I think it also rewards a uh a resilience to change that in my mind was what I signed up for by getting into this industry. 00:11:27.039 --> 00:11:31.279 Like I signed up to be at the forefront of a very changing, turbulent industry. 00:11:31.440 --> 00:11:33.360 And it's true, I didn't predict this amount of change. 00:11:33.440 --> 00:11:39.519 Yeah, what I had prided myself on and what I tried to bring my team up on was the ability to adapt to change quickly. 00:11:39.679 --> 00:11:49.440 Yeah, and I think when you combine that with a true love of problem solving in a creative fashion with computers, that's really where you can maybe feel more comfortable. 00:11:49.759 --> 00:11:55.840 You have to be ready to change your the label that you use, you know, as a way to describe yourself. 00:11:56.000 --> 00:11:58.480 I mean, when I was a kid, I was a nerd or a hacker. 00:11:58.639 --> 00:12:05.600 When I got involved professionally into software, I was for eight, nine years a Java programmer. 00:12:06.240 --> 00:12:13.200 And then I discovered Ruby and Ruby on Rails in 2004, 2005, and I became a Rails programmer. 00:12:13.440 --> 00:12:17.840 And you know, those labels have their use, but I think people get attached to them. 00:12:18.000 --> 00:12:18.320 Yeah. 00:12:18.639 --> 00:12:21.600 And you gotta be willing to make the leap when something happens. 00:12:21.759 --> 00:12:24.639 And maybe this is the death of Rails, you know, at some point. 00:12:24.720 --> 00:12:32.159 Maybe it is the death of the current generation of programming languages as we move into whatever that next level of abstraction is. 00:12:32.320 --> 00:12:35.440 We've been here before and you survive and you adapt. 00:12:35.600 --> 00:12:41.679 I imagine that a lot of people here at this conference probably have only done Rails or have only done web paradigms. 00:12:42.159 --> 00:12:42.879 Yeah, that's possible. 00:12:42.960 --> 00:12:43.360 Yeah. 00:12:43.600 --> 00:12:49.759 When I was a kid, I I did the whole BBS scene thing and video games and like copied source code out of magazines. 00:12:49.919 --> 00:13:09.519 And then in my earliest years of programming, I I did a service dispatching system and small talk and then some database stuff, power builder, heavy-duty log processing and Pearl, Diagnostic Systems, and and Java Micro Edition at uh Daimler Chrysler in the early 2000s. 00:13:09.679 --> 00:13:10.799 It's all programming. 00:13:10.960 --> 00:13:13.039 And it's all super fucking cool, you know. 00:13:13.120 --> 00:13:15.279 Like if you find programming cool. 00:13:15.440 --> 00:13:15.759 Yeah. 00:13:15.919 --> 00:13:16.159 Right? 00:13:16.240 --> 00:13:28.000 And like the things

that I'm doing now with Chad Fowler on regenerative software and like uh, you know, being able to move up that abstraction layer so that the source code that we now consider sacred is no longer sacred. 00:13:28.080 --> 00:13:29.039 It's actually disposable. 00:13:29.200 --> 00:13:29.440 Yeah. 00:13:29.600 --> 00:13:31.120 That's super fucking cool, also. 00:13:31.360 --> 00:13:32.960 That and that's also programming. 00:13:33.200 --> 00:13:34.159 It is also programming. 00:13:34.320 --> 00:13:35.279 Yeah, I totally agree. 00:13:35.519 --> 00:13:37.679 Regenerative software is a really interesting idea. 00:13:38.000 --> 00:13:38.879 Is that the future? 00:13:39.120 --> 00:13:42.320 It's really bad for the success of the well-grided Rubyist. 00:13:42.480 --> 00:13:45.200 So I don't really love this idea, but I think it might be true. 00:13:45.440 --> 00:13:50.399 Yeah, and I'm, you know, I'm super closely associated with Ruby in the Rails world and and whatnot. 00:13:50.559 --> 00:14:01.840 So, you know, it's not necessarily the best thing for me as well, but I think that the attachment to the current paradigm, which is clearly going away, get a fucking grip. 00:14:02.159 --> 00:14:06.879 People, you're not going to be doing code review very, very soon. 00:14:07.039 --> 00:14:11.679 If you continue to insist on doing it, you will be obsolete very soon. 00:14:11.840 --> 00:14:16.879 And by very soon, I mean like within a matter of a couple of years, you will be obsolete. 00:14:17.120 --> 00:14:20.080 People will look at you and they will laugh in pity. 00:14:20.320 --> 00:14:29.840 And a very, very few of you will retain your employability in the same way that people are still doing COBOL and still doing to some degree Java. 00:14:30.159 --> 00:14:37.360 You know, and the rest of us will move on to newer paradigms, which are still programming and are super fucking cool and involve new. 00:14:37.840 --> 00:14:42.960 I see a huge greenfield of programming, and I think that's where people lose the plot. 00:14:43.440 --> 00:14:43.679 Right? 00:14:43.759 --> 00:14:47.919 Because they look at this other thing, they go, This doesn't look like what I was doing, so I don't want to do it. 00:14:48.320 --> 00:14:48.960 Yeah, right. 00:14:49.279 --> 00:15:12.480 The reason regenerative software is so so interesting to me is that it unlocks being able to make that mental shift because it says this code that you thought was so sacred that you need to maintain and make sure it's well architected and make sure that every line is correct, you know, according to whatever your flavor of correct is, you know, well tested, well architected, you know, whatever. 00:15:12.799 --> 00:15:13.840 It's actually bullshit. 00:15:13.919 --> 00:15:15.840 You can delete it and you can regenerate it. 00:15:16.000 --> 00:15:36.000 And anytime that you want to fix something about that code, you want it to be better factored, you want it to be better tested, you want it to adhere to a different uh architectural paradigm, you want to change the programming language that it's implemented in, you can the same way that you compile source code into something else, you do, right? 00:15:36.399 --> 00:15:38.879 And so how is this working mechanically? 00:15:39.039 --> 00:15:40.559 Like, how do you see it working mechanically? 00:15:40.960 --> 00:15:45.679 The reason that people go, hey, you're full of shit, you can't do that, is because the agents are non-deterministic. 00:15:45.840 --> 00:15:49.279 And they go, compiler's deterministic, so you can prove that the output is correct. 00:15:49.840 --> 00:15:51.679 And they go, agents are non-deterministic. 00:15:51.759 --> 00:15:53.679 So you go, fine, put guardrails in place. 00:15:53.919 --> 00:16:02.320 My colleague Sebastian has a metaphor that I I think is really cool, which is like if you take a kid to a bowling alley and you want him to have fun, put up the bumpers. 00:16:02.879 --> 00:16:08.879 Because even if you know throw it at a wild angle, instead of going into the gutter, it's gonna bounce around and it's finally gonna knock down some pins. 00:16:09.120 --> 00:16:09.360 Right. 00:16:09.519 --> 00:16:16.080 Well, right now, agents are at a sufficiently primitive state that they need bumpers. 00:16:16.399 --> 00:16:16.720 Right? 00:16:16.879 --> 00:16:18.240 So, what are the bumpers? 00:16:18.399 --> 00:16:20.080 That's actually a cool software challenge. 00:16:20.240 --> 00:16:21.039 What are the bumpers? 00:16:21.360 --> 00:16:26.480 Now, the reason that you don't see more, and this is this is my cynical, you know, 52-year-old take. 00:16:26.720 --> 00:16:36.320 The the cynical reason I think people aren't actually developing the bumpers is that the lack of existence of those bumpers right now is our job security. 00:16:36.639 --> 00:16:55.120 That's interesting, but it it could it also be that not everybody sees it the same way, like they don't see into the future where it's very convenient for them to not see it that way because there's an expression I can't remember exactly how it goes, but you know, like don't ask someone to work on something that's gonna take away their job or something like that because you're not gonna you're not gonna get it. 00:16:55.279 --> 00:17:03.600 I think as an industry, we're not focusing on that, except for a couple of suicidal wackos who are like, Yeah, we can replace senior developers, no problem. 00:17:03.679 --> 00:17:04.960 We just need sufficient guardrails. 00:17:05.200 --> 00:17:09.279 But the thing is you replace the senior developers, that actually is the unlock. 00:17:09.440 --> 00:17:16.799 Because then once people are like, you know, off the cliff and realize that there's nothing under them, they're gonna fall, then they're like, Oh, maybe I should flatten my wings. 00:17:16.880 --> 00:17:18.000 Oh shit, I have wings. 00:17:18.160 --> 00:17:18.799 Yeah. 00:17:19.440 --> 00:17:20.720 I can fly. 00:17:21.119 --> 00:17:26.240 You

know, because this thing gives you magic powers to imagine whatever reality you want, right? 00:17:26.480 --> 00:17:44.079 Yeah, this does resonate with me because in my conversations, which I you know I have had a lot of, and I've had a lot even in the last couple of days, there is a certain sense of resignedness, or just like, this is the way it is, and I don't know anything else we can do about it, and we just have to do it this way. 00:17:44.319 --> 00:18:05.359 You know, whether it's agentic development or it's the way that the person is working on a team, or the way that software development, like the direction it's heading, which makes me feel well, I feel bad about it because I think we're software engineers with agency, and we can, you know, we have power and we have a team, most of us, with which we can overcome a lot of these challenges that we are feeling existentially. 00:18:05.680 --> 00:18:11.279 I think it's human nature to want to hold on to the thing that makes sense and you know, to you at the time. 00:18:11.440 --> 00:18:16.480 And this is why this meat proxy incident, let's call it, was so instructive for me. 00:18:16.559 --> 00:18:18.400 I'm like, why am I so excited? 00:18:18.480 --> 00:18:21.440 And why are so many people so upset? 00:18:21.680 --> 00:18:24.240 The thing is, if you're used to creating the future, right? 00:18:24.319 --> 00:18:33.440 Like if you're used to creating new open source projects just like that because you wanted to, uh, you know, or just starting companies because you wanted to, like making things happen. 00:18:34.000 --> 00:18:38.960 But sometimes we forget that that kind of person is somewhat in the minority. 00:18:39.119 --> 00:18:40.319 I think it's safe to say that, right? 00:18:40.480 --> 00:18:40.640 Yes. 00:18:40.799 --> 00:18:45.440 Most people just want to be told what to do and do their job well so that they can get on with their lives. 00:18:46.319 --> 00:18:51.759 And in this particular area of of humans, you know, like humans are well compensated to do that here. 00:18:51.920 --> 00:19:05.359 And in fact, I've been in enough corporate situations where people tell me, like, hey, actually, you suck more than the person who's just uh dependable and delivers the same results every day so that I can make proper time estimates based on their performance. 00:19:05.440 --> 00:19:07.119 I'm like, Yeah, but they don't know what they're doing. 00:19:07.200 --> 00:19:11.680 Yeah, and it's like, yeah, no, they may not know what the fuck they're doing, but that at least they do it consistently. 00:19:12.079 --> 00:19:13.519 So you're fired, just knocking off the city. 00:19:14.240 --> 00:19:21.599 You know, and it's like, and then you know, tell me in the comments or send me an email if you find yourself ever doing that because like I want to talk to you. 00:19:21.680 --> 00:19:23.039 Yeah, we're in a club together. 00:19:23.279 --> 00:19:26.160 Those people right now, they're they're so screwed.

00:19:26.319 --> 00:19:36.319 Those good dependable people are so screwed for a little while because right now they can tell that a software AI agent can do their job very, very well, right? 00:19:36.640 --> 00:19:43.680 But maybe it's up to us as a person who is creative and a and a leader and visionary to say, okay, this is the way. 00:19:43.759 --> 00:19:51.039 Because I still me, I still think we need those good dependable people, but we can't count on them to all sprout wings and fly in this new reality. 00:19:51.119 --> 00:19:51.920 They need to be led. 00:19:52.240 --> 00:19:55.519 But the the transition is gonna be bumpy, sure. 00:19:55.759 --> 00:20:00.160 You know, because right now, unless they start at least flapping their arms, you know. 00:20:00.319 --> 00:20:03.519 Yeah, so I don't know, to abuse that metaphor a little bit. 00:20:03.680 --> 00:20:08.559 Okay, so so they haven't been able to manifest wings for themselves, but they start flapping their arms. 00:20:08.640 --> 00:20:12.720 That's like the one that's like sitting there and at least enthusiastically hitting enter. 00:20:12.880 --> 00:20:13.119 Sure. 00:20:13.279 --> 00:20:16.160 Whenever to get Claude to do stuff, right? 00:20:16.480 --> 00:20:18.319 Now there's different ways to react to that. 00:20:18.480 --> 00:20:30.240 So I have another friend, and this is a real friend, you know, who I like, and I would have considered hiring at some point, but like maybe a year ago, he's like, Oh my god, I love Claude and I love coding agents, it's so amazing. 00:20:30.720 --> 00:20:34.160 I got like two jobs and I'm working like two hours a day. 00:20:34.240 --> 00:20:36.880 I'm like not exaggerating, he's like admitting this to me, right? 00:20:36.960 --> 00:20:37.440 Yeah, yeah. 00:20:37.599 --> 00:20:42.079 The engineering manager slash entrepreneur, you know, CTO and me is like recoiling in horror. 00:20:42.160 --> 00:20:43.039 No, don't tell me this. 00:20:43.440 --> 00:20:43.920 I know, right? 00:20:44.000 --> 00:20:47.920 He's like, I'm racing motorcycles and I'm making music and then all this stuff. 00:20:48.720 --> 00:20:56.720 Yeah, so a lot of those stories cropped up of like contractors who took like five different contracts, yeah, right, and was billing quote unquote, you know, oh my god. 00:20:57.519 --> 00:21:03.599 Yeah, no, it was it was horrifying, and I was like, Oh my god, like this is this is uh gonna be an interesting time we're moving into. 00:21:03.920 --> 00:21:04.160 Sure. 00:21:04.319 --> 00:21:16.640 I I guess if your job has certain expectations on you and you're able to meet them in an hour because of AI, and you spend the other seven hours or whatever playing with your kids or racing motorcycles or whatever you do, good for you. 00:21:17.279 --> 00:21:19.200 I hope that lasts for you. 00:21:19.440 --> 00:21:21.039 But it's probably not gonna work for very long. 00:21:21.279 --> 00:21:25.599 Not in the long term, but then again, big companies and big

governments. 00:21:25.920 --> 00:21:29.359 Most government software shops are not really using AI, as far as I know. 00:21:29.680 --> 00:21:34.319 Yeah, I mean, we are in uh a couple of government projects, and it's a it's a slow adoption curve. 00:21:34.480 --> 00:21:40.000 And and to be honest, sometimes it's really like, well, we know you're using it, but we don't want to hear about it. 00:21:40.160 --> 00:21:40.960 Yeah, not yet, right? 00:21:41.039 --> 00:21:52.400 Because we don't have a policy, we don't have a you know we're just so that that adds to the inertia, but I I mean what were we talking about before, you know, in terms of leadership and like cre it's just we're still green, we're still so so green. 00:21:52.480 --> 00:21:59.759 Like there's still many different ways to do this like kind of spectrum how do you reify the concepts that are embodied. 00:22:00.160 --> 00:22:04.720 Involved in doing this next level of abstraction, like what makes sense? 00:22:05.039 --> 00:22:12.079 We're still trying to adapt terms like spec and you know, it's like those kinds of ideas that exist in the current paradigm. 00:22:12.160 --> 00:22:22.880 So, like, if you say to someone about regenerative software that you delete the tests also, yeah, because they're considered part of the implementation, they're like, What? 00:22:23.039 --> 00:22:26.640 Yeah, like how will you know that the thing is correct if you also delete the tests? 00:22:26.720 --> 00:22:29.839 No, no, no, you have to write all this other stuff that's the contract. 00:22:30.000 --> 00:22:30.240 Yeah. 00:22:30.480 --> 00:22:34.559 I think it's a time where you need to coin new terminology, it's a new paradigm. 00:22:35.599 --> 00:22:41.599 It is more helpful to coin new terms at this stage in history so that people can break free of the previous paradigm. 00:22:41.920 --> 00:22:43.119 Yeah, I agree with you totally. 00:22:43.200 --> 00:22:54.720 And in fact, I'm I'm doing a lightning talk later on today, and part of it is based on the fact that I'm striving to look for new principles because I came up with agile and I came up with XP, and I came up with Agile. 00:22:54.880 --> 00:22:56.160 Wait, you came up with Agile? 00:22:56.960 --> 00:22:58.480 I know I'm sorry I didn't come up with it. 00:22:58.559 --> 00:23:00.079 I mean, I grew, you know, I grew up with it. 00:23:00.319 --> 00:23:01.279 I mean I came up with it. 00:23:01.359 --> 00:23:01.920 I grew up with it. 00:23:02.079 --> 00:23:04.079 I was about I was about to say, get him. 00:23:05.440 --> 00:23:07.440 And these concepts don't work anymore for me. 00:23:07.599 --> 00:23:10.240 Yeah um in TDD was something that I loved. 00:23:10.319 --> 00:23:12.079 I mean, I staked my career on TDD. 00:23:12.160 --> 00:23:12.480 Yeah. 00:23:12.720 --> 00:23:16.000 And you know, I gotta let that go. 00:23:16.160 --> 00:23:20.079 And so I totally agree with you that this is a blank canvas to some extent. 00:23:20.240 --> 00:23:24.880 We really don't know what the principles of organizing teams and building software. 00:23:25.200 --> 00:23:31.759 We may see where it's working and where it's not, but we by no means do we have a uh a common language to talk about. 00:23:32.079 --> 00:23:38.319 You know, in terms of people worrying about job security, I think it's really, really silly. 00:23:38.960 --> 00:23:41.599 You know, I do believe in Java's paradox. 00:23:41.680 --> 00:23:43.599 I do believe that an expansion is coming. 00:23:44.000 --> 00:23:48.000 I think people will be able to apply software more and more places. 00:23:48.240 --> 00:23:58.160 And eventually, uh after you know, a bumpy ride, as I said before, even those people who are not, you know, crazy creative and whatever will find a place. 00:23:58.400 --> 00:24:02.960 I can't help going to the dialectic of you know, like capital and and labor. 00:24:03.119 --> 00:24:08.160 I'm a capitalist, you know, like I like starting companies, I like investing, I like the whole system. 00:24:08.319 --> 00:24:11.839 And I I do VC and I work at an Andreasen company for fuck's sake. 00:24:12.000 --> 00:24:28.799 You know, I like that system, and I also recognize that the people with the capital that actually fund innovation and advancement and and whatever, they have huge fucking egos, and they think that they're never wrong, even when they're very, very wrong. 00:24:28.880 --> 00:24:41.839 And you know, if you're talking about venture capital, and more than 90% of the cases, they're wrong, and you know, they're just wrong overall, yeah, you know, in what they're doing, but they're also wrong on a daily and hourly basis with the decisions that they make and that they carry down. 00:24:42.000 --> 00:24:48.240 But their egos cannot handle being that wrong, not that much, not at that level, and not that often, right? 00:24:48.400 --> 00:24:52.160 So like they need other people to be wrong in their stead, right? 00:24:52.400 --> 00:24:52.720 Right? 00:24:52.880 --> 00:24:55.279 Uh aka accountability, right? 00:24:55.440 --> 00:24:55.680 Right? 00:24:55.839 --> 00:25:08.559 They need accountability, so like you know, that's that ensures that there's a next level below capital that is accountable, and that next level below capital that works directly with capital is still very well paid and very prestigious position. 00:25:08.640 --> 00:25:12.720 That's like your CEOs and your directors and you know, yeah, your VPs and stuff like that. 00:25:12.799 --> 00:25:16.480 So they still have big egos, so they also need accountability, you know. 00:25:16.640 --> 00:25:19.599 Yeah, so like they need another level below those and so on and so forth. 00:25:19.680 --> 00:25:20.799 And you get my drift, right? 00:25:20.880 --> 00:25:33.359 Like, there's still always in a corporate kind of situation, even in startups, going to be a need for the people that are actually at the intersection of getting things done and being accountable for it because the

executives are not. 00:25:33.599 --> 00:25:34.559 Yeah, I agree with you there. 00:25:34.720 --> 00:25:38.480 You know, and as the economy expands into like robotics and software, etc. 00:25:38.640 --> 00:25:46.240 etc., like there will be a lot of programming to do, even if it looks it's in a different kind of structured language than the than Ruby. 00:25:46.319 --> 00:25:46.480 Yeah. 00:25:46.880 --> 00:25:55.200 I found this interesting in yesterday's keynote first that we're at Rails World and DHH almost never mentioned Rails in his one-hour keynote. 00:25:55.279 --> 00:25:55.920 I thought was interesting. 00:25:56.079 --> 00:25:57.039 That was probably a mistake. 00:25:57.279 --> 00:25:57.599 Yeah. 00:25:57.759 --> 00:26:10.240 Uh but I don't know if it's a mistake because he also said that only three percent of the code that he's written over the last 10 months has been Ruby, and that a lot of it has been in languages that he doesn't know or read or write. 00:26:10.640 --> 00:26:12.400 I've known David for a long time. 00:26:12.640 --> 00:26:14.880 Rails is his baby, he's not going anywhere. 00:26:15.119 --> 00:26:15.680 Yeah, yeah. 00:26:15.920 --> 00:26:16.880 He's not gonna let it go. 00:26:17.119 --> 00:26:18.000 Oh, I get that. 00:26:18.160 --> 00:26:18.880 I get that. 00:26:19.119 --> 00:26:22.960 But it is interesting, that's his baby, and he's still not writing it all that much. 00:26:23.200 --> 00:26:25.119 Yeah, that that's situational though. 00:26:25.519 --> 00:26:30.240 And if you've been around long enough, you know that there's been periods before where he wasn't actually writing it that much. 00:26:30.480 --> 00:26:30.799 Okay. 00:26:30.960 --> 00:26:39.359 It's not like now where he's like actually talking about Rust and in this way, but I mean, I believe there's been similar kind of periods, and I don't particularly worry about it. 00:26:39.599 --> 00:26:46.640 I do think that Rails is a good paradigm for that shape of software. 00:26:46.880 --> 00:26:47.279 Right. 00:26:47.519 --> 00:26:51.200 I mean, fuck, I'm gonna I'm just gonna say I think it's the best paradigm. 00:26:51.279 --> 00:26:52.559 Obviously, I'm biased. 00:26:52.799 --> 00:26:53.279 We're biased, yeah. 00:26:53.519 --> 00:26:56.079 You know, we're we're biased and we have commercial interests. 00:26:56.240 --> 00:26:56.559 Yeah. 00:26:56.720 --> 00:26:58.720 So yeah, we believe that. 00:26:59.039 --> 00:26:59.519 For now. 00:26:59.920 --> 00:27:05.519 No, I I think you know, as long as this particular paradigm holds, it's still the best. 00:27:05.680 --> 00:27:11.920 You know, as long as the paradigm of LLMs and the way that they're trained and the base knowledge they have, it's still the best. 00:27:12.079 --> 00:27:12.240 Yeah. 00:27:12.400 --> 00:27:27.440 You know, as long as LLMs are in the mix and the way that they think and reason in text, then Ruby is also very viable because Ruby, I believe, has the most readable code of all the major programming

languages. 00:27:28.000 --> 00:27:30.240 So I think it has an advantage there as well. 00:27:30.559 --> 00:27:31.920 So I have a question for you. 00:27:32.160 --> 00:27:36.799 You are one of the foremost orchestrators of LLMs that I know of. 00:27:36.960 --> 00:27:59.759 And by that I mean, you know, you've been on construction of harnesses, knowledge graphs, and other tools to support agents for longer than anybody I know, and have you know built up the techs around and companies are for not the first time for you, I know, that supports agenc development in a way that I uh I think is you know rivals any of the biggest or best larger companies. 00:28:00.000 --> 00:28:18.240 What I've noticed is that there sound it seems like there are two schools of thought around orchestrating agents, and there is one that says, and I would say this is like the Scott Werners of the world, the Jessica Kerr's of the world, that say, okay, well, don't treat agents like people, don't anthropomorphize them, treat them as objects or treat them as tools. 00:28:18.400 --> 00:28:32.559 And then there's people on this other side of the divide where I would put you, Steve Yeege, I would put uh a number of others that lean totally in the other direction and say, no, no, we we want to give them names, we want to make them part of the team, and we want to work with them that way. 00:28:32.720 --> 00:28:38.240 And so I'm curious, because you have had a lot of success, why is it that you come down on that side of the divide? 00:28:38.960 --> 00:28:43.519 Because it is a very useful model, mental model. 00:28:43.680 --> 00:29:09.200 The LOM models themselves understand that model, and we have decades of understanding of how to organize companies, organizational design of companies and software development lifecycles, and and things that lead you to be able to decompose big projects into smaller projects and to tasks, etc., etc., in ways that are easier to do when you use that mental model. 00:29:09.359 --> 00:29:20.559 We don't have that I am aware of mental models around how to direct a swarm of bees or um an antive to get business work done. 00:29:20.799 --> 00:29:22.079 That's what you would have to have. 00:29:22.240 --> 00:29:24.559 Right, I see in order to make it work the other way. 00:29:24.720 --> 00:29:26.079 So I go, no, that's fine. 00:29:26.160 --> 00:29:41.039 Like if we're gonna have a piece of software that is responsible for doing something that looks like a job role that you could give to a human, then I want to be able to say Carlos is working on that, even though Carlos is an AI. 00:29:41.200 --> 00:29:49.519 And it makes it easier to say Carlos is working on that if he has a personality in a picture and we can all kind of visualize who Carlos is and what he's doing. 00:29:49.599 --> 00:29:49.839 Yeah. 00:29:50.079 --> 00:29:56.240 You know, as opposed to the blickety blick, you know, underscore WM 16 running on, you know, whatever. 00:29:56.400 --> 00:29:59.200 I'm not saying that that's the only way to use AI. 00:29:59.359 --> 00:30:06.880 In fact, I got a 700-page book I can throw out your head if you tell me otherwise, because it patterns of application development using AI. 00:30:07.119 --> 00:30:19.200 My my book that is coming out uh in its second edition on Addison Wesley soon forward is literally all about how to do that other part, you know, which is the not anthropomorphizing. 00:30:19.279 --> 00:30:26.559 And it that book actually has fairly little about agentic kind of computing on it, because that's a whole nother book. 00:30:26.640 --> 00:30:41.039 Uh, and a lot about how to use discrete AI components in your software for doing things like decisions or classifying or you know, so on and so forth, generating UI or you know, number any number of things. 00:30:41.359 --> 00:30:50.000 I read the first edition a long time ago when I was really first learning my way in the uh the sort of LLM agenc world. 00:30:50.160 --> 00:31:00.319 There's the LLMs that we're embedding into the software that are if they're not user-facing, then they're user, they're things that are helping the user through you know during the runtime. 00:31:00.480 --> 00:31:07.039 And then there is also the agentic development, meaning how are we supporting the agents that are writing the code that are building the software system? 00:31:07.200 --> 00:31:13.839 And I actually think now that I'm saying this out loud, those things are kind of blurring, and maybe it's not useful to look at them on two sides of the line. 00:31:14.160 --> 00:31:25.200 I think that there's people that are rightfully excited about blurring those lines, and I think that it's a cool area of exploration, but I suspect that it's a cul-de-sac. 00:31:25.359 --> 00:31:29.920 I don't like what it does to the software development lifecycle to blur those two things. 00:31:30.240 --> 00:31:30.559 Okay. 00:31:30.880 --> 00:31:32.480 It's too chaotic. 00:31:32.720 --> 00:31:33.920 It's too unpredictable. 00:31:34.240 --> 00:31:34.559 I see. 00:31:34.799 --> 00:31:52.400 If you're gonna get projects done on time and to budget, which that's a given that you're gonna need that going forward, then you need predictability and you you need to be able to put these components together in ways that are provable and repeatable, and like all the good things about good software. 00:31:52.559 --> 00:31:52.720 Right. 00:31:53.039 --> 00:31:56.079 Good software will continue to still need to be good software. 00:31:56.240 --> 00:32:03.839 One of the reasons that people are revolting now, and that there's a lot of arguments in people's workplaces, is that a lot of the experimentation that that goes on looks like bad software. 00:32:04.240 --> 00:32:04.640 Sure, yeah. 00:32:04.799 --> 00:32:05.519 Plenty of examples. 00:32:06.000 --> 00:32:08.319 Yeah, and many of it is bad software. 00:32:08.559 --> 00:32:16.559 Now, the saving grace of all this bad software and you know, bad ideas and experimentation and stuff like that is that it's become so fast to fix things. 00:32:16.799 --> 00:32:28.079 Like I could probably generate six different versions, uh, you know, each better in the time that it would take me to do like the first feature of the first version these days. 00:32:28.319 --> 00:32:31.279 And people are like, why didn't you just do it right the first time? 00:32:31.440 --> 00:32:33.359 Why didn't you think about it and plan it? 00:32:33.440 --> 00:32:38.640 And I'm like, because every time I'm doing it, I'm learning something about the problem. 00:32:38.799 --> 00:32:41.039 Like, do we really need to go back that far? 00:32:41.279 --> 00:32:42.240 Yeah, maybe we do. 00:32:42.400 --> 00:32:44.319 Maybe we need to go back to some of those fundamentals. 00:32:44.480 --> 00:32:52.640 Like we can iterate faster than ever before, and it if software has never been as uh powerful as it is now, yeah, at this moment in time. 00:32:52.880 --> 00:33:03.119 And and I think if you're going down alleys of, oh, let's write the spec correctly and then you know, and then generate it, I'm like, well, yeah, you could do that over and over again, yeah. 00:33:03.200 --> 00:33:06.160 Right, you know, very quickly because the tools are so powerful. 00:33:06.319 --> 00:33:08.799 Congratulations, you're now doing iterative software. 00:33:09.359 --> 00:33:10.640 Yeah, that's right. 00:33:10.880 --> 00:33:12.720 I don't think we're actually that far apart. 00:33:12.880 --> 00:33:20.079 Yeah, it's just that you know, some of us like experiments having wilder ideas and hence have to throw more of them out. 00:33:20.240 --> 00:33:20.559 Yeah. 00:33:20.720 --> 00:33:21.680 But but that's cool. 00:33:21.759 --> 00:33:25.279 That's how you come up with interesting new ideas that didn't exist before, right? 00:33:25.519 --> 00:33:26.319 Yeah, I agree. 00:33:26.480 --> 00:33:28.720 All right, Obi, it's been really great having you here. 00:33:28.799 --> 00:33:30.880 We got another keynote starting in a minute. 00:33:31.039 --> 00:33:31.279 Okay. 00:33:31.440 --> 00:33:34.880 Um, thanks to Buzzbrough for letting us hang out here. 00:33:35.119 --> 00:33:36.160 Thank you, Buzzbrot. 00:33:36.640 --> 00:33:39.839 This is Bubba Fernandez, the new from Austin. 00:33:40.160 --> 00:33:42.480 Yeah, all right, everybody, we'll see you all next week.