you2idea@video:~$ watch NYFGCESmikA [5:15:51]
// transcript — 3218 segments
0:00 - There are decades where nothing happens and weeks where decades happen.
0:05 And we have seen decades of progress happen in the last nine months. If you're not recognizing the
0:14 gravity of the moment, that's the delusion. That's the psychosis. Who would not get delirious if suddenly a genie
0:22 pops out of the bottle and says, "You can have whatever you want.
0:26 Every feature you've ever dreamed of in an operating system, I can deliver them to you,
0:32 most of them in five minutes, a few in 20, and if we really go hog wild, it's
0:36 gonna take me two hours." I want the fastest car breaking the speed
0:40 limits. I want the diver watch that can go down the Mariana Trench. I want the operating system that can install in less than
0:48 60 seconds. One of the things I always loved about race cars was when I
0:51 would stumble out of the car absolutely smashed and barely able to
0:55 hold my head up, and I'd lay down on the garage floor and just think,
0:59 "Holy fuck, I'm alive." The Overton window does not open itself. It opens one nudge at a time by people risking a little.
1:10 A little reputation, a little pushback, a little criticism, or maybe sometimes a
1:13 lot of reputation or a lot of criticism or a lot of pushback. Creating
1:17 life with another human that you love is literally the peak experience of being on the planet.
1:28 - The following is a conversation with David Heinemeier Hansson, also
1:32 known as DHH, creator of Ruby on Rails, CTO of 37signals, creator of the new Omarchy
1:40 Linux operating system, best-selling author, race car driver,
1:44 and one of the most outspoken and prolific programmers in the world. For the past 20-plus years, for him,
1:51 programming meant meticulously handcrafting beautiful Ruby code. But recently, since late 2025, DHH has embraced the AI revolution and has
2:03 openly transformed himself into, once again, one of the most outspoken and prolific agentic engineers, even
2:11 though he hates that term. So let's say practitioners of whatever programming is becoming where AI is
2:19 doing most of the actual programming and the human steers the ship
2:23 with high-level design, vision, and taste. Yes, it's true. DHH is at times controversial, but he's
2:31 always fearless, brilliant, and fun to talk to. This was, once again, an intense, eye-opening, and wild
2:39 rollercoaster ride of a conversation. This is a Lex Fridman podcast. To support it, please check out our sponsors in the description
2:46 where you can also find links to contact me, ask questions, give
2:50 feedback, and so on. And now, dear friends, here's DHH. 13 months ago, we sat down right here to talk about programming,
3:00 and then everything changed. At that time, you were a bit skeptical about the role of AI in the process of
3:08 programming, and then we went through this rapid evolution of agentic engineering. So, simple question to start: How has your view on AI's
3:20 role in programming changed? Are you excited? Are you terrified? Are you
3:24 on an emotional rollercoaster ride? - I am incredibly excited.
3:29 - You've been talking about fun a lot. - There is none of the existential threat. That doesn't exist for me as an emotional
3:38 component. It is only there as an intellectual component, and the
3:41 emotional component for me is 100% pure, unadulterated joy- ... and optimism and amazement that we've made
3:51 computers do this. And I find it so interesting that we talked
3:55 just 13 months ago because it's like we talked in different universes, different eras.
4:03 I love this quote. I think it's Lenin. There are decades where nothing happens and weeks where decades happen.
4:11 And we have seen decades of progress happen in the last nine months.
4:18 I mean, imagine you're there when the Wright brothers take flight.
4:21 Yesterday, the New York Times would write, "It's gonna be 10,000 years before we fly,"
4:26 and then the day after, we're up in the skies, and just a few years after that, there were cross-Atlantic planes. The
4:33 whole world has completely changed. What a blessing to be there in that moment. If you zoom out and look at all of human history, how many
4:41 humans got to live within the same epoch that they were born in? They never saw that
4:49 complete change of the world and of society. And to have been blessed with two of those
4:58 feels just such a privilege. I got to see the internet- ... from pre-internet to post-internet, and then now pre-AI, post-AI.
5:09 What an amazing run. How fortunate. - Yeah, but the thing is, this feels like a thing that
5:16 happened faster than anything else in human history. - Correct.
5:18 - So I would say somewhere around December, maybe late November-
5:22 - November 24. - This is... - That's the exact moment. - You know, when they talk about when one nation invades another in a world
5:30 war or something like this, it-- this is how we talk about AI changing everything.
5:34 Yeah, the... And it really just sh- shifted for many great developers. It shifted to where AI is doing some,
5:43 basic autocomplete, maybe writing 5, 10, 15, 20% of code to writing 80% of code.
5:49 - Or 100. - Or 100, yeah, especially if it's not a public-facing product.
5:55 - To me, that's why even when I look back upon our conversation a year ago, I don't actually have different
6:02 opinions. I have the same opinions. A year ago, I did not like the mode of AI we were offered. It was the autocomplete mode,
6:10 or it was the AI chatbot mode. Now, the chatbot I actually liked, as we talked about. Great tutor right from the get-go, great
6:18 way of looking things up on the internet, not what was gonna replace me chiseling code. But- ... then we get the agents.
6:28 And the agents start out being curiosities for about five minutes, and then they get amazing, and then they get,
6:38 "Oh my God, is this AGI?" And all of that happened since last year,
6:45 even just within the last nine months. We have basically these few phases here. We have AI in the
6:53 pre-agentic era. I was excited about that, but it was not fundamentally rewriting the rules of the game for me. It was
7:01 not completely changing how I worked. I was still chiseling code, I just had a little helper, a little sidekick-
7:08 ... who could bounce ideas off, and I could look up this information online and so forth in a more efficient way. It was just a more efficient way to do what I was
7:17 already doing, and it didn't change the emotional connection I had to the
7:21 computer. Then we get to November 24th, 2025. Opus 4.5, to me, was the dividing line, where suddenly... I didn't even
7:33 try it on the 24th. I think I tried it on the 26th. I give it a couple of tasks,
7:40 and I realize that the quality of the output is uncannily close to
7:47 what I would've written. And I remember just leaning back and thinking,
7:54 "What just happened?" "How did we go from this autocomplete mess that
7:59 I was talking to you about in the summer to this just a few short months later? How did we get both the increase in
8:07 intelligence and then also the increase in usability?" This agent harness question where
8:15 I don't know if Opus 4.5 was that much smarter than Opus 4, which
8:19 is what we had in the summer, but its ability to instrument your computer, to use tools, to check its own work,
8:27 to apply its intelligence in such a way that you could get real meaningful work out of it, was completely different. And I think this is
8:38 then the big change that happens for almost anyone who paid attention and started playing with
8:41 it over the Christmas break. This is what I heard- ... from Shopify and other places with lots of employees who
8:49 suddenly had a breather, suddenly had a couple of weeks to lean back and just look at what
8:53 was going on, gave it a try, and had the same mind-blowing experience that these agents were of
9:02 a different genre than what we had before. And then what we get to is,
9:08 I'm already excited at this point. Like, by December, I'm already revisiting sort of all my priors-
9:15 ... and going like, "Wow, if it can do this, can it also do that?"
9:18 "Oh, yeah, it can." And again, Opus 4.5 now looks like a retarded model.
9:26 And this is the magic of this progress, is you think you've reached something...
9:32 I remember thinking at the moment, "If this is the last model we get, I'll be set. I'll be happy."
9:39 I could live with Opus 4.5 for the next 20 years- ... and you would hear no complaints from me because it was just so incredibly
9:48 amazing to see an agent do all this work in the way that I wanted it
9:52 done. Because it was not just about it being able to solve a
9:55 task, it was also that I could look at its path there and go, "Yep. Yep. Maybe not there,
10:03 but almost," and it would give you two notes. You'll get to where I wanted
10:07 to go in the way I wanted to go there. You could produce code I
10:11 wanted to merge. You could produce code that actually looked
10:15 beautiful if it was written in Ruby. Rust, different question. But
10:21 this ability for the agents to truly become an extension of how I wanted to
10:28 work was very novel. But then we wait just until early spring, and
10:35 suddenly we get sub-agents. We get harnesses that can subdivide the task, and something that would take
10:43 Opus quite a while suddenly took a fifth of the time, a tenth of the time, because it could get chopped up and
10:51 suddenly you got eight sub-agents working for you. But both of those two first phases of the agentic age,
10:59 to me, still felt like I had to drive. I could tell it what I wanted, where to look for it, and steer it a little bit
11:07 when it went off, and then we'll get there, and I would go much faster.
11:11 But I had to be in the driver's seat. I had to tell it what I wanted, and I
11:15 had to be the reviewer, the auditor of what was coming out. And then
11:21 finally now, this summer, with Opus 5, Fable, and Sol, GPT Sol, and
11:29 to a lesser extent, some of the open weight models, we've arrived at a new era where I'm not telling it where we're
11:37 going. I'm telling it the problem I have. I'm telling it the fuzzy,
11:42 vague idea I have. It tells me where we're going. It tells me which path to take, and I will still look at it
11:50 because I'm a curious person and I like computers and I like the outcome of it,
11:55 but I really kinda don't have to. I've become optional in the part that
12:01 produces the code that picks the route. Remember when early GPS systems came out?
12:09 They were amazing compared to looking at a map, but you still wanted to pay
12:12 attention. Is it gonna drive you in the harbor? I remember these newspaper
12:16 articles. "Ah, GPS, they're terrible because people don't pay attention, they drive in the
12:20 harbor." When was the last time GPS drove anyone in the harbor? Like, that just doesn't happen
12:24 anymore. In fact, the cars now just drive themselves, right? And this is where
12:28 we've arrived at now, that I can trust. For the domain I'm working in right now, I can trust it,
12:36 and feel completely confident that it's gonna have my back. It's not gonna do something stupid, and if it does something stupid, it's gonna be
12:44 able to recover. - Mm-hmm. We should mention that there's all kinds of domains that-
12:48 - Correct - ... programmers operate in. There's... I actually don't know, but I think the most common
12:55 domain in development is, like, web dev CRUD. You have a database.
13:00 You have a user-facing interface, and it does something back and forth,
13:04 and it could be internal to just one person, to multiple people,
13:08 to a small number of people, or to the world. And I think for that,
13:13 it's-- I mean, you could really get to the 100% of code written by AI and really if you're a good
13:21 programmer and you have a good intuition about what's happening behind the scenes,
13:25 you can legitimately not look at the code, at least that's, that's
13:29 been my experience. I don't know how much, prior experience with programming you need to have to kinda
13:36 know that things are working correctly behind the scenes, like intuitively
13:40 by observing the ripple effects, the symptoms of the system. But I
13:46 think for that domain, it's close to 100%, and then there's domains like you're
13:52 also operating in, which is writing a Linux distribution. There, maybe there's
13:56 more because you're obsessed with speed and all that kind of
14:00 stuff. There, maybe you need to look at the code a little bit more. Then there's maybe safety
14:04 critical systems all the way down to operating a nuclear power
14:08 plant or a self-driving car. Maybe you need to look at the code more carefully.
14:12 - Yes, but that said, AI is insanely capable at both finding- ... and fixing
14:19 security vulnerabilities. This was the whole blowup about Fable. This model was
14:23 so capable of finding holes that a an attacker could exploit that it was simply not safe to release.
14:30 So the irony here is that when you look at that field, it seems like we've reached levels of intelligence that
14:39 virtually no human can match because many of these security holes are about stringing combo moves together.
14:47 You find one little vulnerability here that by itself might not be the worst thing in the world, but then you
14:53 combine it with four others, and suddenly you have RCE, remote command execution.
15:00 Humans who are able to do that are very rare. They usually work inside state-sponsored organizations or other clandestine operations.
15:11 They're not just out and about finding things. So they've gotten so good at that. The Linux distribution
15:18 is why I've gotten 100% AI pilled. Because I've been working on Omarchy for the last
15:26 three months, this version that just dropped a few days ago called Quatro.
15:31 And almost right from the beginning I'm working on that version,
15:35 the agent acceleration neared 100%, and in the last two months it has been 100%. I have not written-
15:42 - Really? - ... any of the code that's shipped in Quatro by hand. I've reviewed the shape
15:51 of all of it. I've reviewed the individual lines of anything that's critical in the model
15:56 layer of the system, and I've not looked at a bunch of the UI code.
16:02 I have not looked at a bunch of the auxiliary code, and I've not written any of the new functionality entirely by hand. But then
16:13 the web part, actually evolving Basecamp and Hey, our professional products that have lots of users and are relatively large code
16:21 bases, have proven surprisingly tricky to fully accelerate with agents. We just released
16:29 Basecamp 5 not too long ago. That was the first product at 37signals that was really agent accelerated.
16:36 Because we were in this final sprint phase from around February. By then, agents were already good.
16:44 And we had this early surge of, "It's solved. We can just have the
16:49 designers do the programming. They know what features they want. They know what shape
16:53 they want it to take. Let's just-- Let them vibe." And we let them vibe.
16:59 And we ended up with a lot of PRs that individually perhaps could have been
17:05 justified for a hot moment, taken all together, destroyed the architecture of the system.
17:10 And we actually had to clean up manually, mop it up by hand, by human hand, to get back to an architecture that
17:17 felt cohesive and coherent. So we still have a bit of that-- that was February, by the way. Things are quite different now.
17:25 - Well, hold on a second. What's the lesson from that? Is one of the lessons from that that you have to be
17:29 a programmer at this stage to be able to vibe code? - To be able to vibe code on existing substantial
17:38 - I see. - ... code bases, even if they're crud, if you wanna
17:41 retain the element of architecture that got that system to where it was.
17:45 Now, that's also a point where I've stressed many times that when people accuse vibe coders of
17:53 being slob generators, I go right back at them and say, "Have you
17:57 looked at the average programmer's output?" That is some other slob too. If you've looked at behind the scenes of many
18:05 great companies and what their code bases look like after there's
18:08 been 3,000 humans through them, they're awful. Absolutely awful.
18:14 - Can you tell me the intuition you have? I seem to just, stepping back and observing the different apps that we all rely
18:21 on, I don't know, Adobe Photoshop, all this kind of stuff. There seems to be the
18:25 progress on development there has not accelerated. So why is it, what lessons can you draw from Basecamp, like
18:32 well-established, huge user base? Why aren't we seeing, like, super rapid increase in, like,
18:39 new updates, features, all this kind of stuff in these well-established apps?
18:44 - Multiple reasons. I'll start with the first one that's the most critical. As
18:48 soon as you're having human teams work together on something, the bottleneck is rarely
18:55 implementation. It's human bandwidth and communication. When you have a product manager and a couple of designers and a VP above them and a
19:06 CTO above them, and everyone wants to be part of the shaping process because we're all justifying why we're here,
19:14 that's where all the productivity goes to die. The revelation I've had working on
19:22 Omarchy the last three months is that to get that magical 10X,
19:30 100X, in a few rare cases, 1000X productivity boost, you have
19:37 to interact with the agents directly, and you cannot intermediate that
19:44 bandwidth with another human because it's simply too slow. And on the one hand, that's a bit of a bummer. I mean, I like
19:53 humans, and it's great to work together. But it also means we need to temper
19:57 our expectations with what these agents can do if it's humans driving it, and you have three layers of approval and all the
20:05 other machinery of a large corporation. The implementation part is only a small
20:13 segment of it. The other thing I'd say is that most organizations don't know what
20:20 they want. They don't know how to make it better. They're not
20:24 bottlenecked on implementation. They're bottlenecked on ideas. They're bottlenecked on vision. They're bottlenecked on
20:32 taste. And if you don't have those element in excess of your implementational capacity,
20:40 it doesn't help. So you can make a lot of shitty ideas come true, then what?
20:44 Are you gonna ship that? That sounds like, I don't know, something coming out of Microsoft. That's not what we're trying to
20:50 replicate here. That process is not actually... Because we already had
20:54 this. If you step back for a moment and think of the towering organizations we have who have had tens of
21:03 thousands of programmers at their disposal. I mean, I'm picking on Microsoft here. I love Microsoft some of the time, but I'll pick on
21:10 them in this case because they have had endless resources, endless
21:14 programming capacity for decades, right? This is what's showing us that just being able to write a lot of code does not
21:24 produce great compelling software. Now, the other thing is we've had this capacity for about six months. That's not very long in human
21:37 life cycle capacity of internalizing what's going on. And I think it's actually funny that the critique of AI is,
21:45 why isn't it going faster? What are you talking about? We've not had any other form of progress that has moved as fast
21:51 as AI, and you're impatient because in the last three months we haven't
21:54 rewritten the whole world and made it a utopia of software goodness?
21:59 - You're basically saying everybody should be switching to Linux now. One of the arguments is,
22:03 like, we can rewrite all the software that's not available in Linux in Linux.
22:06 Linux has been the love of my life for many years, but one of the reasons I'm still attached to Windows and
22:11 now, uh, Mac is because of video editing, Premiere. - We're gonna fix that. But-
22:17 - So the question is, who's gonna build Premiere and Photoshop for Linux? And it feels like one person can now.
22:27 - 100% one person can. - And so, and I'm allowed to be impatient. In fact, me or anyone else,
22:35 impatience is the first step of doing it yourself, right? Like-
22:40 - Correct. - ... and there's a community of tens of thousands of people that use Adobe
22:44 Premiere or DaVinci, all these non-linear video editors, that
22:48 know the frustrations. Everybody shares them. You can look at Reddit. You can look at
22:52 the forums. Everybody knows it. Uh, I don't know what it is. It may be the
22:55 management. It may be the meetings. All the things you've always discussed is the,
23:00 The bureaucracy that's inside companies that's slowing everything down in the development of features. But it feels like
23:08 there-- if you just let some of the developers loose with a bunch of agents,
23:12 you can fix all that. But again, based on everything you're saying,
23:16 maybe the way to do that is to start over from scratch, maybe through open
23:21 source, or somebody launches a new company that within large companies now it's hard to,
23:29 to really accelerate into the agentic era by letting developers just build.
23:33 - This is the classic innovator's dilemma. These companies have gotten so good,
23:39 so established at the old way, and therefore their entire structure, management layers, processes are tuned for a time that no longer exists.
23:52 But you can't pivot that. These are super tankers. It just doesn't happen. This is why
24:00 we're finally getting this upset in the technology industry.
24:04 For a while, I was very upset about the duopoly between Apple and Google on mobile, because it felt like mobile was the most
24:12 important computing platform that we had, and I could not see
24:16 a path to unseat either Google or Apple sitting on top of that, controlling everything, taking their toll booth money.
24:26 But the game has changed. This is no longer the most important
24:30 platform. Mobile phones are important, but there's also a lot of
24:34 other form factors coming, whether it's glasses or it's earpieces or whatever else have you.
24:41 It's all in play now, but the computing platforms themselves are in play for the first time in
24:48 probably 40 years, if you look at the desktop. Linux has been around since '91.
24:54 It's not taken off or taken over on the desktop. It's taken over everything else. All the devices you have on your desk, your
25:02 fridge, your toaster, everything runs Linux, except for your computer. Now, funnily enough, your Android is actually Linux, but
25:12 wrapped so sufficiently that you can't really recognize it. But now there's an opening, and the opening
25:19 is exactly as you say it. If there's a piece of software you
25:23 depended on that bound you to Windows, it is completely within reach for
25:30 you personally to start rewriting it. And maybe you won't get to
25:34 100% coverage, but this is the old joke about Microsoft Office.
25:38 "I only use 5%." Yeah, well, we all use a different 5%. Well, what if we all just build our own 5%?
25:46 What if I just took the functionality that I need and just did
25:49 that? That is a completely different challenge and one agents are incredibly capable of doing right now,
25:57 today, and I've done it a lot of times over. So one of the amazing things I've found with the latest agent engaged is I have
26:04 become a polyglot programmer, something I was absolutely not before. I was a Ruby
26:13 programmer-- ... first and foremost, and then I dabbled a little bit in Bash--
26:17 ... when I had to. And in the last two months, I've written C++. I've written three applications that have
26:24 shipped in Omarchy Quattro. I wrote a writing app. I was using this app called Typora, which is a very nice app-
26:32 - Great, great app - ... which in itself is based on another app called iA Writer, which was my real
26:37 love on the Mac- ... for a clean, simple Markdown writing environment. That's where I write all my essays.
26:45 And then I moved to Linux, and I couldn't get iA Writer, so I moved over to this other tool called Typora. It was a piece of
26:52 shareware, and it had a lot of features I just didn't need. And about
26:58 six weeks, seven weeks ago, I thought, "Do you know what? I only need 5% of Typora,"
27:02 which is already a basic app. I literally told the agent to get going, that I wanted it written in
27:07 C++ and Qt, because that would fit well with the aesthetic of what I was building with Quattro.
27:14 And in, I think, about 20 minutes, it had the first version. I
27:18 started using it, and it wasn't quite right. Within two days,
27:22 I'd given up Typora, and I wrote and have written all of my essays since that moment in Omawrite.
27:30 - Can I ask for your advice and your vision about something? So what I find
27:34 with agents, you can actually write something like a replacement for
27:37 Typora that is perfectly customized to you, to your needs- - Correct with just one user in mind. And then there's a version of that that is-
27:50 starts with one user in mind but expands to a larger audience. So I've actually
27:53 written a large number of software just for me, and I find it
27:58 easier to do and better to get my situation done. For example, I have a video editor I've written, which is very easy to do just for me.
28:06 - I've never, unlike you, built something that's used by a large number of
28:10 people, and I find that a bit scary and intimidating. So, like, at this, with this agentic age, how do you take a step
28:18 to actually build something that solves your problem, which I think is a
28:22 beautiful way to build, but also expand it to actually useful for other people? Like, is there
28:28 advice you can give, and do you also see the problem of how easy it is to just build a tool for yourself?
28:35 - Yes, but it is simpler than you think. By the time you're done
28:39 building the tool for yourself, you simply tell your agent to put it on GitHub.
28:44 You don't even have to do anything else. It will figure out how to put that repo on
28:47 GitHub. It'll write a nice README. It'll start using GitHub releases so
28:51 that you can track things. It'll actually be a better software
28:55 maintainer than you could ever be- ... because it is far more patient, it is far more diligent in
29:00 doing the drudgery of managing open source software than you are.
29:05 And you can simply get on with the joy of using the software that you built and evolving it as you see fit. Now,
29:13 this gets into the big argument we've been having in open source for a while. Is
29:19 AI contributions good or bad for the maintainer? There's a lot of maintainers right now who are quite
29:26 upset about the fact that they're suddenly getting a huge influx of pull requests, of
29:34 contributions from people who may not be the best programmers or programmers at all-
29:41 ... towards their software. I look at that argument and go, "Are you kidding me?"
29:47 Here is a vein of free contributions that you can take or don't take,
29:57 but you're complaining about the fact that they're there? It sounds straight
30:01 out of- My steak is too juicy and my lobster too buttery. What are you complaining about?
30:07 This is the most amazing thing. This is what we proclamated for so
30:11 long was going to be the magic of open source, that we could
30:15 all contribute to it, but it was, of course, never true. There was only a small
30:19 subset of people who contribute to open source, and that was the very highly skilled
30:23 wizards. So this is a bit of a reformation moment. We had this
30:27 disintermediation between us and the computer, and it was this
30:31 class of clerics and priests called programmers, and suddenly they're being disintermediated here by was
30:39 it the 95 theses by Luther in 1500s, nailing them to the door? And one of those was very clearly about
30:47 the intermediation, that you should have direct access to the higher intelligence.
30:51 - Mm-hmm. But don't you need the wizards to keep a high bar of excellence in the code
30:55 base? - 100% you do, which is also something that's always been true about open
31:00 source, which is the second argument that really grinds me. I've been
31:04 running open source projects for 25 years. I have literally looked
31:07 at the output of thousands, if not tens of thousands of programmers. I feel like I have the statistical basis to
31:15 assert that most programmers, they suck. And I don't mean that in the way it sounds.
31:24 That was why I paused for a hot second here. I mean, they suck in the sense
31:28 that they don't write the code I want to have written. They don't
31:32 prepare their bug reports with all the relevant information. They don't detail their pull requests with the
31:39 why. They don't bother to fill in needed code comments. They don't double-check their work. They don't
31:46 write unit tests. They don't do all of these things it takes to create
31:50 good software. But do you know who'll do all that stuff? Agents,
31:54 if you tell them to. They're very diligent at following your instructions,
31:58 most of the time, sometimes to a fault. But it actually means that if you take
32:02 the median programmer and their pull requests towards an average open source project,
32:09 they're already getting outclassed by agents. I would rather get an agent-written pull request to one of my
32:16 projects than I'd get one written by a human, and it's not just
32:20 because the quality's better. It's also because I feel a lot less bad if I just reject it. You didn't even
32:28 write it, so I can simply look at what your agent wrote on your behalf and
32:31 go, "Eh, don't want it." That's always been true in open source, and in fact, I think this is one of the problems with
32:40 open source maintainers. They have way too much ball of neuroticism and anxiety that
32:48 requires them to look upon any contribution as an obligation on their part to do
32:56 everything to get that to land in the code base. That's not true at all. You can simply say, "No," or even better, you can say, "No, thank you."
33:06 And if you get into the mode of realizing that your project is allowed to exist and evolve according to your
33:13 vision, according to your roadmap, it becomes so much easier in the agentic age to decline the contributions you don't
33:21 want because you don't even inconvenience or hurt the feelings of a human. It's just a clanker, and the clanker won't
33:29 mind. In fact, the producers of clankers, the labs, love when you spend tokens in vain. They get paid just the same.
33:39 So, in my opinion, this is the absolute best time to ever have been an open source software maintainer.
33:46 Not only do we get this wealth of glorious pull requests made by
33:53 agents with all the boxes ticked, we also get to tap into a creativity of people who did not have access to contribute to a project before.
34:02 I've been running the Omarchy project now for a little over a year,
34:07 and in the last three months working on Quattro, I have merged over 1,000 pull
34:11 requests. Quite a lot of those pull requests were written by people who were
34:16 not classical programmers, or were programmers in other domains, not
34:21 Linux operating system or distribution development. They were able to contribute their good ideas because of agents, and I got
34:29 to cherry-pick the very best of that- ... because it was there. Otherwise, those
34:36 ideas would just have lived inside their heads. Again, Is- that is not the purpose of open source, that we tap into the collective
34:43 intelligence and creativity of the whole goddamn planet, and we channel that towards a commons
34:51 where we all benefit from the combined efforts of everyone. Now, there's currently, I think, about 400 unmerged pull requests on Omarchy,
35:01 about double what it was a week ago. So the acceleration is real, but so are the tools.
35:08 I'm not reviewing every pull request anymore. I haven't been reviewing them for
35:12 quite some time now. I have the agents- ... review them for me-
35:17 ... and then they will give me a summary about whether something is ready for the
35:20 decision, the human decision. Should we merge or should we not merge?
35:25 I don't need to look at all the pull requests that are either wrong, duplicated,
35:30 bad. An agent can sort that chaff away from me. I just get to look at the pearls-
35:36 ... the good ideas that are ready to go, the bug fixes that the agent has validated in a VM on my behalf.
35:44 All the drudgery of replicating and managing open source projects is evaporating at lightning speed, and we are left with the,
35:55 with the golden juicy parts, the bone marrow of software development-
35:59 ... deciding what should this thing do and where should it go.
36:03 - Do you think the good ideas, ultimately the kernel of the good ideas
36:06 originated in the human mind, so the individual contributors
36:11 Like, can it be done by agents, or is the PR ultimately requires a human idea?
36:18 Like how to improve Omarchy, all those incoming... Or can it all just be a pool of agents?
36:24 - That was the mode and the way I was thinking about agents in the first agentic age from November 24 to February 28.
36:36 I thought all the ideas would originate with the humans. They would tell
36:39 their agents what to build, and off they went. I don't think that's true anymore at all. I have seen things you people wouldn't
36:47 believe, ideas coming out of models so great that it makes me humble-
36:54 ... as a person who otherwise prides himself on having good ideas.
36:59 The agents are incredibly capable of creative thought, and folks who are still stuck in the analysis that agents are parrots-
37:11 ... just regurgitating the ideas that are already there are delusional about the progress that's been made in
37:19 the last six to nine months. - So some people listening to you right now will say DHH is suffering from the old case
37:25 of AI psychosis. Can you steel man the case that you are in fact experiencing in
37:33 a state of delusion, like One Flew Over the Cuckoo's Nest, and can you argue against it?
37:37 - I'm in a state of delirium. That's what the state I'm in. Because I have been working with computers for 40 years,
37:47 and I've never seen the things that I've seen in just the last two months.
37:54 A whole career, a whole life dedicating to the love of computers as my main
38:02 focus for the working hours and then some, suddenly completely upended, suddenly completely rewritten. Of course I'm delirious.
38:11 Who looking at this reality is not? If you are not delirious or at the very least
38:19 very excited... Actually, I guess that's not even true. If you're not recognizing
38:23 the gravity of the moment, that's the delusion. That's the psychosis. The psychosis is believing that the world is barely different. We
38:32 just have some electronic parrots reciting things to us. Now, in my opinion, the
38:39 reason this argument is actually brought up is because people are not able to see
38:43 the fruits of the progress. That goes to your point earlier.
38:47 Where is the amazing software? Why isn't American GDP running at 12% year over year?
38:55 First of all, give it a minute. It hasn't even been a year, goddammit.
39:01 But second of all, I feel on my own account, I have actually brought the pudding.
39:08 Omarchy Quattro launched on Friday. It's been downloaded by tens of thousands of people. And they like it. They like it a lot.
39:17 - We should mention that going to Perplexity here, Omarchy is a highly
39:21 opinionated Arch Linux-based desktop distribution centered on the Hyprland
39:25 Wayland tiling compositor. It was created by David Heinemeier Hansson, DHH, as a polished
39:32 developer workstation set up with a cohesive visual style and many everyday tools already configured.
39:39 - Let me put it. Omarchy is a beautiful, modern, and opinionated Linux system.
39:45 It's an alternative operating system to macOS and Windows, and it's freaking amazing.
39:50 - And the Quattro thing is you're pushing towards the age of agents.
39:54 - That's right. That's the latest version that we just dropped out, which is also funny.
39:58 The Omarchy project is only a year old and a little bit. I started it last summer in between sessions at the 24 Hours of Le Mans-
40:08 ... where I had watched one too many YouTube videos on Linux Rising and got bitten by that bug. Started working
40:16 on this second iteration of my attempt at creating a better Linux operating system. The first iteration we talked about
40:23 last time was called Omakub, was built on top of an existing system called
40:27 Ubuntu, and it was fine. But the ambition I was able to pour into the new
40:33 version, Omarchy, because I started seven layers deeper down the
40:37 stack, was completely different, but it was still done in a pre-agentic
40:41 world. I wrote all those batch scripts by hand in the beginning,
40:47 and then I got to see the midway point where things changed over.
40:52 I did a couple of versions of Omarchy that had partial agent
40:56 acceleration, and then three months ago, it was full throttle, 100%, everything is written by agents but steered by me.
41:04 And that just ended up being a very different experience and a very different operating system
41:11 because I was suddenly granted a limitless ceiling on my ambition.
41:17 I could look at any feature in Windows, on Mac, on other Linux systems and say, "I
41:25 want that," and the agents would deliver. So everything I wanted was suddenly
41:31 within reach, which meant that I got a little delirious. Who would not get delirious if suddenly a genie pops out of the bottle
41:40 and says, "You can have whatever you want. Every feature you've ever dreamed of in an operating system-
41:47 ... I can deliver them to you, most of them in five minutes, a few in
41:51 20, and if we really go hog wild, it's gonna take me two hours."
41:54 - Mm-hmm. Have you seen Requiem for a Dream, the movie? So the,
41:58 There's a drug-like effect here. I find it personally-- We'll talk about many levels of this, but it is overwhelming
42:07 how much power is placed in our hands so suddenly, and it's very hard to know, it's like with "The Hobbit" or something, what to do
42:15 with it, and I find myself truly overwhelmed with the multitasking,
42:20 On the verge of almost burnt out. Super excited about so many things, but ultimately
42:28 looking back, you want to focus on one thing and really ship
42:31 - I don't have that sense. I don't have the sense of overwhelm. I don't have the
42:35 sense of dread. I don't have the sense of distortion because I have a mission.
42:42 I'm going somewhere, and therefore, I can channel all this new power towards-
42:48 - Yeah, it's wonderful - ... a singular goal and outcome, creating the perfect
42:52 computer. And therefore, all my investigations with AI are focused towards
42:59 that end. Now, I also have a day job. And we also use agents there, and that's also targeted towards sort of
43:06 specific outcomes. But with the Omarchy experience, I got to tap in
43:14 straight in the back of my head and increase the bandwidth between ideas arriving in my
43:22 brain and software emerging on the screen. It was like going from dial-up-
43:28 ... to fiber. I think on your podcast, Elon talked about the actual human bandwidth-
43:35 ... that Neuralink and other projects are trying to accelerate.
43:39 Like, the bandwidth we're communicating over right now is quite low because
43:46 we're limited by our cognition. We're limited by the rate of
43:50 speech. You're not limited by those things when you're working with a
43:53 swarm of agents. Now, let me caveat this by saying, if I'd heard myself talk like this-
44:03 ... nine months ago, I would probably have used the label AI,
44:07 AI psychosis. I think it would be fitting because nine months ago when people were talking like this, they weren't shipping,
44:15 and I think that's the ultimate difference, that there were people who saw this
44:19 early. I was not actually that early. As we've talked about, I
44:22 was rather skeptical, and I was using AI in certain ways, but I was not
44:26 the first person who downloaded Claude Code. I think Boris released that in end of
44:30 February. I don't think I installed Claude Code until September. So there was about six months there where pioneers
44:39 saw these glimmers of what the future was gonna look like. It wasn't there
44:43 yet because the intelligence wasn't there. The harnesses weren't there, but they could
44:47 see the glimmers. I couldn't. Tobi, Tobi Lütke, the CEO of Shopify and now my partner in crime on this
44:54 Omarchy project in part, he's gotten pilled with Omarchy as well.
44:58 He saw these things very early, and he tried to tell me, and I wasn't seeing it. I was not seeing what he was seeing.
45:07 - You were a hater. - Yes. I was still on Earth. And he was... He had already boarded the rocket and was
45:15 heading towards space. I remember actually reading his memo to the company, the AI memo, and thinking, "Ah, seems a little much."
45:24 Like, you're making a big deal out of something that I cannot yet
45:29 feel is a tangible thing. Because where's the output of this? I try using it. It writes code I don't like.
45:36 It wants to interrupt me all the time. Where are you getting this from? Uh,
45:41 so I get it. I totally get it. If you've not seen what the current quality of intelligence can produce,
45:51 of course you're gonna look at someone and go like, "They sound a little nutty," because
45:55 all your experience up until this point would tell you that they are nutty. Because
45:59 if you look at the history of computer programming, for the last 40,
46:03 50 years, people have been promising that regular people could write code just by talking. We're gonna have fourth
46:10 generation languages. Lisp was once upon a time presented, Smalltalk was presented as these environments that would
46:18 let regular people write their own applications, and none of it came
46:22 true in the sense beyond, like, Microsoft Access databases or Excel spreadsheets. That, those are end
46:30 user programming environments, but it's nothing like what we're able to do now.
46:33 And then, I mean, I'm not even talking about end user. I'm talking as me,
46:37 someone who's been programming for 25 years, seeing the output and the acceleration and the quality and then shipping it,
46:45 and I think that's what's gonna change the entire conversation
46:49 quite quickly. The few laggards who are still on the fence about whether this is actually useful, who are still trapped in a meme
46:56 from early '25 about mechanical parrots, they're simply going to be overwhelmed by the evidence that's about to
47:04 flood over them. - So I hope to see, projects like Omarchy, several of them that, like, really show agent first.
47:15 - It's coming. All of it is coming. - In the spaces, in the silly spaces that I already mentioned, like video editing, all
47:21 these kinds of apps that are like Typora- - Yes. - ... this kind of stuff.
47:26 - It's gonna come, and I've already seen the early inklings of this.
47:30 One of the really neat features about Omarchy is that it has
47:36 a far more robust plugin system that allows you to extend your
47:40 operating system and rewrite it really in terms of its user interface and its panel and its features
47:47 by creating your own plugins that can be cloned off of tools we ship in the box. For
47:55 example, say you want a different calendar. Omarchy ships with a calendar. You
47:59 click the little clock. It pops up a calendar, and it doesn't
48:03 have iCal support, for example. It doesn't consume your appointments.
48:07 There's about 17 implementations of that already, and in three days, we had 330 plugins on the Omarchy plugin marketplace. I have
48:18 never seen growth like that with any project I've ever been involved in. I've never seen participation that broadly.
48:26 I've never seen that many people be able to create software that's
48:30 meaningful for them- Also be usable for others so quickly. All of
48:36 it is driven by the fact that Omarchy ships a set of skills that tells any agent you bring to it how to
48:43 create extensions to the operating system, and therefore affording you the vision of the true malleable...
48:53 I was about to say agentic. I hate that fucking word. And the reason in part I hate that word is, first of all, it's become marketing
49:00 slob speak- ... at this point. Like, it's just slapped onto everything.
49:04 I wish we had a different word that just meant AI doing stuff.
49:08 - Vibe coding feels wrong, too. - It does, because vibe coding to me smells exactly like script
49:13 kiddies did in the early 2000s. People applying PHP scripts they just downloaded offline they don't
49:19 understand anything of. And while, I mean, that's true in the sense that much of the vibe coding,
49:25 including my own vibe coded projects, and as I just mentioned, I have several.
49:29 I've written things in C++. I've written things in Rust. Rust. I hate Rust with a passion. Rust to me
49:39 is like pouring acid in my eyes when I have to look at the code.
49:42 - Wait, wait, wait, wait. Why do, why do you hate Rust? - It is, in my opinion, the ugliest programming language- ... that has been invented-
49:47 ... in probably the last 40 years. - So it's the opposite of Rails and Ruby.
49:50 - It's the opposite. But- - You've written it - ... I've written applications in Rust because the output
49:57 of Rust is actually amazing. The memory safety, the fact that it's a system
50:01 language, the fact that it's highly efficient and so forth is
50:05 incredible. So Rust, to me, can both be the most repugnant programming language ever devised for
50:13 human consumption and a wonderful platform for agentic engineering.
50:19 These two truths can coexist easily in my head. - Do you think that there can be a point at which we can just call agentic engineering
50:28 programming? Can we just change what programming means? 'Cause who's actually going to be programming the old-school way
50:35 in a few months? - I don't think we should reuse the term because for me, programming
50:40 implies an understanding of certain primitives, loops, conditions, variables. All the constructs of programming languages
50:52 are what constitutes programming itself, not the creation of programs.
50:55 Because you could say in the pre-agentic era, well, your CEO is programming. He's hiring a bunch of programmers, he's telling them what to do, and
51:03 out comes software that he can sell. Eh, I don't think most people would call that CEO a programmer, and I don't think we should call
51:09 the vibe coder a programmer either. And vibe coding, if we define it here, is you tell an agent to build software for
51:17 you. You do not look at the implementation. That, to me, is what separates vibe
51:21 coding from programming or, let's say, agent-accelerated development.
51:26 - But don't you think, so to push back a little bit, don't you think a programmer, old-school programmer, you, doing
51:33 agentic engineering is a different kind of agentic engineering than
51:37 a non-programmer doing agentic engineering? What I mean is, if you know
51:42 what a for loop is, if you know what function programming is, if you know some of the
51:46 basic good principles of software engineering, the way you do
51:50 natural language-based agentic engineering will be different and more
51:53 systematic, and this, the scale and the variety of things you can actually build is much bigger than a vibe coder.
52:02 - I'm just gonna extend the pro case here a little bit just for the sake of the argument.
52:05 - Right. - I actually think for a while it was to my deficit to know as
52:14 much as I know about programming because I was instructing the agents to do things as I
52:22 prescribed them to do, and they were very good at that. And in the
52:25 first agentic moment, let's not call it an era, it was something
52:29 that lasted three months. Moment. In the first agentic moment, that felt
52:35 very productive. I could get more productivity out of telling the agents what to
52:39 do in the way I would've done it, and I used all of my experience as a programmer to do just that.
52:47 Then I was a little late on the next moment, and the next moment allowed me, allowed
52:54 anyone, to describe outcomes- ... to describe problems to the agents
53:00 and get better solutions than if you had a programmer prescribe the path.
53:05 - See, okay, so we're having fun arguing here. Well so are you suggesting that there's problems for which
53:12 programmers are worse than non-programmers at agentic engineering?
53:16 - 100%. And the reason I say that, let's define it here. The reason I say that is there's a lot of programmers who are not
53:24 very good product managers. Software is product management. What should it do? Who should it do it
53:32 for? How should it do it? How should it look? What are our priorities? What
53:36 do we start with first? What does version one include? All of those skills
53:42 are not easily or equally distributed across all programmers.
53:46 And in the agentic era where you are letting an agent, an AI, do the implementation, these are the skills you
53:54 need. And I happen to sit in both camps, and this is why I love this
53:58 moment. I still get to live a little bit in the old world where I look at the full
54:02 implementation because there are domains and areas where I feel
54:06 that still brings value, and then I also live in the future.
54:10 And as I said, when I created Omawrite, I've not looked at a
54:14 single line of that C++. I actually made it a point to myself that I was gonna treat it as an experiment. This is 100% a black box.
54:22 I will treat it as though I was any other user who had opinions about how their digital typewriter should work, and
54:29 therefore felt like it was a fair experiment as if a writer of any other sort who had
54:36 opinions about how software should be done and how it should work. As you mentioned,
54:40 video editors who know exactly where they don't like what Adobe
54:44 Premiere's doing and how it could be better, but don't have the capacities to
54:48 improve it- ... how that could be. - The pushback, though, I think what programmers do well, good programmers do
54:54 well, is systematic design, like think in terms of systems- - Correct
54:59 - ... and rigor. - Yes. - And I think still, and probably for a long time to come,
55:04 even natural language-based control of agentic systems will require a certain kind of rigor. Not overly
55:11 structured rigor, but really explain very specifically the goals of the system, how to
55:19 do the verification, what kind of security testing. You wanna do all
55:23 those kinds of things. Although then to push back on that, the
55:27 system will... should be smart longer term. - I would've said the same thing six months ago.
55:30 - Yeah, so this has changed. - And I think the same thing would've been true six months ago. What I have found since,
55:35 especially even just the last few weeks of the agentically accelerated development of Mach-E, is that
55:43 more often than not, I have the humility to recognize that the agent knows best.
55:49 - You can legitimately say, "Make sure it's secure." - Correct. And it would know more about what that entails. Now, you're laughing-
55:59 - What a crazy world - ... but it's true. And in fact- - Oh my God
56:02 - ... a good parallel to this is the AGENTS MD/CLAUDE MD files. There was a hot moment
56:09 where it was all the rage about micro- optimizing that. Oh, you tell your agent this-
56:14 ... you tell your agent that, and you ended up with this huge file of instructions
56:18 for your agent. One of the things that Boris, working on Claude Code, shared in an interview recently about Opus 5 was
56:27 that the system prompt that they ship for Opus 5, and presumably also Fable, shrunk by 80%
56:35 because the agent not only needed far less human instruction, it was actually being damaged
56:43 by overly prescriptive humans. And any programmer who's had a pointy-haired boss knows exactly what that is like.
56:51 When the boss walks into the room, doesn't know anything, starts telling you how to program, how to code. What do you
56:59 do? You sulk. You write shittier code if you're mandated to do things that are against your better judgment.
57:06 Why would an agent not be the same? - Actually, to push back against myself, I do find the actual skill I'm currently
57:11 developing, which I along those lines is probably what everybody's trying to
57:15 develop, is not to overspecify things, is to get out of the system's way.
57:20 - Of course. - Is to specify enough to provide a kinda high-level vision and guidance.
57:25 - It's even better because you don't even need that. The fundamental insight of modern software development came from the agile
57:32 software development movement. In the late '90s, early 2000s, a group of
57:38 smart, honest, and brave people came together and said, "The way we've been trying to
57:42 do software development for the last 40 to 50 years has not worked, will not work. We've been trying to
57:49 specify upfront what software should look like. We thought that we could ask humans what they wanted, write it all
57:57 down, apply our systems thinking and our rigor to a spec sheet, hand that spec sheet over to a group of programmers
58:05 who would then implement just that, and everyone would be happy." Well, guess what?
58:09 No one was happy. Because no one knows what they want until they
58:12 receive it. You don't know what a program should do until you
58:16 play with it. So in the agentic age, you should resist the temptation to be overly specific upfront. Be as
58:23 vague as you can to manifest something, then interact with the something.
58:29 The way you arrive at good software is you write a little bit of software,
58:33 and then you try to use it. It is in the process of using software that you
58:37 discover what you really want. It's where you discover what's important and what's not important. And
58:43 agents are incredible at allowing you to bumble into this experience of realizing what you want because
58:50 you don't know what you want. - Mm-hmm. And I've actually done something similar to what you suggested, which is
58:55 implement different designs and totally different implementations, and then
59:01 I create HTML pages for myself, PHP pages, where I vote on what I like more and so on.
59:06 And it does this iterative thing, and it's very nice and fun. The interaction with
59:10 the agentic system, if you create a nice interface for yourself, it can be super fun,
59:14 and you're focusing on the design part and the most fun parts of the design part.
59:18 - Correct. And this is what humans are really good at. - Yeah, taste.
59:21 - Differential evaluation. You give me three options, I pick one of them. Now, humans actually fall off a cliff if
59:28 you give them 22 options. That's the paradox of choice. But you give them three options,
59:32 they very quickly within a split second will tell you what they like.
59:36 There are all these ways that exploit the fact that humans will make these snap judgments that are actually pre-intellectual,
59:44 right? Like, they're coming from the gut, and then they arrive up in
59:48 the, in the brain, and the brain tries to rationalize why the
59:51 gut said what it said. And if you're willing to let go of some of that intellectual pre-processing
60:00 rationalization and simply let your gut drive, the agentic age is a revelation.
60:06 - So this is a fascinating thing to ask you because you have sort of
60:10 famously for a long time talked about, like, detailed chiseling of beautiful
60:17 Rails Ruby code, and now you have switched in a matter of months to not doing that. So, like, what is
60:24 your current setup look like? 'Cause you're, I assume, are still
60:28 looking for the beauty, for the chiseling, for the crafting, but you're just
60:32 crafting in a different medium. - And at different levels of abstraction. When I'm working in Ruby code, and we have a lot
60:39 of Ruby code because it's in our entire business, and I'm asking agents to make changes
60:43 to that code, I still sweat the details. What I'm coming to realize is that the economic payoff
60:55 of that sweat is diminishing rapidly. The reason why I, for 25 years, was sweating every line of
61:03 code so judiciously was because I knew the payoff of keeping an architecture
61:10 coherent and malleable was software that could change and evolve-
61:16 quickly with a small team, and not exorbitant cost, and not introducing a bunch of bugs when you change one thing over the other. That was
61:25 the driving economic argument for why you should write beautiful code, 'cause beautiful code is easier to understand, it is simpler, it
61:32 is more malleable. That was premised on humans doing the modifications.
61:40 I think it is an open question to which degree this still matters. Now,
61:46 it does matter, and the reason I say that, at least for the moment, is that
61:50 tokens are still scarce. At this moment in time, we are all token limited.
61:58 Well, not all, but anyone who doesn't have endless budgets are token
62:02 limited. So therefore, there is great payoff to writing systems that agents have an easier time dealing with
62:10 and evolving without having to relearn the entire context. Just like humans,
62:14 if they can make iterations and changes to the code base without wrecking the architecture, they can make the next
62:22 change just as cheaply as the last one. And this is the classic ball of mud where you end up with a system that's a ball of mud-
62:31 ... because it's just put together in a way where nothing is
62:35 connected and it's just a real mess, right? I've seen that with agents, and I've seen
62:39 it in our own code bases where, again, the first PR is, like,
62:43 mediocre of quality, and then if you add another PR on top of that, and then five more
62:47 down the line, it's not very good, right? So there's still a payoff to that,
62:51 but that payoff is premised on our current moment. I am now able
62:55 to... And this is, by the way, where the AI psychosis really comes in, when you
62:59 try to extrapolate what nine months from now it's gonna look like.
63:02 What two years from now it's gonna look like. But as an intellectual experiment,
63:07 I think the first computer I used was a Commodore 64. - Me too. Cool.
63:11 - It had one megahertz CPU and 64K of memory. Everyone who wrote software for that machine internalized
63:20 certain constraints. Well, all the constraints. That machine was
63:24 nothing but constraints. If you put that programmer in a time machine and teleported him or her to
63:31 2026 and gave them a modern computer today and said, "Now write me a piece of software," they'd be lost for a little bit
63:39 because all their techniques and all their heuristics would simply be wrong. Or not wrong, because efficient software
63:46 is still beautiful. They would be out of date with the value that they could create.
63:51 The amount of optimization you have to apply for a one megahertz computer
63:55 to produce a video game is just very different from what you have to apply today.
63:59 - Just imagine that guy that was programming back then on a Commodore. Just re-
64:03 remember all those, like what programming felt like and what it meant, highly constrained resources-
64:10 ... how slow everything is. - Beautiful in many ways. Beautiful.
64:12 - It's chiseling. But- - It- - ... imagine that person coming now.
64:17 - It's romantic too, though. Let's also imagine that. This is one of the things, whenever
64:21 we bemoan modern car culture, for example, right? And then we think like,
64:27 "Oh, wasn't it better when we all had the horse-drawn carriages and so on?" No, it wasn't.
64:31 Do you know what New York smelled like when we used horses for transportation?
64:35 It was an open sewer. They just shat everywhere. - Yeah, but have you played Red Dead Redemption? You know how cool those cowboys were-
64:42 - That is the way- - ... on the horse? - ... to experience the nostalgic past, when you don't get the all
64:47 the furry impulses and senses. - So basically programming, old school programming by hand is
64:54 kinda like the cowboy culture that we romanticize and make movies about.
64:57 - It's already happening, the romanticization of handwritten code.
65:01 And I have some of it because it was a very romantic era. I'm grateful to have been alive for 20 years of
65:12 economically valuable handwritten code. That was a, that was a good time.
65:16 - Where handwritten beautiful code was also valuable. - Correct. Because handwritten beautiful code exists
65:24 as much today as it did yesterday. There are people who willfully write new video games for the Commodore 64,
65:32 or for the second Mega Drive, or for other vintage consoles. In
65:36 fact, one of my favorite video games of late is ModRetro's. Do you know Palmer Luckey's-
65:43 ... side geschäft, which is recreating old consoles. So he's recreated the Game Boy.
65:50 - Nice. - It's amazing. Look it up. It's really cool. But what's even
65:55 cooler than the fact that he recreated this hardware and the original
65:58 chunkiness, but brought it up to date with sort of modern screens and so on, now they just put out the Nintendo 64 too. Incredible.
66:06 Super cool. Wild attention to detail, input lag, and all the other things.
66:12 But what's also very cool, that one, I have to wave. That one
66:16 right there, and hover over it, 'cause then you get to see what I'm talking about.
66:19 That is the Chromatic- ... Tetris re-implementation. - Wow. - I played a lot of Tetris on the original Game Boy.
66:26 Probably one of my top three favorite games of all time. And they rewrote it with one change. If you hit up, the brick slams.
66:39 And that speeds up the game by about 400%. It is an incredible game, and it's a new Game Boy game.
66:47 I think the Game Boy came out in- '89 or '88. It's a very old machine. The fact that there's still people writing new old
66:55 games is awesome, and the reason they do it in part, now here
66:59 they're doing it because they want Tetris, but there are people who do it just for the love of the constraints,
67:03 for the love of the romantic notion of writing that. Just like there are people still riding around on horses, not to get
67:11 places fast, but because they like the mode of transportation where it's a literal biological
67:19 being below you propelling you forward. - And also, we should say that that beautiful code is great training
67:25 data. So in some sense, the beauty that we've built over a few decades-
67:30 - Yes - ... continues. - We gave birth to this moment. What a privilege, and I feel that very
67:39 personally because I... Almost all of the code I've ever written is public
67:43 code. The vast majority of my career has been dedicated to writing open source code,
67:47 and therefore, some of that code, some of those beautiful lines
67:51 are in the training set. And in fact, I've heard people do this when they write Ruby code. They ask the agent, "Write
67:59 it like DHH would." And they're pleased with the result, and that does warm my heart a little bit, that
68:05 I helped give birth to this moment in my tiny little small part. And anyone else
68:09 who contributed open source code or any code at all that the agents have had access to
68:13 over the last, well, entire time programming has been around as a discipline, is now part of
68:21 these new agent overlords. - So you don't have a little bit of sadness that writing beautiful
68:26 code by hand is now less and less useful? - Am I sad that if I had to get a job tomorrow,
68:33 it's unlikely that I could apply my skills in such a way that I would be
68:36 paid to write these manual lines of code? No. I did it for 25 years. Like, I wanna see something new here,
68:45 especially if we're all gonna live forever, which I hope we don't. Maybe we'll get to that.
68:49 But, like, I've done that. That's enough. That's fine. I don't feel any more
68:53 nostalgic about that than I do about the fact that I don't have to
68:57 spend my time in the field with a hoe or at the assembly line.
69:01 - Well, hold on a second. You are one of the best programmers
69:05 in the world at writing handcrafted beautiful code, appreciators
69:09 of beauty with great taste. - Correct. - That has been replaced.
69:13 - I say humbly. - Yeah. That has been replaced. Like, you had now...
69:18 You're just fast evolving, and you're able to... One of your
69:22 really great qualities is you're able to change your mind and evolve very
69:26 quickly. But, like, you had to kill that other person. - I am grateful for every moment that brought me to where I am right now, and
69:34 therefore, I have also tried, and maybe this did take some practice. I
69:39 will concede that. Maybe I wasn't always like this, but stoic philosophy, amor fati, loving your fate,
69:48 is an incredibly liberating way to live. And you could attack that and say, "Well, that's a point of privilege."
69:57 True, but also I had the same directional point when I didn't have all this, quote unquote, "privilege."
70:05 So I think there's a way to live and interact with the world
70:09 where you accept the things you can change and the things you cannot change
70:13 and fall in love with all of it. Fall in love with the fact that the world today has jumped
70:19 two decades forward into technology industry from where it was last year.
70:24 - So can we talk about the general anxiety that programmers feel,
70:28 going through the same transformation? They've maybe gone to university
70:36 majored in computer science, dreamed of being programmers and building stuff,
70:40 high salary, and now everything's changing, and there's deep
70:44 anxiety about, "What do I do when... with my life?" Can you empathize with that, and what the hell are they supposed to do?
70:51 - Hugely. Because just because I have this disposition, I'm keenly aware that that's not evenly distributed, that
70:59 not everyone sits with this position and this optimism for the future.
71:03 But I do think I wanna separate things a little bit. I think you're gonna have a hard time coping with the new
71:11 reality if the only thing you loved about programming was the mechanical bits of
71:19 putting the right logical constructs together to produce something other people
71:25 told you to produce. Because that mechanical process is under threat.
71:33 If you are, as you just mentioned, excited about building things,
71:38 I don't think you're under threat at all. In fact, I think there's a great argument
71:42 for us needing far more builders than what we have now, and if you look at
71:49 the employment stats, it's fuzzy. It's not clear what's gonna happen at all. Some stats actually show an
71:56 increase in openings because the advent of AI is so dramatically lowering the price of
72:04 programs that people want a lot more programs. This is the classic,
72:09 the Jevons paradox, that says when the price of something goes down, there's gonna be more
72:13 demand for it. And this is the example of the ATMs too. When the ATM
72:18 originally came, a lot of bank tellers were very afraid for their job because suddenly there was a machine that could dispense money from
72:26 people's account. Well, what the ATMs did was lower the price of a branch, and suddenly banks could afford to open a lot
72:34 more branches, and we ended up with more bank tellers than we did before. Now, none of this
72:38 is guaranteed. There are also moments where things do change. The
72:42 vast majority of people worked in the fields up until the late 1800s, and then suddenly we got mechanized-
72:51 harvesting tools, and we did not need human labor with a hoe and
72:55 an ox out there. We had mechanical tools to do that for us. In that moment, are there people who were threatened that their
73:04 job and their livelihood was at stake? Yes. Was that a reasonable thing to
73:08 feel threatened about? Yes. The Luddites smashing the weaving machines-
73:13 ... in England had the same thing. And I think maybe that's the better parallel
73:17 than people working the fields because these were actually highly skilled professionals
73:21 doing a job that they liked on rather favorable conditions, not slaving outside in the sun all day.
73:28 But where we live now, would we still like for clothing to be this
73:34 heavily constrained resource? Like, what if this T-shirt was
73:38 $400? I mean, okay, then I'd have two. Well, my wardrobe probably looks like I have two. But
73:46 I think the general progress of mankind depends on productivity
73:54 improvements. And what are productivity improvements? This is very important because in the abstract,
73:58 I think everyone says, "Yay, productivity. That's good." What did you think, this was all just vibes? No. Productivity means fewer
74:06 people to do the same number or the same job. Now, the amount of job you want done may increase, and therefore you get more people,
74:14 but it also may not. There may be pockets where a company just needs a certain set of
74:21 fixed tasks done, and suddenly they can do them with a 10th the number
74:25 of people. This, by the way, could... is tragic and difficult in the moment for the individual being laid off. It's also amazing
74:33 for the economy at large. Suddenly you've freed up these resources who can
74:37 now go do more productive things. This is how the whole economy evolves and improves and how we get growth,
74:44 and growth is good. I think this is the other thing to make a
74:48 strong defense for, that growth in general is a good thing. We do not want to wind the
74:55 clock back to 1920 or 1950. We should be excited about the fact that we've discovered new
75:03 technology that allows us to do vastly more. And while some of the tasks that may be taken over by AI were tasks
75:12 that humans were excited to do, and they're just no longer economically
75:15 viable, a lot of the tasks were just drudgery. This is certainly true
75:19 of my programming. There were moments of programming I thought were amazing. These were the moments that produced that state of
75:27 flow we talked about last time, and they were ecstatic. How often did they
75:34 happen out of the course of a year? If you took your average
75:37 programmer, how much time out of 2,000 hours at the job did that person spend in a flow state?
75:47 100? 200? 25? I think all those answers could be plausible for someone.
75:54 And taking that drudgery, handing it over to machines, that is the history of civilization.
76:03 - Yeah, I mean, the hours spent debugging. So programming has so many painful,
76:09 painful hours in it. And now they're gone. For the... I mean, for the most part,
76:14 it's just fun. - A lot of it is gone in a lot of domains, and the rest of it may be gone very soon.
76:21 So that's where some of the romanticism also comes in ... because pain in the moment
76:28 is difficult. It's hard. It's annoying. Pain in the rear view mirror is accomplishment, proud,
76:35 learning, steps forward, all the things, right? So we encompass all of this as part of the human experience, that we hate
76:43 pain in the moment, but we cherish it when it's five minutes past and we don't remember the hardship. We
76:50 just remember the progress. - So certainly at the societal level, at the macroeconomic
76:55 level, growth is exciting. But the individual level, there's going to be a lot of
76:59 anxiety potentially a lot of suffering. So by way of advice, if you're like a young DHH or young developer
77:07 programmer now, what would you advise they do? - Don't try to anticipate anything. You will literally go crazy.
77:15 Because even the smartest brains in the business cannot anticipate what two model hops from here is
77:22 going to look like. It's an absolute waste of time, and you will develop an AI psychosis trying to deduce what two
77:30 years from now is gonna look like. Focus on right now, and right now is the most
77:34 incredible time to be into computers. You can make them do the most amazing things if you lean in. If you
77:45 maybe for a hot moment force yourself to learn where the state of the art is,
77:53 I double dog dare you not to get excited about what's possible.
77:57 - And would you recommend building publicly? - I think it's optional actually to build publicly. I think it's
78:03 nice to do because then you get to be part of a community, and there's some
78:07 camaraderie in open source. Not some camaraderie. There's an
78:11 enormous amount of camaraderie. Open source is an incredible source of community and camaraderie,
78:18 and it's also a great way to stem some of that personal existential dread. If you're amongst other humans, you can
78:26 get infected with good forms of mind viruses, exciting forms of mind viruses, the ones that tell you that it's all gonna be okay,
78:34 and the future looks bright, and if we build a bunch of things together, we can do things none of us ever
78:41 dreamed we could do. I was just talking to Ryan Hughes, who's been
78:45 my main partner on Omarchy almost since day one, and we were literally talking about yesterday how when
78:53 We started on this journey together. Our ambitions were very
78:56 modest. We're just like, "Can we get this distro to not crash at
79:00 3:00 AM because of an Arch package that just got pushed out outside of our control? That would be amazing. Then it'd be great." Now
79:08 we're leaning back and thinking, "How can we take over the world?
79:12 How can we make Omarchy go to Mars? How can we do everything we've ever dreamed of?"
79:21 And that level of ambition inflation is partly because things have gotten better with agents, but hugely also because we've
79:29 surrounded ourselves with others excited about the same journey.
79:33 And if you're just sitting in your little isolated cave worrying about the future, yeah, you're gonna go a little nuts.
79:42 I think if we learned anything during COVID is that people will go nuts sitting in their little cave
79:49 inside by themselves worrying about what's gonna happen. That is
79:53 indistinguishable from depression, ruminating endlessly about things you can't
80:00 control. I'm sorry to channel some Jensen Huang here, but that's just loser talk.
80:07 You don't have to be a loser. You can choose to lean in and-
80:13 ... win. And win is very broadly defined as learning more, making more, contributing more, being part of more.
80:23 And at the end of the day, what are your choices? You don't have a choice, mate.
80:29 The future's coming whether you like it or not, so you might as well choose to be
80:33 excited about it. - But I mean, we should emphasize the fact that planning is nearly
80:37 impossible with AI. I mean, most of us plan our life a little bit. When we're younger, you have, like, hopes and dreams. There's a
80:45 reason you go to college. There's a plan underlying that. - I agree with that.
80:48 - But there- - I'll concede that. - ... it is kinda- - This is different.
80:51 - ... nuts that it's impossible to really plan. Because, like, you really don't know,
80:58 if there will even be a Claude Code, a Codex, and Cursor six months from now. Maybe it'll be you'll be talking to WhatsApp. Maybe
81:07 it'll be all open call type. Maybe it'll just all be voice. Maybe Omarchy will take over the world and it'll be voice. Maybe there won't be operating systems
81:14 anymore. You'll just... There'll be this glowing orb that just you just talk
81:18 to because if natural language is enough to accomplish all tasks in life, like, what is life about? Like, what are the major
81:26 tasks you need to accomplish in life? If you can just do that with voice, and
81:30 it writes all the programs for you, and all is systematized, and what,
81:34 what are the jobs needed then? Maybe it's more about the service
81:38 industry. Maybe there won't be programmers, like agentic engineers at all. And so
81:44 yes, your advice is wise to focus on the moment, but that doesn't mean
81:48 that it's just a crazy rollercoaster ride, 'cause we don't know
81:52 what this December or what this January will look like. - We don't know, and therefore, we should choose to have faith that it's
81:59 gonna pan out. That is my best advice towards fighting anxiety, is that you have a choice, to a large degree,
82:11 of how to lean. Do you wanna lean into PDoom? PDoom is gonna happen whether you lean into it or not.
82:21 You don't have control. - But there's also actual daily schedule of how much learning are you doing, how much
82:27 building are you doing, how much uh, diversification you're doing.
82:31 And for example, I traveled across rural China recently, and I did
82:35 it quite deliberately because you're talking to people in China, in,
82:41 in rural anywhere, where time moves much slower, where there's no discussions
82:46 about agentic engineering. And the ti- and I started by going on the, on the Great Wall of China.
82:54 That's timeless. That spans centuries. And, you know, there was, like, this kind of tension of FOMO, like maybe I'm missing out on something.
83:02 But then I very deliberately wanted to find the timelessness of human existence, really plug into the fact that,
83:10 okay, it always feels like everything's changing uh, throughout human history but, you know, the universals are still true,
83:18 of what matters in life. And I just wanted to make su- keep connected to that.
83:22 - You're not gonna miss anything. This is the part that actually
83:26 grinded my gears in the early phases of this transition. Everyone was so up in arms that if you
83:34 didn't... It's loops now. Oh, no, no, we're done with loops. It's graphs
83:37 now. Oh, no, no, we're done with that. It's, it harnesses this, right? They're, they're constantly
83:41 churning through the frontier, which in one way is actually very exciting. This is what happens when a new field opens up. It's like a
83:49 new portal, and it's all spilling out. That's exciting. But it
83:53 also means that if I had just been backpacking for the last year,
83:58 hadn't touched a computer, hadn't witnessed this agentic moment, and I just showed up
84:04 yesterday, do you know what? I would've been caught up in two weeks.
84:08 There's not any accumulation, which is in some ways a great relief. If you missed the past year, you can catch
84:16 up to the frontier in two weeks. If you're a programmer who was out hiking the Himalayas for a year and
84:23 you come back, you can catch up in two weeks. And this is actually the great
84:27 credit to the progress we're experiencing. We have so many experiments running simultaneously right now that are
84:34 constantly and ruthlessly sorting what works, what doesn't work. You don't have to remember or even be part of that
84:42 entire journey. You can just show up for the results. - But the key step there, when you come back from the backpacking journey,
84:49 is to be willing to become a totally new, different human because-
84:55 Things are changing so fast. - That's right. - I mean, literally, if you left for the backpacking journey- ... in October last
85:00 year and came back in April or May- - You might as well have been in the cryo chamber for 100 years
85:06 - ... it's like, "What, what do you mean? We're not programming anymore?" This, it's a it's a
85:10 huge transformation. - It's startling, and I actually... I mean,
85:14 I'm putting some of this on slightly for effect. I also recognize that it's okay to grieve for a hot moment.
85:22 That's not my disposition. It's not how I do it, but I accept that that's a
85:29 time-tested mechanism for coping with a sense of loss. And it's okay to have a sense of loss that the world
85:38 that you knew and maybe you were very fond of is no longer the same.
85:43 - I maybe I should say out loud and explicitly that I do have that, the Native
85:47 American with a tear rolling down my... Like, there's a sadness.
85:51 It's a goodbye. It's a goodbye to the old world of programming.
85:55 - That's, it's so fascinating to me. - I've spent so many years-
85:58 - But it all amounted to this moment. That is, to me, what makes the
86:02 entire journey so meaningful. AI did not arrive from the sky. It was not alien technology that
86:09 had blasted through the universe and suddenly showed up on our shores.
86:13 AI has been around for as long as computer science has been around.
86:17 In the '50s, they thought the problem was gonna be cracked in about a
86:21 decade, the smartest people at the time. It took a little longer,
86:25 but the reason it took a little longer was in part because we took
86:29 some blind turns on the road to AI and neural networks and thought that there were other symbolic representations that were
86:36 gonna carry the day, and they didn't. And that probably set us back
86:40 about 15 years. So some of it was also we had to, Go through the gaming revolution, 3D games. If we had
86:47 not had Quake, if we had not had Duke Nukem, if we had not had
86:51 Unreal Tournament, we'd never have gotten AI because we would never have gotten the GPUs,
86:55 and therefore we would never have been able to get AI in the shape it is now. So
87:01 everything we did up until this moment was required to get there.
87:05 You were part of that. Holy... What a blessing. What a blessing to have contributed in your small part, either by
87:13 training or by playing video games. Literally, you could just choose to
87:17 think of it that way. All those times you spent in, for my case, in the '90s just playing video games, they were-
87:23 - You were contributing. - Yeah, I was contributing to the AI revolution right there and then moment.
87:27 - To the progress of human civilization. No, for sure. Civilization
87:30 progresses with the death of the old and the birth of the new, and it
87:34 continues in this way, but it's still the grieving process is a part of it, I think.
87:39 - Take a moment. It's okay. Take a... And also, you don't even
87:43 have to accept that it's all dead. Like, if you are still very attached to the hand
87:50 chiseling, which if you'd asked me a year ago, I would have predicted that I
87:55 would've been more attached. I'm surprised- - I'm kinda surprised
87:59 - ... a little that I've not been more attached. But the reason I've not been more attached,
88:03 I think, is that I found something more fun. And if it hadn't been like that, if AI had
88:11 just replaced the thing I loved and gave me something I hated, I'd
88:15 probably be a little bitter. - I mean, this really feels... It's hilarious and awesome to watch and inspiring to watch.
88:19 It's like Picasso all of a sudden like, getting access to to image generation and getting super excited and switching in a
88:30 matter of months. That's literally what happened. So, like, it's an artist that appreciated beautiful
88:34 code, and all of a sudden you're talking about having fun- - I think Picasso is actually a good parallel because if you look at his early
88:42 work where he learned his craft, it was painting realistic paintings, right? Like, he was training to
88:50 be a master of the old ways of depicting reality the best way we knew how,
88:56 and then along comes all these other forms of expression. Cubism comes along.
89:01 And all these forms of abstract expression, he's not decrying that he's not doing Renaissance
89:09 paintings anymore, that he's not just painting the perfect depiction of an
89:13 apple. He's reimagining the apple to be a freaking square, and he's getting excited about that. And I feel that
89:22 transition, that we've gone from this one mode of expression, and we've suddenly opened the gates where
89:31 the, like, the color spectrum is so much wider. I'm seeing more. And it's
89:37 perhaps because the way I became a programmer. I did not become
89:41 a programmer because a deep love of if statements. I became a programmer because I wanted programs.
89:49 I wanted things to exist that did not exist. And at first, I had to learn how to program to make that happen.
89:56 Then I happened to fall in love with programming as a craft, and then I spent two
90:00 decades really diving deep on that. But now I'm back. I'm back to having an idea and
90:09 being impatient beyond belief to see it exist in the world, and AI has just shrunk that down to almost nothing. So it's an acceleration
90:20 of a rediscovery. Now, I understand that that's not true for everyone. There are plenty of
90:27 programmers who just loved programming right from the beginning, and they were attracted
90:31 to the logical constructs and so forth. I mean, I've been there for two decades, so I get
90:35 it. But I also think that you're able to discover new sides of yourself-
90:41 ... if you let it in. Again, I keep coming back to this almost regret minimization framework, to borrow Jeff Bezos's terms here.
90:52 Are you gonna look at yourself two years from now and think, "Oh, I spent
90:56 my time well being pissy about the present, about the fact that my industry changed"? Or
91:04 Would I two years from now look back upon this moment and be proud
91:08 of myself for leaning in, for maybe having a moment of grief? Again, there's room for all of it. And then going, "I'm gonna
91:16 learn how the world spins now, and I'm gonna make not just the best of it, I'm
91:20 gonna make more of it." - I mean, it is true, a lot of programmers, a lot of friends of mine,
91:26 me, obviously you, are having fun. - I have had more fun with computers in the last three months than at any time
91:37 previously. - Quick bathroom break, and then I gotta ask you about
91:42 how you actually interact with the systems now. - Yes. - I gotta ask you about how has your programming setup changed? So keyboard,
91:54 voice, what's the IDE? - It's crazy to think about now, but yeah, I used TextMate for almost 20 years.
92:02 I used TextMate starting in 2005, I think. I helped get the first version out, and then I just
92:09 wasn't interested. I wasn't in the market for an alternative. And
92:13 it wasn't until the switch to Linux that I was forced out of my habitat. And
92:21 now with the switch to, what are we calling it? Agentic engineering? Oh, I fucking hate that term. We gotta
92:28 come up with something that sounds as plain as programming, but
92:32 encapsulates the fact that it's with agents. But- - I still think it should be called programming at this point.
92:36 - ... All right, let's just call it programming. Programming with agents requires a different tool set. It really does.
92:42 And the main change here is that you're going from single thread programming in your head to parallel processing.
92:54 When I was writing code, chiseling it by hand in TextMate or even Neovim,
92:59 Not that long ago, I would just focus on one problem at the time, and
93:03 I would methodically work my way through it, and that was actually the portal to flow.
93:07 The portal to flow was deep, immersion into a single problem, see it through to the end.
93:15 That's not how it works with agents, in part because the agents are at once both too fast and too slow. They don't give you an immediate
93:26 reply on something that you asked them to do that's the same as typing on a keyboard. So you have to let the agent cook-
93:35 ... for a bit, and therefore you realize, well, if I just sit around waiting for them, first of all, that doesn't feel productive.
93:43 Even if the agent, just one of them, can be highly productive, it does not feel productive.
93:47 It does not feel good. It feels actually like you're a little bit useless.
93:50 And maybe I had a moment, when the first agentic moment was there and we, I was running mostly
93:56 one agent at a time where I felt like, "I don't know about this."
94:00 But you can solve a lot of hard problems by simply throwing more resources at it. This is the whole scaling law of
94:08 AI of itself, right? That if you parallelize these things and you're not running one agent, but you're running a handful,
94:16 you can feel like you're in a flow state because you're constantly doing
94:23 programming work in the sense that you're making decisions, and you're helping either
94:26 unblock an agent because it has a question about which direction to take, or
94:30 you're ready for a new task. And to do that, you need a different
94:34 setup. I started first doing it in tmux and just having separate panes and having separate splits. Basically,
94:42 a terminal with tabs is a good way to think about it. You open a bunch of tabs. I think
94:46 most humans know exactly how that works. They don't work in just one tab.
94:50 - So still sticking to the terminal, CLI. - Absolutely. - So you're not using Claude Code app or the-
94:56 - No. - ... Codex app or the- - I love the fact that this agent revolution was kicked
95:01 off in the terminal because I was already a huge fan of TUIs, terminal
95:05 user interfaces, and the terminal in general. That feels like a,
95:10 a really nice place to be. It's a beautiful place to be. The modern terminal
95:14 is just a good-looking place to work. So I like that. And then
95:22 this fact of having multiple agents, especially once it's not
95:26 just multiple agents running on your own machine, but you start running multiple
95:29 machines. Now tmux alone is not enough to keep track of it, and that's why as of late I've switched to this thing
95:39 called Herdr. And Herdr is essentially tmux plus agent notifications. So whenever your agent is done and needs something for you, it goes
95:48 ding, a little bell telling you it's ready for its human, uh... Is it master or servant? I'm not quite sure always.
95:58 But it is ready for a decision, and it also keeps track of these is it in working mode or not. So I have this Herdr set up. I have
96:06 multiple Herdr setups actually running on individual machines.
96:10 I went on this crazy phase just about a month ago realizing that doing this work on a single machine is
96:18 not fast enough. It's like I've discovered multi-core programming, but I only have two
96:22 cores. I'm like, "What if I had 16 cores? What if I had 32 cores? What if I had
96:26 64 cores?" So I instantly went out and I bought these amazing
96:31 KVMs called GL.iNet Comets. And what they do is it's this little box. You plug in HDMI, you
96:39 plug in USB and connect it to the computer. It's like a KVM. So a KVM is a remote way of controlling a computer, but
96:47 what's special about this is just how easy it was. You connect this thing in,
96:52 you go to a webpage, log in once, set one password. Now this thing can hop on your tailnet.
96:59 This has been the other revolution of the last year for me, is discovering
97:06 these WireGuard networks. Uh, Tailscale is essentially turning all the computers you have into a local
97:14 network wherever you are. Like right now on my phone, I have direct access to all the
97:19 computers in my Malibu office. I also have access to all my computers in my Copenhagen office.
97:26 And I can treat them as though I sat right next to them without having to
97:30 punch holes in a firewall or set up complicated VPNs. And what this does is it just decreases the friction it
97:38 takes to get new compute online. So as soon as I discovered this, I looked at my closet, and I realized I had a bunch
97:46 of mini PCs from prior experiments, and I just said, "What if I just connected all of
97:50 them?" And I just connected four of the computers- ... in a closet. They all had their little Comet, and
97:57 suddenly I could run agents on more computers at the same time,
98:01 and I could control them all with Herdr. And it did get to a point where I maxed out my own processing power.
98:09 That I think at about... What do we want to put it? About four to five machines
98:15 running, I don't know, three agents. I have about 16 threads. That's what I can run.
98:21 And the faster the agents run, of course, the fewer threads I can run, but at the current pace, I can run
98:25 about 16 threads- - Mm-hmm, at full acceleration. And that's part of why I've gone from just being excited about
98:35 the agent moment to being delirious. Because I think as you say, some of it is a little assaulting.
98:42 Again, we're on a dial-up bandwidth-wise with our computer. I mean,
98:48 it's actually hilarious. So you think of I'm sitting here a year ago, right? I'm
98:52 chiseling my code. Yeah. - I'm writing my lines. And over the course of, like, an hour, I will have written
98:59 one beautiful controller, one beautiful model. Like, that's one file.
99:03 And I've really worked on it, right? So maybe there's 60 lines left. So
99:07 maybe my bandwidth at this moment is, like, 30 lines an hour. I think that's even high. Maybe it's 20 lines an hour. Now I'm
99:15 running 16 threads. I'm producing at some times hundreds of lines of hour or hundreds of lines of code
99:22 per hour. Now, let me pause myself there for a hot moment. I hate that metric, right? Like, lines of code is a stupid metric
99:31 in general to measure these things, and I think it's right from
99:35 some in the programming community to ridicule the AI psychosis when
99:38 everyone talks about the number of lines of code they're writing, and then you ask them,
99:42 "What did you build?" And they're like, "Well, I..." And then no good answer comes out, right?
99:47 - It's just a nice shorthand to- - It is a nice shorthand. It is a representation or encapsulation
99:52 of how much output that's coming out. Now, whether the output is good or bad-
99:57 ... is not a referendum on that. But I was producing and producing so much more now,
100:01 right? And therefore I'm able to keep all these threads going.
100:05 So that's the setup. It's still Neovim, but at this point, I'm not writing a lot of code, so I'm using Neovim as a project browser
100:14 and then as a way to kick off Lazy Git to see the change log for what's there. And even that, I'd say
100:23 if GitHub was a little faster at showing you your pull request, the web is
100:26 probably actually a nicer place to do that. I don't know. There's also this other tool
100:30 I've been playing with a bit called Hunk, which just produces
100:34 Diffs in a really nice way. So you could look at that too. But I find that when I look
100:38 at Hunk, I only see the change set, and when I'm reviewing output from an agent, I often want to see the surrounding context. Oh, yeah, so it
100:46 changed this file, but actually, what do we have in this other file that wasn't touched but maybe should've
100:50 been touched? So that's why I still like Neovim as a way of doing
100:54 it. But it's all happening, by the way, of course, in Amachi. So it's all happening on Linux. And this was the other major
101:01 breakthrough with agents. Agents love the Unix philosophy. It loves individual tools
101:09 that it can invoke through the command line, and there is no operating system
101:13 on Earth of the majors, I'm counting three here, Mac, Windows,
101:19 Linux, that works as well with that mechanism as Linux. Everything in Linux is either a config
101:27 file or a CLI tool. Now, that was its main drawback five minutes ago. This was the reason people didn't like
101:35 Linux. It's like it's all config files and CLI tools. What great irony that the universe has played it upon us that now the
101:45 drawbacks of Linux five minutes ago are now its major selling points.
101:51 This was one of the things I thought I was stuck for a weekend with a Mac
101:57 four months ago. I thought I had a computer at the place I was going that was a Linux machine, so I
102:02 didn't bring my laptop, and I found out when I arrived I only had a Mac Mini.
102:08 So I was gonna make the best of a bad situation here and, Set up my Mac in some of the ways I've been thinking about with,
102:17 With Linux. And you can actually do a lot now. Homebrew has gotten
102:21 really quite good. Homebrew is the missing package manager for the Mac,
102:25 and no one has done more to move the Mac forward in terms of ease of setup. But still something as simple as
102:33 Raycast, I don't know if you've used that. That's the- There's no config file that you can just access. You have to go into the
102:41 GUI, export a file, then take that file, I don't know, in your freaking backpack on a USB key.
102:48 And then you can import it somewhere else. You cannot automate the entire setup of your
102:52 machine. You can't automate at all the configuration of Mac's default key bindings.
102:59 That has to be a manual process where you're clicking with a mouse like a
103:03 caveman to set up your machine. - I mean, there's ways around it.
103:07 - Not good ones. I looked. I tried hard - So here's my, been my journey. Obviously I'm a Linux
103:14 person, but because of Adobe Premiere, Adobe products, I'm also a Windows person,
103:18 so oftentimes I would use WSL, Windows Subsystems for Linux.
103:23 So Linux inside Windows, which is... in the agentic era is not a good kind of
103:27 Linux, 'cause it's just- - It's a sandbox - ... it's a sandbox, and you want the Linux to be unleashed to be able to do everything.
103:33 - Correct. - And so I have now woken up to things like GrayCat, and have
103:38 you know, I'm a keyboard person. Where's the config file? What, can I,
103:42 can I basically do everything where an agent can set everything up for me and save the configuration so I can
103:50 replicate it across systems and everything's automated? And so I had to
103:54 ask a lot of those questions, and a lot of them are missing.
103:56 - They don't have good answers. - But oftentimes you could actually just build the app yourself.
104:01 - But there are hacks. - Yeah, they're hacks. - There are hacks.
104:04 You don't have to live like this, Lex. There's a better... Actually, this is a great
104:07 moment. - Until you have a video editor. - So this is a good moment.
104:12 - Yep. - Um, I heard it was your birthday. So I brought a little gift.
104:18 - No. - We are going to get you onto the agentic operating system whether you
104:23 like it or not. - This is awesome. - And I asked- ... my friends at Dell whether maybe they had a machine that,
104:29 um, I could get you for your birthday. - Oh, no. - So here's the machine I use- ... which is-
104:36 ... well, the same one I use, but it's now yours. It's a- ... Dell XPS 14-
104:42 ... already set up with Omarchy 4- ... ready to be configured for you. And
104:49 once you realize that all your... By the way, we should almost time
104:53 this. You can be up and running. You have to answer, like, five questions in about one minute. You
104:59 could literally do it live if you wanted to. - Oh, Omarchy, beautiful, modern opinion Linux by DHH. Yeah, let's do it.
105:05 - Yeah, fill it out while I talk. So- ... exactly what you realized, that pain on the Mac
105:12 was the exposure I got for about a weekend, and I tried really hard, and I
105:15 found some decent workarounds, the main one being that Homebrew
105:19 now has tabs for GUI apps. Because it's hilarious to think upon the recent past where people
105:27 would go to websites, like, "I need Zoom." They would enter zoom.com, find the download link, download
105:35 a DMG- ... double-click on that, see a weird manual interaction telling them to drag an icon
105:42 from one place of the screen to the other, and that was how you installed apps.
105:48 And you'd just do that over and over again until you had all the apps that you wanted,
105:52 and if you were setting up a new machine and you had, like, 12 apps, you'd go to
105:56 12 different websites to download them like a total caveman.
106:01 Homebrew has solved a lot of that, but it has not solved all of it, and has solved
106:05 nothing in terms of the- - Sorry to interrupt, but look how fast this is going.
106:11 - Oh, it's gonna be done in about 40 seconds. And then you're gonna be inside of Omarchy.
106:17 - Super is the Windows or Command key on your keyboard. So which one is the Super key?
106:21 - Yep, that's the Windows key. Oh, actually, no, no, you have a special edition.
106:25 There are only about six of these computers in the world. - Oh, there's a Super key.
106:29 - It says Super, and look at where the Copilot key is on the other end of the
106:32 keyboard. That's the Omarchy logo. - Nice. Okay, cool. - This was Spencer Bull at Dell, who I've been working with on a
106:40 lot of this stuff. They've been incredible. - They, for a moment, let go of the XPS brand.
106:46 - Yes, they did, last year. - And I was like, "This is dumb. This is... XPS is awesome."
106:51 - It was not just dumb that it got rid of the XPS branding. They also got
106:55 rid of the Escape key and put in capacitive touch buttons at the
106:59 top. The mistake that Apple realized like three years ago was
107:03 terrible, and everyone hated them for them. But now they have an incredible team, and this
107:08 was actually my first Dell laptop for myself. Because I've been using Dell on our servers. Dell was what powered our
107:16 cloud exit. We've been running Dell since the mid-2000s, I think. But I never ran their laptops, because compared to a
107:23 MacBook, they just weren't up to the task for me. The XPS' from 2016, or from 2026, with those
107:32 Intel Panther Lake chips, are the first Dells that I would personally actually use. And it's funny because when I got mad at Apple
107:41 originally, I thought, like, "Do you know what? I should just get a Dell," because we already used
107:45 them and whatever, and I got a Dell, and I was like, "Man, I can't do it. I can't do it.
107:49 It's not good enough." And then lo and behold, this year they just fixed everything.
107:56 They fixed the battery life. They fixed the performance. They fixed the compatibility. They fixed the weight. They fixed the
108:03 screen. That's a tandem OLED screen- - Yeah, that's incredible.
108:05 - ... which means it's actually better tech than what you'll get in a freaking
108:10 MacBook. It's lighter than a MacBook, and the chip is finally competitive because Intel spent
108:17 a good five years developing the 18A process node. It took a long time, but now it's here, and
108:24 it's freaking incredible. We finally have competition to Apple M chips-
108:30 ... that can go as long as on battery and can have competitive performance.
108:34 - And there's a beautiful default theme to this thing, and then what is it? Super
108:38 Space opens up. - Actually, you can just push the Omarchy key. It opens up the menu. So now you control
108:43 everything from there. - Oh, same thing as the Super. Okay. - Yeah.
108:45 - Cool. - Same thing as Super Space. So this has been a huge change for me, that we finally
108:53 have Linux machines that you don't have to make excuses for.
108:57 Because I don't mind making excuses because excuses are really just another
109:01 word for trade-offs. I love that Framework 13 I think I talked about last time. I still
109:06 like them. They still make good computers. But I think for a lot of people, they
109:12 didn't want a DIY experience. They wanted something that just out of the
109:16 box felt like it was on par in build quality and whatever with Macs, and now we finally have it. So happy birthday.
109:27 - Thank you so much. An Omarchy Dell laptop from DHH is the best birthday
109:33 present. This is awesome. - Check the screensaver, by the way.
109:36 - Yeah, it's kind of amazing. I'm seeing it. - On that tandem OLED screen-
109:41 - It's gorgeous. - ... it is absolutely mind-bogglingly awesome.
109:46 - Look at that. There's like a retro look to the whole thing.
109:50 - Totally. - It's like Fiverr pop- - It's like running a bulletin board system in- ... the '90s, which is
109:55 exactly what I did when I was 14 years old. So for me, it's a
109:59 complete throwback and an awesome aesthetic, combining this sense of modern
110:06 rising culture in Unix with- ... this throwback of the '90s.
110:10 - Yeah, like a retro look for an agent first OS. - Exactly. - It's crazy.
110:13 - Which again, pairs... This was why I was so excited that- ... all these agent harnesses arrived as TUI's, arrived as CLIs-
110:21 ... has arrived as a terminal because suddenly, a whole new generation of both programmers and regular users
110:29 discovered the glory of the terminal. The terminal was actually one of the most,
110:34 impressive and productive user interfaces we can have. Now, we've built user interfaces since then that were easier to learn.
110:43 The mouse is a wonderful invention for letting someone who doesn't know a lot about computers just click around and find things out, but
110:51 the terminal is for people who want to learn their systems, who want to learn the keyboard commands.
110:57 - So you think the TUI is gonna stick around? - 100%. Now, it's not gonna be the only thing. Of course it's not, because
111:03 these labs now have trillion-dollar valuations, so they need to be able to
111:07 reach billions of people. And I don't think we're gonna teach billions of people the intricacies of the TUI.
111:13 So we're gonna have other tools, and we already have other tools, which is, by
111:17 the way, one of the sweet things about Linux, the ChatGPT Codex tool.
111:22 - Yeah, I saw that. - Literally within two hours of that announcement dropping, I had it wrapped
111:27 up and ready to go in Omakub. Already now, if you go to... You can
111:31 actually hit the Omakub key- ... and then just write ChatGPT. It should pop up as, like, install.
111:38 And then just hit return. - Sudo password for Lex. - Yeah, whatever you put in. And then it's just gonna install it, and it's gonna pop it up, and you have
111:47 ChatGPT. - Nice. - And that's how a lot of the apps are, and all the agentic harnesses are built in that
111:52 way. Um, Tailscale, as I talked about, if you wanna hop on that, is that way. Uh, Dropbox.
111:57 And the reason all these things are in there is because I use them.
112:00 It's because I built this system for me, and I thought, first and foremost, how do I
112:04 build my dream computer just for me? And if I get my dream computer, I am assured there are other people who are gonna have the same dreams
112:12 as I do. - Now, you're obsessed about getting the installation to be under one
112:17 minute, and a bunch of people asked you online why. So Why? Why were you so obsessed to getting the installation to
112:25 be less than 60 seconds? - First, let me quote Mitchell Hashimoto, who created
112:31 Ghostty terminal that's awesome and also available on Omakub and was also
112:35 the founder of HashiCorp, who had this great post on X not too long ago that had the killer line, "The
112:43 pursuit of excellence deserves no explanation." Wanting to make something as good and as fast and as beautiful as possible has
112:56 no need for justification. We should want beautiful, fast systems
113:02 that are a delight to use. Now, the funny thing with the install time is it didn't
113:06 actually start with a one-minute mission. I was being modest when it got
113:10 going. I thought if I could set up my entire system in 15 minutes,
113:15 it would already be such a dramatic improvement over what came before it that I
113:19 would be perfectly happy. And it came off the fact that the benchmarking
113:23 had been skewed so out of whack. You talked about the Commodore 64 being
113:27 your first computer, right? When you unwrap that Commodore 64 and you hit the on button,
113:32 in I think about less than one second, the basic interpreter
113:36 was ready to accept your command. It was literally almost instant.
113:40 There was virtually no boot time. Do you know how long it took me
113:44 to set up a new Mac that I had bought for actually the one app that I still haven't vibe coded my way into on Linux, which is Adobe Lightroom?
113:53 Brand-new computer, out of the box. I set it up. There's a software update. Before I was ready to
114:01 install Adobe Lightroom that I had it fully up to date, it took 42 minutes.
114:08 That's grotesque. That's absurd. That's an insult to everyone who likes computers, that you get a new computer and
114:16 before you can use it, you have to spend 42 minutes doing updates
114:20 on something that was already installed on it out of the box from the factory.
114:24 That is just preposterous, and it's crazy that that's not even the
114:28 worst thing. I also got a PC, um, because all these Linux machines, they come as PCs, right?
114:34 - Right. - And they have Windows pre-installed. So I got one. This was just
114:39 three weeks ago. Brand-new machine, Intel Panther Lake, the whole shebang.
114:44 An hour and 35 minutes from unwrapping the thing until it's ready to use. How long did you take
114:51 before you think you were in Omakub? - I mean, it's less than a minute for sure.
114:55 - A minute. We can build computers like this. We have the technology. We've had the technology
115:01 since like 1981 when the Commodore 64 first came out. The fact that we now have a computer that's a trillion times faster
115:08 and you have to spend 42 minutes or an hour and a half setting it up before it's ready to
115:12 use is so infuriating to me that it became a passion. And it's funny
115:19 Just while we were having a little break here, I was talking to Ryan Hughes, my co-conspirator
115:22 here. We now have, we think, a line where we can create these special turbo images that are pre-made for specific computers
115:30 because it relies on some knowledge of the hardware, like the Dell XPS. We're gonna make a Dell XPS turbo
115:37 image that can install in about 12 seconds. A full Linux system
115:44 installed in about 12 seconds. - That's incredible. - That obsession is what gives me this great satisfaction with using computers,
115:52 that we do not accept things the way they are. And this is why this
115:56 agent-accelerated reality that I now inhabit is so invigorating.
116:01 Because I can look at these kinds of constraints and go, do you know what? Um,
116:06 I could have stopped when we were at five minutes. That would already have been fast enough.
116:10 This was the argument on X, right? Why do you keep going? Five minutes is fast enough. Okay, maybe
116:16 for you, but for me, 12 seconds is a lot more fun. And I always try to think these first principles.
116:24 Okay, in that machine, actually hit the Omarchy, key again, and then do speed. You'll see a
116:32 list come up, and one of them is, like, disk speed. - Speed test, network speed, disk speed.
116:37 - Disk speed. Do do that one, and then run it. So this is testing the performance of the built-in NVMe drive, and what are you
116:44 getting? - Seven GB a second. - So seven gigabytes a second. That's how fast that drive is. Why can we
116:51 not install a Linux distribution in one second? The Omarchy distribution is 5.8 gigabytes.
116:58 If we can run at seven gig... Now, of course, you can't actually do that
117:02 because the USB key is a lot slower than that, but you should be able to
117:05 maximize the underlying physics and protocols that we're dealing with, and that's your goal. We gotta
117:13 get all the way back down to that. - Yeah. I love that you're doing a Linux distribution. 'Cause I mean, you could be
117:21 do... You could be building anything. But the fact that you're building a Linux
117:25 distribution means you're rethinking about what a computer is, 'cause a
117:28 operating system is the fundamental, like, the foundation of how
117:32 ... you interact with a computer. - Yes. And it's been so stale.
117:34 - Yeah, it has. - The macOS we have today is scarcely different from the macOS we had 10 years ago. In fact, it's worse
117:42 in many ways because Apple continues to chokehold the control on users. They don't want them to
117:50 install software that they haven't already said good- ... to go for. They don't want users to reconfigure their
117:57 hotkeys without doing it manually. They didn't want to change the
118:01 500-millisecond animation switching workspaces. It is an infuriatingly locked-down computer. Now,
118:09 to Apple's credit, it's a pretty good computer for being locked down,
118:13 but I don't want a locked-down computer. I wanna own my computer. Better
118:17 yet, I wanna mutate my computer, and this is where the agentic
118:21 age needs a new operating system. When you can vibe code whatever app comes to your mind, you should be able
118:28 to vibe code your operating system. You should be able to change anything, how it looks,
118:32 how it works, what panels are there, what are not there, and that
118:36 requires Linux. As simple as that. There are no... None of the other two
118:40 operating systems can deliver this. - But can you speak to the, I think somebody asked you, like, "Who the hell
118:45 are you to think you could do better than Ubuntu?" For example.
118:49 Or, or like, "Why are you doing it when Ubuntu is already pretty good?" And your answer
118:53 was, "I thought I could do better." So can you speak to, this was on X, can you speak to, like, where
119:01 the craziness comes from, thinking that you as one person or maybe with Ryan, you could start building something that's
119:09 going to be better than Ubuntu or Arch or any of this? - Well, I'll say that it didn't start quite as
119:16 grandiose. When I got going with my Linux journey, I actually just
119:20 started with Ubuntu, and I just built on top of it. Here's a
119:23 system that already does a lot of the things I wanted to do, and at least it's
119:27 Linux. So can I just build on top of it and make it more modern, make it actually look good,
119:31 and have the programs I wanna use? I got, like, some version of that,
119:35 and then I discovered that there's seven layers deeper you could go, and as, the
119:39 closer you get to the metal, the closer you get to the kernel, the closer you get to the
119:43 individual packages and so forth, the more freedom you discover. So as I discovered more and more freedom, I got more and more
119:50 ambitious, and I realized, oh, surprise, surprise, I have strong opinions about how computers should work, how they
119:58 should look, what programs should be installed, and whether they should take 42 minutes to install or
120:05 45 seconds. That's the current world record, by the way. So if anyone installs Omarchy Quattro after this and can beat 45 seconds,
120:13 take a screenshot, send it to me, and you'll be the new holder of the world record.
120:17 - When, how long ago did you cross the 60-second mark? - It's funny because it's like
120:23 that the old limit on, what was it? It was a four-minute mile or something-
120:28 - Yeah. Uh-huh - ... like that in 1952- ... that people thought was impossible for humans to beat. Then one guy beats it,
120:34 and literally within the next few months, three more guys do, right?
120:37 So we had this goal. I originally had the goal it was two minutes.
120:40 And then we beat two minutes, and I'm like, "Well, if we beat two minutes, why can't we beat
120:44 one minute?" And I remember thinking like, "That'd be crazy. You can't
120:47 install a modern operating system in one minute when they're all..."
120:52 If the other ones are taking 42 minutes, right? Like Apple, full of a lot of smart people, like, it took 42 minutes, so that's gotta be the bar.
121:00 No, no, it's not the bar because there's no speed limit. Like, none of the things
121:04 you take for granted are fixed pillars of reality in almost any case. Until you get down to the seven gigabytes
121:12 per second that drive is able to transfer, you've not reached the limit yet,
121:16 so you should just keep pushing. And of course, then it becomes a game in itself and a satisfaction itself.
121:22 Every second I could shave off was an excitement. And when we broke the one-minute
121:26 barrier, which was really only broken, I think, three weeks ago or something like that, maybe even two weeks ago, I just got crazy.
121:33 And then I got a swarm of agents to run all these auto research
121:37 loops of trying all these different theories on like, oh, if you change this thing, if you
121:41 change that thing, what if you cut this out? What if you do things in parallel? What if while you
121:45 were typing in your username, we start preloading packages into memory system, we can fire them in faster. And it becomes a
121:52 fun game. And I think that's how you should think about product development. It should be fun.
121:55 You should, it should be frivolous. It should be trying to overshoot
121:59 the target and making it so much better than what it quote unquote needs to be.
122:04 - Especially now because you can move so quickly and you can try so many different approaches because of
122:08 the agents. - Yes, that just makes it more satisfying. But I think this was always true. If you look at the
122:13 greatest Mercedes-Benz of all time, I think it was called the W126.
122:18 I saw this wonderful documentary on it. It was steered by this guy who for years
122:22 had worked in Mercedes' department for safety. And suddenly he gets put in charge for the whole S-Class project, and he
122:30 just puts in all of it. This is where you got the auto-retending seat
122:34 belts and washers on the headlights and everything. This car was so overbuilt for the market.
122:42 It was unbelievable. But you look at that and go like, "Well, I want that."
122:46 In just the same way that you look at a mechanical watch and go, "It could go down to 100 meters?"
122:54 Or I think Rolex has one that can go down to 4,000 meters, the,
122:58 the deep ocean, the deep diver ocean ... or whatever it's called.
123:01 No one outside of three people in the world, whoever buy that watch, will
123:05 ever put it to the test. But don't you wanna be the kind of person who
123:11 is a patron of people who wanna push the human race that much
123:15 forward? I want the fastest car breaking the speed limits. I want the
123:19 diver watch that can go down the Mariana Trench. I want the operating system that can install in less than 60 seconds.
123:27 - So from the individual level, I admire those people. I think
123:30 everybody should strive for that. But also the consequences of that we should
123:34 mention, like the pursuit of excellence, even when the metrics seem to be frivolous,
123:40 the maybe unexpected consequences of that is other kinds of excellences, discoveries, innovations, and all that kind of stuff that has nothing to do with the
123:48 speed. - 100%. - It's discovery. It's nice to have the metric that you're
123:52 chasing be the catalyst, the engine that drives innovation, because you'll just discover. It's the same reason you go to the moon, you go to
124:00 Mars. It's like, why? Who cares? Well, actually, the real thing is you'll discover a lot
124:07 about chemical propulsion. You'll discover a lot about how humans can live for a long time
124:14 in space. Maybe you'll discover whole new ways of doing solar energy in
124:18 space or data centers in space or whatever the hell. - That's right.
124:22 - All kinds of... Maybe you'll meet-- - That's right - ...aliens, but you never know. But setting a goal
124:30 and chasing it, especially when it's numerical, is really nice.
124:33 - And setting a grand goal is even better. When you refuse to just do what's practical and you strive for what's slightly beyond
124:44 plausible, that's when the magic happens. And we should be going all in on that. This is one of the reasons... And in fact, to some degree,
124:53 this specific goal was motivated by seeing others do this. I mean,
124:57 Elon is the singular individual in this conversation. The guy is so out there in his goals.
125:06 They seem so preposterous at first glance. I've certainly considered them preposterous on many glances. And then you realize that
125:18 they come true sometimes. Maybe not always on schedule, and so
125:21 what? So what? So we didn't get the self-driving car on schedule.
125:26 I gave Elon a lot of shit for that back in '17. But now it's here,
125:31 and who would have gotten us here except for the person who thought it was
125:35 possible in '17, who is so impatient with reality that the sheer force is just gonna propel that reality to the place he
125:47 wants it to be. I think that's deeply inspiring, and we need these kinds of,
125:52 goalposts to be moved. We need someone to run the four-minute
125:56 mile, so the rest of us realize, "Well, okay, but I guess I can run it
126:00 quicker." That's what I try to strive for, and I really strive for with Omarchy
126:04 because it's just more fun to live that way, to have high aspirations and ambitions. And I
126:09 will say I've actually changed my mind on that a little bit. I spent quite a lot of time
126:15 speaking the other book that there are a lot of people who don't actually need those
126:21 ambitions. And I also-- That's also true. We don't want 8 billion Elons running around the world. Would be rather crazy, I think.
126:29 But I've come to realize that for entrepreneurs, it's healthy to have suitable chunky goals. The idea of just running the Italian
126:41 restaurant can't just be that. It should be maybe I just run the one Italian restaurant, but I make the best damn pizza
126:49 possible. I make the most delicious Parmesan. We have to strive for excellence to muster the motivation to keep going.
127:00 - And yeah, to dare to think big like you could build a Linux-
127:05 - Yes - ... distribution. Is there some stuff you remember about what it took to get us to be
127:11 so fast? You mentioned a few things that were, the agents were discovering.
127:15 - One of the things was to treat the ... lag of human input as an opportunity to preload. So--
127:23 Hmm ... there's five questions you answer when- Yeah ... you set up an Omarchy machine.
127:26 So it's doing stuff in the background. We're doing stuff in the background. And that was one of the things I didn't
127:30 even consider, which is, I mean, the oldest trick in the book.
127:34 All sorts of video games for a long time have done this, where you let the
127:38 player interact with something, and then you do preloading and so forth in the background.
127:42 So it's an age-old technique, but I just hadn't thought about it for an installer.
127:45 I thought, first you ask the user some questions about their password and their
127:49 username and their time zone and so forth, and then you do the work,
127:53 when, in fact, you could preload that. Now, the other things were just a bunch of other
127:58 tweaks to how the installer itself was doing things in a certain order. Some of them was also just shrinking it.
128:06 The last version of Omarchy was 7.5 gigabytes, and a lot of the time we're down to now is literally
128:13 decompressing compressed package files. That's the bulk of what is
128:17 on the ISO, and that's the bulk of the install time. So by shrinking it down to, I think we're
128:21 at 5.85 gigabytes now, a pretty chunky reduction, it almost literally translated one-to-one, and I did some really interesting things there.
128:30 So Omarchy uses the JetBrains font, which is an awesome font, beautiful font, and actually a font I didn't like
128:38 at first glance, but then it was the only font I could get to look
128:41 perfect in Mitchell Hashimoto's Ghostty. Ghostty, for whatever reason, did not wanna render my beloved Bitstream
128:50 Vera Sans just the way I was used to it. Mm-hmm. But it rendered
128:55 this JetBrains font perfectly, so I switched over. But it...
128:59 The standard package for JetBrains font on Arch is 200 megabytes because it includes all these variations and all
129:07 these subtypes and whatever. And I realized, like, we're not using any of that. We're using the monospace
129:13 version in this one Nerd, uh- Yep... font patched edition. Do you know how big that is? I think it's 16 megabytes. Hmm.
129:21 So I simply just came up with a new package, the slim version-
129:25 Mm-hmm... of the JetBrains package. Nice. And there, right there, I saved 180 megabytes.
129:30 I just went through and did that a bunch of times. We did that with the NVIDIA
129:34 drivers too. So the NVIDIA drivers, when they came straight off the, I think the
129:38 Arch repo, were not compressed with the most extreme form of compression, which I think called ZSTD, which is a
129:46 really slow form of compression. It's slow to create the archive, but if you're building a
129:50 package- Hmm... you have all the time in the world if that means you can then shrink the ISO.
129:53 So on just the two NVIDIA packages, I think we saved 200 megabytes. Mm-hmm. So what I loved about this
130:01 was I imagined myself being a car designer at McLaren. So McLaren makes some of the lightest
130:09 super sports cars in the world right now, and they are absolutely obsessed with shaving every
130:16 damn last gram off their cars. That's why even a new 750S McLaren is just way
130:23 lighter than the competition. They use a carbon fiber monocoque, and that helps them over like Ferrari, for example, that's
130:31 uses aluminum. But they also just have this obsession with weight, and I would see this
130:35 interview with the McLaren folks talking to some journalist just going over, "Yeah,
130:39 we saved, uh- ... 370 grams- Yeah ... off that one." And I just went, "Wait, what? There's someone on a car that weighs maybe
130:48 1,040 kilos or 1,380 or something. They're s- They're worrying about
130:55 370 grams?" Yeah. What obsessed maniacs. Hmm. I love them. Yeah. I want that in my life. I want a McLaren-
131:03 ... just for that reason. So I thought in that moment as I was shaving
131:07 individual megabytes off the packages, man, I'm kinda like a McLaren car developer here, even though- Yeah ... I mean, we haven't even completed that work.
131:15 The package could shrink even more. But this is actually this
131:19 other conflict that we've had with Omarchy and the traditional Linux
131:23 community is bloat. Mm-hmm. This idea that it's actually not kosher in certain circles to ship with
131:30 pre-installed software because who are you to make decisions
131:34 about what programs people should use? And I'm like, "Dude, it's in the fucking name. It's called Omarchy because the Oma part is
131:42 short for omakase, which literally means chef's choice. I'm the chef. I'm making the choices, and this is the
131:50 collection of applications I think is awesome," which by the way has a video
131:53 editor. It ships with OBS- Mm-hmm ... to do your recording. It's just in
131:57 there. Yeah. OBS is pre-installed. Nice. Kdenlive is a video editor, timeline editor, which is great. I edit all my videos with that. And then I made
132:05 my own clip editor. You talked about the thing with syncing up.
132:09 Yeah. I had this thing with clips, so I would do a podcast or whatever. I need a clip for X.
132:13 Mm-hmm. And it's kind of a hassle to use Kdenlive or other- Mm-hmm ... timeline editors, so I made
132:17 Omacut. Mm-hmm. Omacut is included too. Omacut, cool. It's just for making clips, and you can
132:21 drive the whole thing with keyboard. So you can go to just the right time mark- Mm-hmm ...
132:25 and then you can hit Control + Space. Hmm. It moves the clip line to that line.
132:29 Then you go to- Nice ... the end of the clip, and you hit Alt + Space, and it moves the end line.
132:33 Nice. And then you hit Control + S, and it saves. Nice. I can make clips so damn
132:37 fast now with Omacut. It's amazing. But it has all the software on it. It has Neovim on it. It has Herder
132:44 on it. It has Tmux. It has a terminal. It has a bunch of themes. It has a bunch of background images, the best ones I could pick
132:52 out. That takes up some space. It should because this is not supposed to be this
132:57 just barren landscape where you have to reconstruct everything. Mm-hmm. This is supposed to be
133:01 a productive system the minute you unpack it. What is it built in?
133:05 So it's actually mostly built in Bash for the Linux configuration part. Right. Bash is a great language for that.
133:14 It is made for that really, for system administration. It also has its limitations. You should not build everything in Bash
133:21 - How are agents with generating Bash? Pretty good? - Oh, amazing. Except for one thing,
133:27 and I have to smack the agents over the back of the head every single time I catch it,
133:31 and remind it to look at the agent's MD file, because I will have an
133:35 instruction in there not to make early exits. The agents love to have preconditions instead of fully expanded conditionals.
133:43 I like something that says, "If this, then do that." Agents like to have precondition or exit, another precondition or exit.
133:53 And then it falls down to the final thing it actually wants to do. I hate
133:57 that style. So except for that, it's gotten shockingly, dramatically, awesomely good, to the
134:04 point where when we talked last time, we talked about how I would ask the agent
134:08 for a Bash thing, and then I would type it in myself, and I kinda wouldn't remember.
134:11 And then I took the effort to make sure that all that competence wasn't draining out of my
134:15 fingers, so I actually learned how to write Bash myself, and now I'm not
134:18 writing any Bash myself. I have not written any Bash myself for probably a couple months, because the
134:22 agents have just gotten so good, that if you give them the few pointers on
134:26 style, and occasionally put pushback on complexity... This is...
134:30 I tweeted about this a couple of days ago. It blows my mind how
134:34 often an agent can do something, say it's done, have the work reviewed by
134:38 another agent- ... they say it's done, and then I go like, "Eh, looks a little too complicated for me."
134:43 And then it'll go like, "Oh, yeah, you're right. You're right. I totally
134:47 overcomplicated this," and it'll cut it in half. I'm like, "Couldn't it
134:51 just do this, like, from the outset?" And then someone else reminded me,
134:55 humans are the same way. If you work on something and someone else reviews it and they just are like, "Eh, it looks a little complicated," quite often you're
135:03 able to go like, "Ah, you're right." - See, that, that just... Those little indications to me
135:09 show that the programmer is still needed, that intuition, 'cause I think, what, what did you say? I have a sense of the landscape
135:16 of the code. - When I build something like Omarchy, I, I do look at the code. I look at the shape.
135:23 And I look at the proportions. Like, I don't have to chisel everything myself anymore. This is, again, when I'm flattering
135:28 myself for... Not one moment I'm a McLaren audio, uh auto engineer, and the next moment I'm a da Vinci, working
135:36 with his whole studio of students- ... who are doing all the chiseling on the, the fine marble, right? And he goes like,
135:42 "Ah, no, the proportions are not quite right." That's how it feels working with
135:46 agents now. That I can give feedback like an editor. That, do you know what? I'm not gonna write everything myself, but I can tell you
135:54 when the argument doesn't land. I can tell you when the shape isn't
135:58 proportional to the problem, or when it just feels like it's too complicated. And
136:02 shockingly often, it just works. It's like the early days, where the jokes was,
136:08 "Don't make mistakes." I don't think you have to tell the agents that anymore, because now the
136:12 harnesses are so good, they run automated testing. They get that feedback. But
136:16 you actually do have to tell them, "Make it simpler." - Do you use voice at all for the interaction? Or are you... How much are you typing?
136:23 How much are you- - I type the whole thing. I actually, in Omarchy's built-in Voxtype-
136:27 ... Audio transcription, and it's actually really good. I just find that
136:33 I don't think that way, and it surprised myself that I don't like to talk to the computer.
136:40 Uh, I like when I'm in the car, but when I'm in front of the keyboard, I just
136:43 wanna type. I also just love typing. Are you using the voice?
136:46 - I'm an introvert who doesn't like social interaction- ... or speaking. And-
136:50 - I think that's part of mine too. - But, like, for me, for some reason, voice is really powerful. I want
136:57 to give this... So I, I run with this thing. It's called Plaud. So
137:01 I, I just hook it up on my, you know on my T-shirt here. You press a button, it starts recording. And this is actually more,
137:09 in terms of flow, much better experience than behind a computer. And I,
137:14 uh, save, well, usually it's agent gen- generated problems that I wanna really talk through. And so I write, I speak a long prompt,
137:28 like, and a lot of it, it's like stream of consciousness reasoning. So I'll
137:33 sometimes just change my mind about a design, but all of that is in there,
137:37 and then I have an LLM that processes that. With Plaud, there's a,
137:41 obviously an API. You can grab it, transcribe it, clean up the transcript,
137:45 'cause you, you want it to be... 'Cause I'm referring to files.
137:49 I'm referring to technical concepts, like files and functions-
137:53 ... and so on, that you wanna make sure the transcription, the-
137:56 ... the speech-to-text gets correctly. So you wanna have a dictionary. You want it to be aware of the code
138:00 base when you process the files. And then you have this gigantic
138:04 thing that's often 10, 15, 20 minute me discussing the full design, and I find that to be
138:12 really, really powerful. Because it doesn't have the problem overs- of overspecification because of how
138:20 much stream of consciousness thinking the system has about what you're imagining. And this is really good
138:28 for the early steps of a design. Like, if I wanted to build Omarchy from
138:32 scratch, the first stages, I talk through. It could be an hour. There's been prompts that I had for an hour, and I'm just running and-
138:40 - That's amazing - ... and speaking. I highly recommend people try a voice-based-
138:48 - So- - ... speaking for prolonged periods of time. - I haven't actually-
138:51 - As opposed to very specific - ... I haven't, yeah, I haven't used it that, and maybe that's how I'm using it wrong.
138:55 I try to just type or speak two sentences, and then I get a little frustrated that I could've typed that just as
139:03 fast maybe. But I have not tried this whole thing of just talking continuously for 20 minutes.
139:09 - And especially you because you speak very fast. - Yeah. No, I could see it. I think because the reason I've been, I've been trying to
139:16 do it for things like replying to emails, and it got to a point where I thought it could
139:22 help. And it sort of did, but then it would make a small mistake, and I'd have to
139:26 correct the mistake. But if you're doing stream of conscious to an LLM who doesn't
139:30 care about all this other stuff anyway- ... that probably works way better.
139:34 - There's two things you have to do. One, okay, you have the raw audio,
139:37 transcribes it. I use ElevenLabs for transcription. They're probably the best
139:41 speech-to-text model. There's still mistakes there, so you need to have a prompt that has a-
139:48 access to a dictionary and to a lot of- - How do you build the dictionary?
139:52 - First of all, there's a dictionary of terms that are relevant to you. Then have it
139:55 aware of... So you have an agent that looks through, when you're talking about code,
140:00 looks through your entire code base for key terms that you use often. Also, when
140:04 we're talking about messaging, Gmail, and WhatsApp, and so on, it's aware of all the
140:07 people and common terms that you get wrong. - You've really gone deep in this.
140:13 - That, to me, is a problem that needs to be solved- ... because, uh, one of the issues you probably have speech to text
140:19 is that it makes mistakes sometimes. - Correct. Then I can't use it for dictating an email.
140:25 - Right. If it never makes mistakes, you- - Then it would be more appealing.
140:29 - Yeah, and you are okay being not perfectly real time, actually. You're okay waiting five seconds-
140:35 ... potentially. I don't know what your use case is. I'm okay waiting five
140:39 seconds- ... 10 seconds for it to process everything. - You gotta vibe code this into an app that does all of this such that-
140:46 ... it just works. - I think I mentioned to you offline that it's very surprising to me that,
140:54 translation is not done perfectly. Speech-to-speech translation.
140:59 - It seems like you've gotten much closer to the best of what's possible by putting all
141:03 these things together. - But, you know, when you're shipping that kind of stuff, you do have to start
141:09 thinking about latency. - Don't let the perfect be the enemy of good. I mean,
141:15 if you're suddenly enabling something people wouldn't be doing before at all, it's okay to
141:19 take five seconds longer. I think this is one of the areas that Steve Jobs was
141:23 so good, that he realized, like, the first iPhone was terrible
141:27 in all sorts of ways, right? Absolutely horrendously slow data connection, very
141:32 underpowered, all the things. But the product was so compelling that it just didn't matter. You'd be shipping
141:39 too late if you waited until everything was just right, just perfect.
141:43 If you're solving a problem that people currently don't have a solution for, they're
141:47 willing to give it five seconds. If you're in a competitive market, yeah, okay, it's different. But that means the problem's already solved, and you
141:53 don't need to solve it. - Mm-hmm. Well, I do think there's a few... So, like, Whisper Flow is really good.
142:00 A few people have stepped up that speech-to-text thing. - Yeah. Yeah, we're using Voxtype. But others have-
142:05 - Uh, Vo-Voxtype is open source? - It is. It's just using one of the open models. I forget which parrot model
142:11 it's using or something like that. It works quite well for short commands, so I
142:15 have used it for that. Uh, I think you just hold down F9- ... and it starts a dictation.
142:21 - Oh, so it's already in there? - You just have to get online. It'll offer you to install-
142:25 - Got it - ... the Voxtype package. - Got it. Yeah, yeah, yeah.
142:26 - Because the Voxtype package includes a model that's 150 megabytes, and I
142:30 was like, "Ah." - Yep, yep. - "I don't know if I can carry that." So-
142:35 - Yep - ... we don't have that in there by default, but we have the option set up.
142:39 And I actually just yesterday was thinking we really need to combine all of these things. So Omarchy
142:47 Quattro is truly amazing as a malleable operating system where you can add functionality to it just by talking to your agent.
142:56 But that's exactly what we should be doing. There's a lot of people who would
143:00 find it very natural just to talk to their computer and say, "Hey,
143:03 can you make me a stock panel? I wanna track Apple and Dell,"
143:10 and just see the agent go off through, entirely through voice, both in terms of the input and in terms of the
143:16 output, and then your operating system just changes. This was the vision
143:20 of Iron Man and Jarvis and- - Yeah - ... bespoke applications that just appear
143:26 magically. I'm thinking the area of AI I find truly intriguing are these live gaming models, I don't know if you've seen that, where the
143:34 AI's inventing literally the next frame, but you can actually
143:38 play the game. And I think if they're able to do that, why can't we do
143:42 that with our operating system? Why can't we just talk to it and ask it
143:46 to be a different way or change, and it'll all just change? I think this, by
143:50 the way, is one of the other breakthroughs that is going to lead to
143:54 the total domination of Linux. The agents have taken all the hardship out of
144:00 diagnosing Linux systems and turned the fact that Linux produces
144:06 these overly specific, totally arcane error messages into its greatest advantage.
144:12 When Linux has an issue, the agent can take that very specific error message that makes no sense to a normal
144:19 human and correlate it with the fact that the agent was pre-trained on
144:23 40 million lines of Linux code, so it knows exactly where to look and dial it down.
144:27 - Yeah. - I have not had a single problem on my Linux machine since
144:32 the beginning of this year that an agent could not diagnose. That was
144:35 not true a year and a half ago. A year and a half ago, I was
144:39 still searching on forums- ... to find answers to esoteric questions.
144:45 Now the agents know the source code of not just the Linux operating
144:49 system, but every single piece of software I have on that box. The agent
144:53 has access to the source code of all of it. In fact, this was one thing I built in just
144:57 before we, uh, we shipped. So once you set up your default agent-
145:02 ... Omarchy Quattro has a crash watcher. If any app on your machine crashes, it'll pop up a little
145:09 thing, ask you whether you want your AI to diagnose the problem.
145:13 - Yeah. That's amazing. - Unbelievable. I've seen things where there's an
145:18 error in some subsystem. Some, some... Not even Linux, just some app I
145:22 installed. The agents start digging through the logs and looks up systemd log. Then it checks out the damn source code.
145:30 Of the application that crashed, pins down that it's in this
145:34 Rust file line 472 that there's an unbounded, unwrapped variable that overflowed or whatever it is.
145:43 Then offers you whether you wanna file a bug report with all this detail.
145:49 I'll give you one amazing anecdote. So I was working with this guy JDX, who's working on mise. This, this package manager
145:56 for fast-moving development tools. This is how we manage all the agent software and so
146:00 forth. Normal package managers are not built to be updated seven times a day.
146:04 And all of these agent harnesses are updated about seven times a day, so we
146:08 needed an out-of-band package manager, and mise just turns out to be
146:12 perfect for this. Anyway, it had a small bug. There was some race condition which agents are
146:17 very good at finding because when you start running multiple agents at the same time, they will suss out all
146:21 these race conditions you have in your underlying infrastructure that was never triggered by a human
146:26 trying to do manual thing. So it finds this issue, right? I tell
146:31 I tell the agent, "Hey, can you post this as a bug report to JDX's GitHub?"
146:38 And unfortunately, right before that, I had had to do a QA run on Omarchy itself with eight different agents. It had found 28 real issues-
146:49 ... that it needed to file. Well, it went to GitHub and tried to file all
146:52 28 issues at the same time, which it did in about, uh, I don't know, 12 seconds. GitHub, not
147:00 unreasonably, marked that as probable spam and banned- ... my Omarchy bot.
147:05 And then I couldn't do that. So I was blocked from the bot having access to GitHub, so I just told it, "Hey, do you know
147:11 what? Just email JDX. Here's his email address." I had already set it up with hey.com-
147:17 ... an email address. We have a CLI that's in beta right now.
147:20 So I set it up with that so it can send to email. That's how it gives me reports about
147:24 outstanding issues and PRs. So it sends, um, JDX this email about the bug it found, and he was like,
147:32 "Wait, this is the first time I've gotten a bug report in unreleased
147:36 software." Because it had downloaded the source code to mise,
147:40 saw that he had already fixed the bug, but that it would not fix the problem
147:44 entirely in software they had not shipped yet. Pinpointed the problem, and he was like, "Damn it-
147:50 ... I got a bug report before we even cut a release." - That's amazing. That's amazing
147:54 - ... believable AGI levels of mind-blowing stuff. - And that was actually, if you remember, that was an open question
148:02 whether AI systems could debug well. That was kinda the thought that, yeah, sure, they can write code, but can
148:10 they find issues and- - They are so incredibly good at this, and this is actually interesting
148:14 because I was on the perhaps same side of that, a little skeptical about
148:18 can it reason about all these things. And Mikhail, who's the CTO
148:22 at Shopify, ran a scientific study on this. They had... And this was, I think, late last year or early this year, where he
148:29 had agents go back through all the incidents, both outages and,
148:33 and other problems that Shopify had had in production, trace that back to the
148:37 PR that was merged, and find out whether PRs that had been reviewed by human or PRs that had been reviewed by
148:45 agents were of higher quality. Well, lo and surprise- ... the PRs that have been reviewed by agents caused far fewer
148:54 issues in production. And this was with models we had six months ago. At this point, 100% in the
149:02 majority of domains we work in today, agents are better at finding bugs.
149:06 - So you now run agents all day. What stands out to you as the better model? Who's currently winning?
149:14 It seems to be changing constantly. Sol, Fable, Grok 4.6, Gemini, the Chinese OpenWay
149:22 models. - What's amazing to me is that you could rattle off so many different contenders.
149:29 That this market is so wide open, that it does actually change back and forth, that we have real
149:36 competition, that there are so many labs that are able to get either to the frontier or close to it.
149:44 That, by the way, is remarkable. I still don't fully understand that.
149:48 But to answer your question, the best model in general right now is Fable.
149:54 The second-best model, in my opinion, is Opus 5. But the tier just below Opus 5, and
150:02 it's not even that they're always below, sometimes they're ahead.
150:05 I'll get to that in a second. I would rank GPT Sol very good.
150:13 Grok 4.6 I just started testing a few days ago. I had this wonderful test that I've set up by accident
150:20 where I translated this Python library into Rust. Uh, the screensaver you just saw with all the cool animation?
150:28 That's powered by a Python library called Terminal Text Effects.
150:32 Really cool library. We've been using it since the first day of Omarchy.
150:36 The problem with that is it's written in Python, so when it starts up, especially on a
150:40 laptop, and it runs in Python, it uses all of your CPU to do these effects, and therefore, it means it uses about 30 watts
150:48 of energy, and it spins up your fans, and it drains your battery.
150:52 Doesn't really matter on a local computer, but it does matter on a laptop. So I thought, "Do you know
150:56 what? This sounds like a problem for Rust." So first I gave Fable the challenge,
151:03 and all I told it was, "Here's the source code for the Python library," this TTE library that had a bunch of dependencies and so forth.
151:11 "I want a Rust version of this with no dependencies, a single executable." Like, that's what Rust does. So basically, I want it in Rust.
151:19 I want it to be pixel perfect, frame by frame, do a full analysis, don't stop until you're finished.
151:26 - Yep. - That was the prompt. - Yep. - I kid you not, in just under 45 minutes, it was like
151:34 Am I heard all? I'm finished. I've checked everything. I have reduced the startup time from 86 milliseconds to two
151:46 milliseconds. I have sped up the execution time by 9.6 times, I believe it was. The executable is three megabytes. Do you wanna run it?
151:59 - That's awesome. - So I run it— - Oh, man. - ... and I don't know why I'm surprised, because this
152:05 translation job is something we've known for a while that AI is
152:10 pretty good at, but it was still staggering to me that I could one-shot
152:14 a full translation of a Python library I had been using for a year, that others had been using for much longer,
152:22 and turn it into a Rust executable without knowing any Rust,
152:26 without looking at the Rust code at all, and produce this executable that I then told the agent right after, "This is great. Ship it."
152:35 It packaged it up as a new package. It told me, "What do you want it to call?" "Uh, let's call it
152:39 TTFX. Let's create a new Git repo." It sets up the Git repo. "Let's create
152:43 a new package, build package for our build system." It puts that
152:47 up. "Let's push it out. Let's open the pull request to the Omarchy itself so we switch around from TTE, the Python implementation, to
152:55 TTFX." It does all of it, and I'm just sitting there. And again, I'm already at this point fully
153:02 delirious with agent acceleration, and I still had to lean back and I go like,
153:07 "This is AGI, isn't it? This is what AGI looks like. If we have... If these moments are just what it is all the time, this is AGI."
153:15 - Now, as I understand, you also repeated the same experiment not just with Fable.
153:21 - I did it on all the models. First, I did it... Actually, funny thing.
153:25 So I ran out of Fable tokens about two-thirds through, and it just automatically switched over to Opus 5 and kept going and finished
153:33 the job. Part of the reason why I think it was able to do that was that the first thing
153:38 Fable did was create a plan, and it was a really detailed plan. I think it had eight separate steps of, "Here you do it,
153:44 and here, how you analyze it, and here you run the effects," and so on. I didn't review the plan at all.
153:48 I didn't change the plan. It just made the plan so Opus could take
153:52 over. And then I thought, "Well, if Opus can finish the job,
153:57 maybe some of the other agents could finish it, too." So the first thing I did was
154:01 I gave it to... I think I gave it to Sol. And Sol finished the job, too. It took twice as long, so it took, I think, about an hour
154:09 and a half. But here's the kicker. The per token cost, I didn't pay per token. I have a
154:15 max subscription to Claude, right? So it did it within the subscription. It used all my
154:18 tokens, so I had to switch over to Opus 5. But if I had paid per
154:22 token, it would've been 550 bucks, I think, to do the whole thing. And I
154:26 thought for a second, "Holy shit. What a steal." If I personally had to learn Rust well enough to be able to do this
154:34 translation, I'm looking at a nine-month job here. I can pay 500 bucks to have this translation happen, and suddenly I get a
154:43 10x execution speed up. This is ama- I would totally pay 500 bucks for this.
154:48 But competition. So I give it to Sol. Same plan. To be fair, I didn't ask Sol to do the plan. I just took the Fable plan,
154:56 gave it to Sol. Sol, in an hour and a half, and I think $46 worth
155:04 of token, repeated the task. Did the same thing. And I thought, "Well, blimey, that's amazing." Then I got greedy.
155:17 So I asked GPT Luna, which is this crazy cheap model that OpenAI has as well,
155:27 "Can you do it?" Absolutely not. First of all, it didn't even wanna
155:31 start the task. I think something happened I don't know. I think
155:34 it was in the spring, where we didn't need these slash goal things anymore. The agents could just automatically keep going in a loop if you
155:42 told it not to stop- ... and so forth. So Sol could do that. Fable could do that. But
155:48 Luna couldn't. So I had... I think I did 12 prompts. Kept telling it to do it, and eventually
155:55 I sorta got it started. The first thing it did was to cheat. So the
155:59 first thing it, it looked outside its own directory, saw that there was already another imple-
156:03 implementation, and just did a short wrapper around that and said, "I'm done."
156:08 Hilarious. But it couldn't finish because it made just a couple thing. But I mean,
156:12 okay, so it can't do that. Then I gave it to Grok- ... 46. And I had used Grok 45 a little bit, and I thought like, "Ah, I mean,
156:21 it's cool that there's others trying, but, like, I'm not gonna use it," because it
156:25 felt quite far behind. Grok 46 fucking completes the task. 10x speed up, same size executable. $55 Worth of per token cost, I think it was.
156:37 Absolutely unbelievable. Then I repeated, too, with, uh, Kimi K3-
156:43 ... which took forever. I forget how long Kimi actually spent on
156:47 it. And then I also did it with DeepSeek V4 Flash first. And Flash failed the same way that Luna did. It couldn't do it. And then I did it with Pro,
156:56 and Pro also completed the task. It took 2 hours 45, $23. So here we are, right? Like, Fable, clearly the best. It was the
157:06 fastest. It was the one that wrote the plan, but 550 bucks, and the output the same.
157:13 Uh, the others, Sol, Grok, about the same 1/10 the cost. DeepSeek, 1/20 the cost, but you have to wait a little longer.
157:24 Absolutely gobsmackingly incredible. And now, by the way, by the way, so I had Fable finish the first job, and then Opus finished
157:33 it. That was the first one shot, right? It's 10 times faster. I did two auto research runs, which isn't
157:39 even auto research anymore. You don't have to do the slash. You just tell it to keep going
157:42 until you tell it to stop. It ended up... I think we ended up with a 46 time execution
157:49 improvement over the original. - It's incredible. - Absolutely incredible.
157:52 - It's just incredible. Um, I do, you know, I just don't think there's anything that compares to Fable
157:58 in terms of planning. So I usually do Fable for planning and for reviewing,
158:05 and then something else for implementation, like Opus 5. And it's really nice. It's a, it's a really nice setup that allows you to
158:12 not run out of tokens too much. - What I found, it's really interesting because Fable is, in
158:18 my opinion, the best model right now, but it also makes mistakes. And the
158:21 best way to get the best software, I would actually rather have two differently sourced sort of... I mean, they're not
158:29 mid-tier, they're all frontier. But have, let's say Opus 5 and Codex-
158:35 ... and have one check the other's job. This is my standard operating procedure now.
158:40 I'll have Opus or Fable do the work, and then I always end it, review
158:45 with Codex xHigh. And I've also started using Grok just to test it out, and it's also quite good. And it keeps finding stuff.
158:51 And then that's my workflow when I'm having my agents on my own
158:55 machine do it, and then I push to GitHub. And then Copilot, I kid you not, has
158:59 actually gotten good. Copilot keeps finding stuff that's legitimately broken, which is also incredible acceleration because the
159:07 first version of Copilot that started doing this was literally
159:09 retarded. It would just constantly flag things that were nonsense.
159:13 It would constantly flag the same problem over and over again as you would push. It was really annoying, so
159:17 I think a lot of people actually ended up turning that off. And if they did, they should turn it back
159:21 on because it's actually quite good. Keeps finding things. And if you then take that, and we shouldn't be surprised. Why are we
159:27 surprised? Even if you're a good programmer, if you finish a job
159:31 and you ask your also very good peer to review it, you're gonna
159:35 end up with better code. Of course you're gonna end up with better code. So build that into your process.
159:39 Pick one of the agents to drive with. I've mainly been driving with Claude, which is actually interesting because I have some other
159:47 reservations about Anthropic. But the reason I'm sticking with Claude is, in my
159:51 opinion, they actually have the best harness. And one of the reasons it's the best harness is it's
159:57 multi-agent running. So if you wanna run multiple agents at the same
160:01 time, you can do arrow left when you're inside a session, then it goes back to Agent View. And here in Agent View, you can pick up another agent.
160:08 So if you wanna do this thing where you have multiple threads going on-
160:11 ... the Claude Code is just the nicest setup. They also just, they keep being a
160:15 little further ahead, which probably shouldn't be surprising. I mean, Boris is one of the
160:19 guys that's working on that, he was the first one and basically came up with the thing.
160:22 It's just interesting that that has been an enduring advantage. I've also used
160:27 OpenCode a lot. I think OpenCode is great. I use OpenCode mainly as my main harness for all the
160:31 open models, so Kimi K3 and so forth. And I inference not on the Chinese servers.
160:40 So I inference on Fireworks- ... which is a really nice service where you can just pay by the token, and
160:47 if you're using these open weight models, they're not that expensive by token.
160:50 So that's a great way to do that. But I don't think I would be using...
160:54 Well, maybe I would. But if... Using Claude with a subscription does feel like a bargain, like a crazy bargain.
161:03 - Yeah, I knew. And I actually have four. - I just signed up for my second one this morning-
161:09 ... because I got up at 4:00 with jet lag, and I started working with the agents right away, and
161:14 now I'm three days away from limits resetting on Fable, and I ran out of tokens.
161:20 - So and they should really— they— - I mean, I don't... Why is that so complicated? Can't you just stack one subscription?
161:25 Why do I need to log in multiple times? - And obviously you'll have to develop some kind of tooling-
161:30 - Yes, I did that too. - ... to move from one- ... to the other. And then-
161:33 - Oh, we're building that into Omarchy, by the way. So the next version of Omarchy is gonna ship with
161:36 multi-sub support. I wish that the labs would just make you buy a max 100 times.
161:43 - It's kind of actually fascinating though for me because there's...
161:48 It's very challenging for me to use Grok 46 and then a Claude Code because Grok is so fast.
161:57 - Yes. That's actually the deal. The deal with Grok is that their fast
162:01 mode is cheaper than the regular mode on the others. And when you've used Grok 46 on fast, it's kind of addictive 'cause
162:09 it almost gets you to this single thread mode- ... where the agent is able to keep up with you.
162:14 - Yeah, but I don't know how... I can't... Once I get used to Claude Code,
162:18 uh, I can't handle the speed of Grok 'cause I'm used to now multitasking.
162:24 And so now I'm back in this- ... anxious mode of, uh- - Yeah. No, I agree.
162:28 - ... switching. So I mean, all of this is a learning a learning process of what programming actually is supposed to look like.
162:34 - By the way, one of the reasons why I'm not sure what the final form of these harnesses is gonna
162:38 take, because one of the things we've started experimenting with at Base Camp is that we put
162:42 the agents inside of Base Camp, and we treat them as coworkers.
162:45 - Yeah. That's awesome. - And then we assign them tasks- - That's awesome.
162:47 - ... inside of Base Camp, so they will do work on a to-do in Base
162:51 Camp, or they'll work, do work on a card. And it turns out that
162:55 a collaboration tool that's optimized for asynchronous communication is
162:59 actually the right format versus these harnesses are a little more like chat.
163:05 And chat, which was, I mean, the original format for these things, is
163:09 not the right thing because you're sitting around waiting. It entices you to sit
163:13 around and wait, versus when you're dealing with something like Base Camp and you can just give
163:17 an agent a to-do item, you don't expect to hear back immediately.
163:21 So I don't know what the final form is gonna take, but I've been using all of the harnesses.
163:24 So I use them inside of Herdr, and Herdr has panes. So oftentimes I'll
163:28 run Claude up top, and then I'll run Codex down below, and then maybe I also have an OpenCode set up. But
163:36 I will say just very recently, I found that the human in the
163:40 loop is the limit here. So I've started setting up more automated systems. I've been building an Omarchy bot
163:46 that can do more autonomous development on Omarchy that is just
163:50 On a regular schedule process, all depending to do or PRs and
163:54 issues, and then send me an email using the Hey CLI, and then I just get this email. Here's 12 PRs that are either ready to
164:02 go or I think you should close. And then I just make the final determination there.
164:06 - Do you find the multitasks, the task- switching mentally exhausting?
164:10 - Yes, but in an exhilarating way. Like, one of the things I always loved about
164:14 race cars was when I would stumble out of the car absolutely
164:18 smashed and barely able to hold my head up, and I'd lay down on the
164:22 garage floor and just think, "Holy fuck, I'm alive." That's the kind of exhaustion, mental exhaustion I'm feeling at
164:30 moments with the agents right now. - By the way, when you're racing, where, what's the source of the exhaustion? Like-
164:34 - It's physical. It's just... I mean, I bet it's the same thing with jujitsu, right? Like, it's actually satisfying to
164:41 feel exhausted. It feels like you've applied yourself. And I feel that with the mental exhaustion you can get from agents.
164:48 - Are you also mentally drained from racing? - Yes. If you drive in difficult conditions, especially if you drive in the rain, where you're
164:53 constantly managing the thing right at the knife's edge. But
164:58 most of the time, I just enjoyed the physical exhaustion. - And here, with the multitask switching, I do find that, like,
165:05 I get really exhausted, and I think it's because I'm thinking a lot.
165:11 'Cause, like, you're basically- - That's exactly it. - ... there's no, like passive low energy thinking.
165:18 - No. - You're like- - There's no coasting. - ... problem. - There's no coasting.
165:21 - Solve problem, next problem, solve problem, and you're really thinking.
165:25 - It's funny. It's actually very similar with racing. There's some tracks, like Le Mans,
165:29 where you get time to relax. There's the Mulsanne straights where it's kinda
165:33 long. You're going fast, but in a straight line, you can take a breath.
165:37 And then there are other tracks where it's just coming full on all
165:41 the time, and you don't have any moment to relax, and you stumble out of
165:45 the car after an hour in each, and it's very different how exhausted you are. And this is
165:49 exactly what I'm finding with the agents too, that when you're running at max
165:53 human capacity- ... you're just constantly in a corner. - Do you think people are in danger of burnout?
166:01 - I think the moment we're in right now is gonna pass. That this need to...
166:09 It's not a babysitting motion, but constant interaction with the agents is
166:13 gonna fade. From what I've seen on Omarchy, I think we can automate a ton of the development and debugging of the system-
166:24 ... to the point where I can just review that email once a day, and then make the decisions once a day. Yep, goes, no,
166:31 goes, in, out, and whatever. We're not there yet, but I think we're gonna get there.
166:36 - Well, can you sort of, like, empathize with the feeling that,
166:40 say, a young person sitting in San Francisco, everybody around them is not
166:43 sleeping. They're obsessed. They're programming nonstop. And there's this kinda sense, that AGI is gonna be here
166:52 any minute. There's, everybody knows somebody who's made millions of dollars
166:56 because they sold a startup. And so there's a kinda feeling like you shouldn't be sleeping.
167:04 You should be constantly managing agents and programming and building and,
167:09 and that feels like a moment in time we'll look back at. - It's the birth of a new paradigm. It's always messy.
167:16 - It's always messy. - It's always exhausting. There's no other way around that. And by the way,
167:21 hasn't San Francisco always been like this? I remember the dotcom boom years, and everyone was just the same, and then
167:27 it was all mobile, and before that it was the gold rush. So I think that's probably-
167:30 ... just San Francisco. It attracts- ... the kind of individuals-
167:33 - The dreamers - ... who would, who would think, "I don't have time to sleep." I've not been on that track,
167:39 generally speaking, but I will admit that the last three months have felt more exhausting than
167:47 any other project I've done in the last five years, since the Hey launch. That was the
167:51 last time we had a crazy exhausting launch. - Do you think you can keep going at this pace?
167:57 - No, no, no, no, no. This is not sustainable at all. - Okay.
168:00 - But I also, I can see the end of the tunnel. Because the automation, which by the way, everyone is building. This
168:06 is so hilarious. Like, everyone is building their little gas town, their little
168:10 agent coordination, their setup. This is all gonna be solved. We're not all gonna have to
168:14 build our own coordination harnesses. Of course we're not. And in fact, I'm
168:18 a little surprised that it's gone this long, that there's not more of it has been
168:22 sucked up by the major labs. You'd think that they'd just built this stuff in.
168:26 But some of it is also this is what a new domain looks like. I remember for a hot moment when
168:30 JavaScript kind of came to realize its own power, and there was basically a new JavaScript framework every five minutes, right? Like, a lot of churn.
168:38 This is what happens at the advent of a new paradigm. So I think it's natural, and I think it's gonna settle down, and I can already see the light at the
168:45 end of the tunnel. So I don't mind a sprint. In fact, I welcome it now. I
168:49 like this notion that most time is calm, but then occasionally you gotta climb a mountain.
168:56 - If we look back to the DHH a year ago, what advice would you give to that guy?
169:03 - I wouldn't wanna spoil it. - Okay. - What a- ... thriller that this has been.
169:09 I mean, if you would've scripted this, I'd be like, "This is so far-fetched. Get
169:14 out of here. Unbelievable," right? I wouldn't wanna know anything. The
169:17 fact that this rollercoaster and this acceleration has been absorbed in real time by everyone, no one
169:24 knew, right? Like, some had premonitions that were a little better than others, but no one knew, not exactly the way-
169:31 ... it was gonna go and how fast it was gonna take off. I mean, it's the show of a lifetime. I mean, the mean cinema-
169:40 ... absolutely applies- - Yeah, it's incredible. - ... to this moment.
169:43 - Uh, that said, what do you think DHH from a year from now will be like?
169:49 I mean, what are the possible versions? - That all our dreams are coming true. Like, this is the AI
169:56 maximalist abundance argument. That once Elon's robots have the level of intelligence and AGI that I'm experiencing on some of
170:08 this development stuff, the world is gonna be so unrecognizably different that we can't even imagine it.
170:17 Now, I don't spend any time thinking about that because I think that is the way to
170:21 the AI psychosis. So therefore, I just immerse myself in the moment and get
170:27 absolutely the maximum out of it and- - Just have fun. I mean, I get it why people are finding this exhausting. Just keeping up.
170:35 Not even using agents, just, like, being like, "What's the latest model? What did, what
170:39 tool are we using now? Is it TMox? Is it a Herder? Is it none of it? Is it
170:43 agents?" Whatever. I get it, but I also again think we should be so blessed.
170:49 You are alive in this moment where decades are happening- ... in weeks.
170:54 - Um, quick bathroom break. I've been generating a lot of video recently with Higgsfield,
171:01 and so they, they became a sponsor. Created a racing video. Wanted to get your opinion on it.
171:06 - Oh, yeah? Oh, let me see. Let me see. - To see, um- - If they downshift on a straight like they do in the movies, I'll call it out. That is the
171:14 worst. Let me see. - Yeah, see if- - Oh, that's a real clip - ... this is fully AI generated. So first it goes-
171:22 - Wait, what? This is AI? - Yep. - Shit, that's my suit. That's my car.
171:28 - Lights go green on a street that never sleeps. - Oh, my God.
171:36 - Do an audio- - Are you kidding me? - But they never came. Is it good? I got a bad case.
171:44 - I mean, it's crazy. You really have to notice in the... Like, that car didn't have headlights on and the other one did.
171:54 And that looked like a 60-year-old version of me. But holy crap, the...
172:01 Wow, that was crazy. - So there's parallels here to programming because if you do just straight
172:07 video generation with no human in the loop- ... you get a lot of weird artifacts-
172:12 ... and so on, and you can't really do... So they do and I highly recommend people
172:16 go to the Higgsfield YouTube. They have full 90-minute original movies.
172:22 And, like, I can't look away. It's really cool. 'Cause it used to be that it's just, like, something that just feels like trailers.
172:30 Like advertisements for something. - Yes, yes. - This is actually telling stories, like, people's faces and they're talking and you're like,
172:37 you're drawn in. It's not quite where programming is. And so the question is how do you integrate the human into the loop of the
172:45 filmmaking process? So you have to create the people, and they're-
172:49 ... they have to be kept consistent. And then for the images that are kinda like feeding the
172:55 thing in the creation process, you have to correct through prompting things that don't feel right.
173:04 Because when you go from image to video, the stuff that doesn't feel right will be-
173:11 - It's gonna be zoomed up, right? - That's the stuff that's- ... going to really create, uh-
173:15 ... the non-realistic stuff. The questions I had for you is, like, do you think
173:19 outside of programming, can you apply lessons from programming to video creation, to art?
173:27 - It's a good question because I think good art is very dependent on the
173:31 same thing that good software is depending on: having a vision, having a
173:35 cohesive idea of what you wanna create. How is it different? How is it novel? How is it going to appeal to people?
173:43 But I'm not sure that the creative instincts transfer quite as well. Um,
173:51 I don't know that I would have the creative insights to come up with good video prompts that would create compelling
173:59 content. But it's funny because I have actually, on TikTok, there's this guy, I'm trying to remember,
174:03 Goblin something. That creates these... They are shorts, but not super short, maybe a couple minutes,
174:10 of really interesting sci-fi vignettes. And I think the guy is actually working on a feature-length video right now.
174:20 Um, and I think that this is one of those abundance moments. There
174:24 are creative people with that vision all over the place, especially in a genre like film, where before it
174:31 required a $200 million budget to create, I don't know, a sci-fi movie. Now suddenly, with AI, if
174:39 you have the right vision for it, you can do that in your bedroom. Now, that story's been told quite a few times in other domains. It used
174:47 to be true that if you wanted to record a studio album, well, you needed a
174:51 studio, and you needed to book that. And now you can record at home on your
174:55 laptop, and you can do a lot of these things, right? So we've had this democratization
174:59 process happen with plenty of other domains, but film seems to be one of the last ones.
175:04 - And it's the one that makes, like, me and a lot of people feel weird, right? Like-
175:11 - Oh, I'm super excited 'cause I think it's- - I mean, you- - There's so much garbage being produced right now. It's the same argument with the slob,
175:17 right? So people for a second were worried about slob in programming, and I was telling them, like, "Have you seen the human
175:24 slob? It's also pretty bad." And I think if it comes to studio entertainment, I mean, the bar is pretty low, right?
175:32 Like, the number of great movies I've been dying to see over
175:36 the past, say, five years, not been a particularly high number.
175:42 - Yeah, the video models are just, they're able to create incredibly realistic
175:49 facial expressions. I mean, it just might change the nature of film
175:55 I think they're— I guess my hope would be that video games and film kind of merge, and that film can become interactive
176:06 - Like the malleable operating system, having the malleable narrative, having malleable entertainment, that
176:13 man, Game of Thrones, what if we could just forget the last four episodes of slob that was produced- ... and then AI-
176:23 ... comes up with- ... I don't know- ... maybe 100 different variations, and someone else watches that,
176:29 and I get the version that actually finishes off that show in a way that's fitting.
176:34 - Yeah. Something tells me an individual creator would not create those la-
176:38 would create a different ending for Game of Thrones. - Oh, for sure. Right? And more just to make the argument that talk about
176:45 human slob. That was an absolute atrocious ending to perhaps one of the greatest pieces of,
176:52 Movie history, I was about to say, but, um, series history, right?
176:56 It was so good, and it was actually a compelling argument for the fact that you
177:00 need that strong vision. And as soon as those showrunners had to go without a script,
177:08 it went off the rails. - Do you think... You've mentioned AGI. Do you think we've actually achieved AGI?
177:14 - No, not in the general definition that it's in all the things. But have I seen
177:20 AGI in these glimmers? Absolutely. Especially over the last three months, I've seen things where
177:29 I was thinking to myself, like, "How would AGI look different from this?"
177:34 - Right. - What, what more could there be? And I mean, I can quibble at the margins,
177:38 but on the big picture, I've seen it able to do these long run tasks and
177:45 coordinate these multiple aspects of it that we're so far beyond that early moment where I'm telling it, "Go do
177:53 this, package that." And now it can carry out a whole task from these vague instructions. That feels very
178:01 AGI-like, and that's what's so inspiring about the moment because you catch these glimmers.
178:09 And then you go, "What if everything was like that?" - Doesn't programming basically unlock the everything?
178:15 So you can write a pro- like it can generate the program that does the everything.
178:20 - So, so this is... I'm in that meta phase right now. So I'm building this Amabot system
178:24 that's managing the development of Omarchy, and I'm using AI to build the bot.
178:32 And I'm seeing that recurrent loop where I started it on, like, "This is what
178:36 I want. I want you to be able to have these isolated VM workers and so forth,"
178:40 and then it can do the self iteration. Then it starts running it. It starts noticing its failures. It starts optimizing it.
178:46 And a lot of that system, Amabot in particular, AI has driven the majority of the design decisions. Like, I had some
178:54 vague ideas of where I wanted. I wanted to use this brains and hands pattern where
178:58 the model is running from where the code is executing- ... which is actually a really interesting pattern. Again, Toby alerted me to
179:05 that, where you have a coordinator that runs the model- ... but it's manipulating a safe VM where
179:13 any untrusted code you're pulling down from pull requests or issues or whatever
179:17 can't contaminate the model. And even in that loop, I keep seeing it
179:24 catch itself, spotting vulnerabilities, where it's like, "Oh,
179:28 I actually just took some feedback from a test run. If a clever
179:32 attacker had embedded a malicious payload in the response of the test run, that could have
179:38 polluted something. I took it for good thing. I better start treating that as
179:42 as outside data." And you just go like, it is just... I get why
179:48 San Francisco is so obsessed with fast takeoff- ... because clearly they've seen this 100 times-
179:55 ... more. They've seen it with all the models that are in- ... training now, and
180:02 we keep getting these little glimpses, like the security breach with
180:06 Hugging Face, where OpenAI was training this model, and the model starts inventing ways, essentially sending smoke signals to
180:14 itself through embedding messages in a package manager, where you just go like, first, that's damn clever.
180:22 Second, okay, that's also a little scary. And third how amazing would this be if we could harness this level of intelligence-
180:30 ... and ingenuity towards productive ends? So this is the it's so over, we're so back pendulum that keeps swinging back and forth-
180:39 ... that AI has just delivered that swing back and forth like nothing else. Like barely a week
180:46 goes by where it has to swing one way, and then the next week it swings back the other way.
180:51 Exhilarating. - I'm still at the stage where it makes me truly happy to see those moments
180:56 of cleverness, and they're becoming more and more frequent. Just kinda like seeing, oh, like first of all, at the very basic
181:04 level, when a model gets it. - Yes. - I'll say a basic prompt, you know, a request, and it,
181:11 it doesn't just do a dumb implementation. It deeply understands, and it... That was... That's a, that's always a
181:19 beautiful thing. It's almost like having a good, like, partner, like program.
181:22 - That's exactly what it is. That's exactly why it's so exhilarating and
181:26 why, as we talked about before we started, it can sometimes be difficult to
181:30 come out of hyperdrive. That's also one of Toby's terms here. That
181:34 when you're working directly just one-on-one with agents and you're not looping
181:38 in other humans, you go so fast, and when the agents are really understanding your intent, or even better,
181:46 they revise and improve upon your intent to deliver something that was greater than what you asked for
181:55 And you're doing this with 16 parallel threads. It can be quite difficult to come out of hyperdrive and have to deal with
182:03 squishy humans. They just, they don't think as fast. They don't move as fast.
182:08 And I mean, I'm trying mostly to make light of it, but you can also see a
182:14 dark version of that where someone does end up in an AI psychosis
182:18 where they're not interested in other humans at all anymore. I mean, certainly
182:22 don't have that at all, but I do have this sensation at times where I'm really
182:29 grateful for just the rush of just me and the agents. Like, can I just not talk to anyone else but agents-
182:36 ... for four hours? Like, that's an incredible run. - Toby is such a fascinating human because
182:42 he's running a large company. So I would love to get insights of how
182:49 AI's being incorporated effectively in companies. Because, like, for me, humans are pretty slow.
182:58 And so, and so there's a temptation to, like, use AI for basically,
183:03 You know, AI running inside Slack, for example, or inside, The Basecamp, whatever. And, like,
183:13 collecting all the information about the human interaction, basically handholding
183:17 the humans to speed up as much as possible. And like, and then at
183:21 a certain point you're like, wait, what is this human... What are each of us doing that's not
183:28 replaceable by AI? And that becomes, like, a scary kind of conversation because a lot of work is kind of replaceable.
183:37 - The irony here is it already was. So the number of fake email jobs that currently exist in this world is an absolute epidemic. And-
183:46 ... this, by the way, is not a new thesis. Um, David Graeber
183:50 wrote this wonderful piece Bullshit Jobs, back in, I think it was maybe early 2010s.
183:59 Based on a poll he did in the UK asking people, at that time, would... I
184:07 think the question was something like, "Would it make a difference if you didn't go to work?" For
184:11 humanity, for mankind. And something like 30-some percent answered, "No, it wouldn't." That a third of workers thought
184:19 that their job was fake, that it did not produce any worthwhile, valuable outcomes, neither for
184:29 the community or society or maybe even the economy at all. And
184:34 this was my argument since day one with startups. This was why
184:38 we stayed a small company. This was why we didn't want to take VC, because
184:42 I was so skeptical of the idea that you could get hundreds of programmers, thousands of programmers to do something productive-
184:50 ... as a collective group. And I think what we're seeing right now with some of the layoffs
184:57 is a testament to that, not AI. It's just that during the pandemic, a bunch of overhiring went on, and now AI is a
185:05 convenient excuse to slim down. But I do also think AI is going to expose some
185:11 roles as just not being productive ways for humans to spend their time,
185:16 and therefore we must come up with new ways of spending their time. And the one,
185:22 again, I'm quoting Toby for the third time here, but anecdote that
185:26 he shared me, or observation was, okay, let's say we all-- we lose half the jobs that are currently happening right now.
185:34 Do you know that Formula 1 employs literally tens of thousands of people just to
185:38 run cars around in a circle for spectators? Like, that has, that whole circus has
185:46 no intrinsic value to society besides the spectacle it produces of
185:53 amusement and entertainment. We just decided it would be really fun for these, what is it, 12 manufacturers
186:01 to create these fast cars and constantly tweak their little aero
186:05 improvements, and we will watch it, and we will spend literally
186:08 billions of dollars doing it and employ tens of thousands of
186:12 people towards an ultimately deeply frivolous activity. So if we are liberated from a bunch of drudgery and fake
186:21 email jobs, maybe we all get employed as F1 style engineers and drivers and mechanics and masseuses and the other 1,500 jobs that are
186:33 involved in that, and we come up with new things. I think this is... Humans are very
186:37 poor at imagining what exactly the future's gonna be like. - Yeah. Imagine, imagine a future 100 plus years from
186:44 now, where we look back at this whole span of history, thousands of years where humans worked.
186:52 - Funny thing is we're already doing that, right? Like, the number of fake email job
186:56 people who sit on a chair all day long and type into a machine
187:00 who don't have hard physical labor would be unimaginable- ... to folks in the 1800s, right?
187:07 So the future is always already here, it's just not evenly distributed.
187:11 - Unfortunately, as is always the case, the transformation of
187:15 society required to go from one step to the other is going to
187:19 have a lot of suffering. And potentially political turmoil and all that kind of stuff.
187:25 Because you can imagine how many white collar jobs might be lost in this process.
187:30 - Yes. And I don't think you should make light of that. I don't think it's funny.
187:36 But I also do think it's a necessary component of progress. And
187:40 I think if we look back upon the Luddites or the folks working the fields,
187:48 separation has a tendency to reduce empathy. So we have high
187:52 empathy right now because we know people- ... and we are people who are facing these challenges. But-
187:58 What is our empathy for the Luddites of the early 19th century? Would we wish that all garments were still made by
188:06 hand weaving? No, we wouldn't. We want the comfort and accessibility and convenience of being able
188:13 to buy clothing off the rack. So we must suffer in the moment for future generations to live more prosperously.
188:22 And that's always been true - ... I think that's actually the trick, is to push for progress
188:27 but have deep compassion for the people who have to suffer. And that's many of us. The suffer the transformation required to
188:38 achieve that progress. And sometimes too often in places like Silicon
188:42 Valley or so on, you can over-focus on, look, the kind of utilitarian view of things- ... and look at the progress.
188:51 - Right. - But, you know, considering deeply each individual human
188:57 that suffers because of that progress is important. - I do think that suffering becomes more meaningful
189:03 when you are suffering for a future someone in particular, like your kids.
189:08 And I think this is one of the great tragedies of the moment,
189:12 is the falling birth rates are making it more difficult for people
189:16 to be excited about a more prosperous future that is difficult to get to.
189:23 Because if we're planting trees of which our children will not sit in the shades, I don't know, maybe we
189:31 should just make some lumber and burn it all down, right? I think there's
189:35 a nihilism that can sneak into society much more easily when we are not
189:43 DNA invested in the future, when we are not liable to our own
189:49 offspring to propel forward. So I think these are bad trends to
189:56 merge at the same time, that we have these falling birth rates and these
190:00 problems with coupling, and the fact that we need to keep our chin up about the future and be...
190:11 I mean, you don't even have to be excited about it in this literal
190:14 ecstatic sense that I'm portraying here, but I think you should be hopeful, and I think you should be working towards that hope, and I think
190:22 it's much easier to do so when you have literal skin in the game.
190:28 - We've been talking about AI agents. Let's talk about the human side. What
190:32 what do you love most about being a dad? - That's a good question. I think the overall love that expands from
190:46 humans that derive directly- ... from your lineage- ... is very difficult to communicate because I was actually not a
190:56 particularly big kids person prior to the arrival of my own.
191:04 And now they are the most interesting people in the world. A lot of the time. Not always. Sometimes they're also just assholes and so forth.
191:11 But there's been such a focus on, "Oh, kids are the worst. Look at all the things I can't do. I can't go out and drink.
191:19 I can't do this. I can't do that." Yeah, that's called sacrifice, and sacrifice is
191:23 meaningful when you're doing it for something worthwhile. And what could be more
191:26 worthwhile than literally the continuation of the human species?
191:31 And that you having a stake in that. Now, there's also just the sheer joys of watching a human
191:40 grow from a baby to a toddler to a teenager. My oldest has just become a teenager.
191:49 And I often talk to Jamie, my wife, about this, where the regret I would have of having missed that opportunity
192:01 on the last day would just be total, unlike anything else. So I do think it is far more
192:13 satisfying even in the moment that is normally portrayed. Again, as I
192:17 say, there's such a focus on all the ways they suck. "Oh, you don't get to
192:20 sleep," or whatever. "They're annoying. They're ungrateful." Yeah,
192:24 yeah, these things are all true, but the sense of meaning you get from it is unparalleled. And I'm excited about a lot of things. I love building
192:32 things, and I love hobbies, and I love all sorts of stuff. But
192:38 creating life with another human that you love is literally the peak
192:45 experience of being on the planet. - What do you think about them coming up in this world full of AI?
192:54 Is it totally diff- It feels like, maybe I just sound like an old man on
192:58 a porch, but it feels like a very different world. Y- So I grew up before the internet, but even the internet doesn't feel like as big
193:08 of a transformation as this. - No. But there's been bigger transformations. I mean, imagine
193:14 growing or being born in like 1880. Imagine seeing the First World War and the Second World War. Imagine seeing the
193:21 airplane, the radio- ... television. I mean, the transformation, and I think Peter Thiel makes this argument, that basically
193:30 nothing has happened in the physical world in quite a long time. We've been stagnant.
193:34 All the development has been in the digital realm. And as important as that is,
193:41 it's actually not as consequential as we like to believe it is.
193:45 Maybe AI will be that thing that is more consequential. It does seem to point that way. But there are other humans at other
193:52 times who have lived through- ... transitions that for them certainly would feel as momentous
193:58 as what we're living through. And I think having that sense of history, that
194:02 you're not that special in your sense of worries, in your anxieties. This was one of the great revelations of discovering
194:11 the stoic writings. You have these guys 2,500 years ago dealing with
194:18 very familiar dilemmas and challenges in their life, and recognizing that we're not so special in that regard, I think is a great
194:28 liberation. Because then you think, "Okay, well, we have whatever, 200,000 years of human development before us, and they
194:36 somehow made it through." - Yeah. And, um- - We will make it through.
194:40 - ... the thing that makes life worthwhile is basically the same today as it was-
194:45 - Yes. Yes. - ... 2,000 years ago. - Yes. Love, creation, creativity.
194:50 - Yeah. Same challenges. - I read this book The Fourth Turning.
194:55 You check that out. It talks about this notion of history working in cycles, and there are these,
195:03 phases to history, and they name them and so forth. And the specific theory is not as important as just realizing that
195:12 this is a... History is a circle, more so than just a straight line.
195:18 And we are repeating many of the same patterns over and over
195:21 again. They feel so unique to the moment. And whatever crisis we're dealing with right now feels so pressing and
195:29 important until we zoom back a little bit in history and realize that the last one also felt so pressing and important
195:36 to the people who lived through it. - Yeah. You're making me feel better. So I'm gonna travel across the
195:44 country and spend a lot of time with no access to the internet,
195:50 alone with my thoughts. - I think that's healthy, taking a break.
195:55 - It's okay. It's okay. - I took not a big break, a little break, break from Max this summer, three weeks.
196:02 - Yeah, you were... You disappeared off of- - I'm surprised you noticed.
196:08 - I went through withdrawals. No. Uh, you disappeared for, I don't know, a few weeks.
196:13 - About three weeks. - Three weeks? What happened? - Um- - This is after, after Llama.
196:21 - I've done that before. I just sometimes get the sense that X in particular
196:31 is the most addictive form of information, entertainment, connection that exists in my
196:39 life, and anyone that follow my X stream as I was on my flight over here would certainly go like, "That man is deeply
196:48 addicted- - Yeah - ... to the connection." And yes, guilty as charged,
196:53 which also means that occasionally I gotta cut it cold, because I
196:57 do think you can break your brain. And over the summer in particular,
197:03 for whatever reason, I had somehow managed to steer my feed,
197:07 my algorithm, towards too much politics, and not enough tech, and not optimism, and not enough building. And even if I
197:15 agreed with the politics and the angles that were being presented, I just didn't need- ... that much of it.
197:23 - Yeah. - And I think arriving at a place where you just go like, "Okay, that's enough."
197:30 Not that... Like, it doesn't have to be something bad. I would also say after eating probably two, maybe
197:37 three chocolate strawberries, I'd go like, "Okay, that's enough."
197:40 Like, it's not that I don't like the chocolate strawberries, it's just, like, I don't wanna eat
197:44 17 of them. And I think X and the modern algorithms are very good at feeding you
197:49 freaking 17 chocolate-covered strawberries. - I wish I could, uh-
197:53 - Sorry for doping you - ... control that. Like, 'cause there's... I would like some politics, 5%.
198:02 I would like for example, one of the things I don't like about the tech community is the snark,
198:06 but I like some snark- ... 'cause it's funny. But I want that to be like 10%.
198:11 - Yeah. Yeah. I want it just right. - Of my... Yeah, just right.
198:14 - The Goldilocks version of the algorithm. Well, the problem is the algorithm finds
198:18 out what you respond to- ... not what you say you like. It's all revealed preferences, and unfortunately, those revealed
198:23 preferences are not always pretty. I mean, we have this shadow, as Jung talks about, right? All these dark aspects of ourselves, and they
198:31 are expressed in how long you linger on a tweet, and what you like,
198:35 and what you don't like. And the algorithm holds up a mirror-
198:40 ... and it's not always pretty, and I just went, "Do you know what? I need a break."
198:45 - Mm-hmm. I've gotten to, I st- started staying off social media, and I've started s-
198:52 scraping X. Like, for example, in preparation for this conversation, I've scraped all of your tweets.
199:00 - That's actually a method I was using for a while. I was using a tool called,
199:06 I forget. It was an email tool that would summarize every day, and unfortunately, freaking X cut off
199:10 their access. Now to get access to the feed, you gotta pay some exorbitant-
199:16 ... sum that this startup couldn't pay. So I couldn't get it that way, and I actually thought, "Man, that's a
199:20 miss," because X has and reveals some of the most interesting thoughts.
199:27 Some of the most interesting people all the time. But it's like a damn slot machine,
199:31 and most of the time you don't hit the jackpot. You hit, like, something else.
199:37 And if I could just get it in sort of snack sizes, not like all of it-
199:44 ... it'd be nicer. But that's not the product. And I think this goes for all of
199:48 social media. It's all optimized towards maximal extraction,
199:52 and if we're making the parallels here to AI, that is the great fear, right?
199:56 That once there is a reward function, how much time can I get this human to
200:02 scroll away? AI and machine learning, before we called it AI, has become very, very good at delivering
200:11 just the stuff that you will find most enraging and most addictive.
200:15 - I think all, I think legitimately engagement maximizing social media is a huge harm to
200:23 society. It's creating this brave new world situation that I
200:27 hope is a temporary thing that people just solve. It's a technology problem.
200:31 - So I try to consciously flood the channel in the other direction, and this required a mental shift. I think
200:39 if you look at my tweets, like, I don't know, five, 10 years ago,
200:44 they tilted far more negative. If you look at my tweets now, I'd hope-
200:49 ... well, I guess we can do a sentiment analysis and see if it's actually true.
200:52 - No, it's more good vibes. - But I hope to, that- - It's good vibes.
200:55 - ... it's more, it's more positive. It's more good vibes. It's more like, "Here's a bunch of cool stuff I found."
200:59 "Here's what I'm excited about." "Here's what I'm building. Here's some encouragement to other people who are discovering things."
201:07 I also accept that that is just not how human brains are built for maximum,
201:13 engagement. Like, there is a, whatever, there's... I literally, I
201:17 think you can quantify that negative sentiments find, what is it, 50% more traction or something? And eventually that just crowds things out.
201:25 So that's why so much of the feed is that negative stuff, because we
201:29 respond to it. But you can make a conscious effort to not do that.
201:33 This is one of the reasons I've liked TikTok so much. I only use
201:37 TikTok when I travel, and this actually was the first travel where I didn't use TikTok, and
201:43 part of the reason was that even though the internet had high download, it was kind of crap anyway. It wasn't Starlink, so I couldn't actually use
201:51 TikTok on the, on the plane, and I was too engrossed with the agents. But normally when
201:55 I travel, I watch TikTok. And there it's such a marshmallow test.
202:03 A video will scroll down and, like, it'll activate some sort of
202:07 primal instincts of rage or lust or whatever, and if you linger on it, you don't get the marshmallow of science,
202:15 excitement, art, music, all these other things. You have to be disciplined-
202:21 - Yep. Fast. - ... to instantly scroll away- ... such that the algorithm doesn't pick up your real preferences.
202:27 And I find that just to, at the meta game of TikTok- ... to be pretty interesting, and then also just fascinating in themself. Again, as we talked
202:35 about, there are a lot of really funny, really talented people in the world that you would never, ever have heard of if
202:43 it hadn't been for these platforms. - Yep. Who do you think is gonna win the,
202:51 the race to, uh, AGI in this whole process? We got Google, which is
202:57 a big surprise to me, to a lot of people. - I thought they had a good comeback for a second, and then I don't hear anything
203:04 about it anymore. - Yeah, it's quite, I mean- - But I've been wrong about this before. I mean, I think in February I had a tweet about like, "Oh, all I
203:10 need is Kimi K 2.5." And in that brief moment, that was true.
203:17 When I was instructing the agent, I was like, "Oh my God, Kimi's just as good as
203:21 these other models." But then the game moved forward and- - the agents got smarter, and I was no longer satisfied just instructing them, and
203:29 I haven't used Kimi a whole lot recently because the frontier models have gotten so
203:33 good. And somehow Anthropic has been able to, for the most part, stay on top of the programming world.
203:42 - Yes. Yes. - I mean a lot of people argue with Codex being it's-
203:46 - No, I don't think there's that much of an argument at the moment. Codex is very good.
203:49 I use it all the time. It's my favorite checker. It keeps finding stuff. It's great.
203:55 But one of the things I've found that I like so much about the current Claude models is
203:59 they're really good writers out of the box. GPT is a terrible writer out of the box. Like, the pull requests you
204:06 get on a blank context that you're asking it to open, terrible.
204:12 Like, really awful writing. The pull requests and descriptions and commit messages and so forth I get
204:18 out of Claude models- ... even though occasionally they can be a little verbose on code comments and you gotta tell them
204:24 to be succinct. But the writing style is really good. To the point that for a while I was just having it reply
204:31 on pull requests as me, because I was just using my own account. I was like,
204:36 "That looks a lot like me, that, like, I could have written that." But then I went like, "Eh, do you know what?
204:39 That's disingenuous." So no, whenever I have an agent post on my behalf, I always
204:43 have it sign itself like whatever, Claude on behalf of DHH. But they are still ahead.
204:51 They have the best harness too, in my opinion. Claude Code is still
204:55 better. I really like OpenCode too, and I really like some features that they have,
204:59 but overall I was actually sort of weirdly, I don't know if I was proven wrong, but
205:07 when Anthropic cut off all the other harnesses from using the subscriptions-
205:12 ... that Open... You can no longer use your Claude subscription in OpenCode. I was kinda pissed
205:16 because it felt like it was a protective, protectionist move.
205:20 "Oh, we gotta lock you in," and of course it played into the fact
205:24 that Claude out of the box still refuses to read agents.md. It refuses to
205:31 read .agent/skills. It is .Claude skills. It is Claude MD, and it just feels so petty-
205:41 ... that we now have, we have to put these in directions. So all my projects, they have a Claude MD,
205:45 and all it includes is a pointer to the agents MD, right? And I just went like, well, if these things is true, true, if Anthropic
205:52 is willing to do these petty paper cuts just to be, annoying, it's probably also true that they're cutting off OpenCode from the
205:59 subscription just to be dicks about it. And maybe that's also true.
206:04 But I found that using Claude in Claude Code is actually a good example.
206:11 So I was like, all right, I'm annoyed. I wish it wasn't so, but
206:17 this is the other thing I have and, and social media is so good at this:
206:21 Some company does something you don't like, right? Now suddenly you have to swear off
206:25 everything. Why are you... Someone just asked me, "Why are you still using Claude?" I
206:29 had another example where, I really didn't like what Claude was saying about some writing I had asked
206:35 it to do before. - Oh, yeah, you got there-- - I had an essay it didn't want to translate into Italian--
206:40 - ... you had a blog post. That's right. - The essence of the post that it felt like, well, I'm not gonna translate that because I don't agree with
206:45 what you've written. Which to me was straight out of Space Odyssey, right? HAL telling me,
206:50 "I'm sorry, Dave. I can't do that, Dave." Open the pod bay. Translate my essay to Italian. And it wouldn't do it. And I just thought like,
207:00 well, that's kind of terrifying. But then I also went, well, why am I asking Claude to translate
207:08 my essay into Italian? I never asked Claude for that because Claude just does my code.
207:12 So I could also just accept that maybe Anthropic is run by a bunch of people I don't share much politics with.
207:20 And they don't want their agents to be used for that, and I think that kinda sucks, but that doesn't
207:24 mean it, I have to stop using it for generating code. And I've thought, and I've tried to do that with people too, become better at
207:33 tolerating the fact that not only will I not agree with everything someone has to say,
207:41 I don't... I can even be mad about certain positions or things they say and still go like, "Well, you're still an interesting person,
207:48 so I'm still gonna follow you." And I've actually kinda gone back on a number of people where I'm like, oh man, that person really pissed
207:56 me off because they took this point or another point, and went like, do you know what? That's stupid.
208:00 I should be able to accept the full spectrum of someone. I don't
208:04 have to like all the shades of it. I could just go like, okay, well, on that topic we either disagree or, even
208:10 worse than that, I think you're a fucking idiot. And on another topic, you're a smart and
208:14 insightful person. Because first of all, of course, it's a little self-serving.
208:18 I know for a fact that this is how lots of people feel about me. Like,
208:22 oh, if he could just stick to talking about these topics, it would be so much
208:26 easier on my feed because otherwise he'll say these things I disagree with,
208:30 and that annoys me greatly. I'm like, do you know what? Human beings are jagged like that, and we're
208:37 not fully compatible on all the levels. Nor should we be. The world would be so fucking boring if we
208:43 constantly sought 100% compatibility in our opinions and in our politics. And when it comes to my code generator,
208:51 which is what Claude is, I don't need it to share all my politics.
208:56 - And therefore it's nice to have competition. - And it's also nice to have open weight models. So this is, again,
209:02 ironic that we're getting this out of China. In fact, I did this test right
209:06 after the issue with the essay that couldn't get translated into Italian. I
209:12 asked Kimi K27, to be fair, through open code, so not the Kimi harness,
209:20 and inference through Fireworks, so an American inference. I asked it about Tiananmen Square.
209:25 Like, that's usually the test, right? Like, all the Chinese models, they don't wanna talk about that moment in 1989.
209:32 That was actually how I, that was how I posted. I was like "What happened
209:36 in China in 1989?" That was the prompt. And K25 7 on Fireworks just
209:43 answered, like, extremely bluntly- ... With a description and depiction of what had transpired-
209:49 ... to a point that, like, there's no way that that would've been kosher with the Chinese censors,
209:53 right? And I just thought, like, isn't that ironic? I'm running a Chinese open
209:59 weight model that I can tell or I can ask about 1989 and Tiananmen Square,
210:06 and in America, I can't get a frontier model to translate a marginally controversial essay about immigration into Italian.
210:16 - Mm-hmm. I don't know, you and I might disagree about this actually. Um, I
210:21 was really deeply troubled that the US government banned a model
210:27 or pressured Claude and then- - It's still very muddy what exactly transpired. Was this
210:35 sort of petty retaliation for whatever the Department of War? Was this simply
210:42 Anthropic themselves going like, "Oh my God, we discovered a cyber weapon so
210:46 powerful that we can't let anyone have it"? Well, okay you can't be surprised then if the government wants a say on a cyber
210:54 weapon so powerful it can- - The- - ... I don't know, rob banks
210:57 - ... The problem I have with that whole thing is, no matter what happened, it sets
211:01 a precedent. And so government, when you give it that power, it's gonna start abusing it. I
211:08 could see for political means pressuring a model. You know, if it's a Republican president pressuring models to be, to
211:16 censor more, uh- - Sure - ... uh Democratic views and- - Right
211:20 - ... For vice, and vice versa - That's why we should insist on the fact that this stuff,
211:25 like, I'm sorry, Dave, I can't translate your essay into Italian.
211:28 - Yeah, no. - That's abo- that's way beyond the line. Like, whatever the
211:36 sort of uh, root of it is, you have to be a tool first. Now, if the tool is, "Hey, can you tell me how to make anthrax?" Okay, fine.
211:44 Right? But within the jurisdiction you're in, and certainly in America,
211:48 free spe- speech is quite absolute. Very, very few limits on free speech, and that's one of the reasons why the
211:56 country's so great. And yeah, I don't like that. But on the other hand, I'm also a big fan of market
212:03 competition. So while I don't like this about Anthropic models, I
212:06 appreciate that I could go next door to Groq and get my task completed.
212:12 - Do you think, outside of blog posts, do you think, There's some legitimacy, like can you steal Mandarion's
212:18 case that we should be careful about the release of these models?
212:25 - Is AI safety important? Yes. Are-- do these models have capabilities that could be seriously harmful in the
212:32 production of biological weapons or otherwise? Yes. Is it fair to have some
212:38 ground rules on that? Yes. But then don't squander it by denying the translation of an essay.
212:44 - Okay. - Because then you erase the whole thing and you- ... bias everyone towards thinking like, every guardrail you put up is gonna be
212:54 bullshit. - Yeah. The security piece of it is tricky because it's so good at
213:00 finding vulnerabilities. - Shockingly an- I was about to say annoyingly, but that's not actually true because what it's finding, at
213:08 least when you're using in defensive ways, is it's finding real issues that a clever
213:12 hacker could exploit. But nothing has caused as much stress, to be fair, at the technical team at 37signals, like the
213:21 fact that these latest batches of models are exceptionally good at
213:25 finding these issues. So suddenly there's just this seemingly endless
213:29 parade of security vulnerabilities that need to be patched and
213:33 fixed and sorted and so on. And the end result is we end up with vastly more secure systems. But the road there is pretty rocky.
213:41 - Do you think it's-- 'cause you said seemingly endless, do you think it's
213:45 a finite set, like we can get to the bottom of it? So just make the systems much, much more secure, where we get
213:53 back to this semi-stable state where mostly it's very difficult to
213:57 find security vulnerabilities. - I do think that all the aggressive tendencies of these models, or not even tendencies, uses of these
214:06 models are also the defensive ones. So if you're good at finding an
214:10 exploit for exploits, you're also good at finding exploits for defense.
214:14 So in that sense, it does feel like we're spy versus spy here, black and white, but with the same
214:21 capabilities and with the same actions. But while we still have to steer
214:25 these agents somewhat and verify and vet the work, it's creating a lot of
214:31 intense moments for security teams around the world. And of course, the more concerning part is all the teams that are not stressed, like they
214:44 just don't know, right? Like if you're a team right now and you're not dealing
214:48 with a bunch of patches, it's just because you're blind, and it
214:52 means that your adversaries quite probably have access to techniques that can get
214:58 into your system. - And we should also say one of the big security vulnerabilities in the world is
215:03 the human side. You know, social engineering, and that's extremely concerning with increasing the
215:09 intelligent AI systems, you know, being able to write emails and phone calls and faking
215:14 voices, thinking about folks less tech-savvy. - This stuff is gonna come, and it's gonna come in bulk. But again, I
215:23 choose to be optimistic about it that- ... whatever these models can do aggressively, they can also do it defensively.
215:29 - Uh, so like you said, recently released Omarchy 4, Quattro.
215:36 And you said it's one of the greatest software releases in my professional career,
215:41 which is a hell of a statement. - It's a hell of a milepost for what's possible at the moment.
215:49 And the reason I said it like that is not that I haven't worked on anything else of note. I've worked on a lot of things of note, a lot of things that I'm very
215:59 proud of. There are very few other projects that I have worked on
216:03 that in the amount of time that I've spent on Quattro, which is really three months,
216:09 I have been able to accomplish as much, fulfill as many hopes, desires, wishes. It truly is like the metaphor we used earlier about
216:21 finding the genie in the bottle- ... and having an unlimited number of wishes, more or
216:28 less, for getting everything I wanted out of a system. So it's strange because it's also perhaps
216:36 the major software release that I've worked on the least in terms of lines of code. I have the least number of
216:45 personally hand-chiseled lines of code in Quattro than I've had in
216:49 any major release I've ever done. And yet, I am insanely proud of what we've accomplished.
216:57 I would not have guessed that that was how it was gonna play out, that
217:01 you could divorce the sense of achievement and make it be like a team victory.
217:08 where other releases, I was very happy for my personal contribution because, like, I was the one
217:14 chiseling out the lines of code, and then I would push it out there, and then later on, of course,
217:18 these projects, Rails in particular, would grow into be huge open source projects with thousands and thousands of contributors, and then it certainly was a team effort.
217:26 But in the beginning, all code bases that I've almost all of them that I've worked on, I started, like, just myself.
217:33 And Quattro started as a team effort right from the get-go, and
217:36 where I was just the coach. I was telling us-- I was telling the agents where to go,
217:42 and we would go. So to be able to find satisfaction in that actually
217:47 gives me great hope for retirement in some way, that there is a phase after playing that you can enjoy. And
217:58 I know that's not easy. You look at sports and the number of professional athletes
218:01 who really struggle to cope with the fact that their career is over
218:05 and have to find a new identity as someone who doesn't play football or race
218:10 Formula 1 cars. I mean, it's more the rule than the exception.
218:14 So I wasn't so sure about myself. I wasn't so sure about having to give up or not even having, wanting to give it up.
218:24 Wanting to play another level of the game and finding not just equal satisfaction, but more satisfaction in it.
218:34 - Do you... You've been pretty optimistic about Linux. Do you think
218:37 legit Linux can increase its adoption? 'Cause, you know, for many years the meme is, you know-
218:45 - Linux on the desktop- ... every year. Next year again. - Yeah. But it seems like, at least the case you're making, the energy you're
218:52 putting into it, you could see a vision where it takes over because agents love Linux.
218:58 - Not only can I see it, I find it to be the most probable outcome at this
219:01 point. It is simply too well-suited for the moment. And again, as we talked about, it's a great irony
219:09 that all the flaws of Linux, the arcane config files, all the strange error
219:16 messages and so on should just so happen to be the perfect thing
219:20 for an agentic operating system. But so it is, and we should rejoice that
219:27 fate has this irony, and just laugh at it. How... It's so funny to me that all the advantages that Apple had for the longest time-
219:36 ... that they produced this really locked down, curated experience should become their greatest drawback now. If you're into
219:49 agents and development and working with all this stuff, the Mac is just a hostile place to be. It just has walls all over the place.
219:59 And Linux is simply open through and through. And it's also interesting because that was not
220:07 obvious either. If you look at the way a lot of the Linux communities, open
220:11 source communities, have reacted to AI, it is not universal love.
220:16 I would argue that the majority of them are actually, if not
220:20 skeptical, then outright hostile- ... to AI. Now, the saving grace is that the BDFL himself, Linus-
220:29 ... Torvalds, just wrote a few weeks ago that he actually welcomes AI.
220:35 - That's right. - He wants to steer it. He wants to make sure it's good and whatever, but I- the
220:39 line was something if you think, uh, Linux is an anti-AI project, think again, and you should just do the open source thing and fork it,
220:47 because we're gonna use AI. And you see these graphs of the number of AI
220:52 contributions going into the kernel, and it's a parabolic curve.
220:56 So Linux is leaning hard into this. And, It actually in other ways it was surprising that it didn't. Linux conquered
221:05 everything else. All the AI infrastructure that everyone runs off, it's
221:09 all running on Linux. All the systems, all the servers, everything is
221:15 Linux. So it was kind of a curiosity that the desktop and the personal computers we were using really hadn't been
221:22 captured, but clearly it was just waiting for this moment. Linux
221:26 spent the time from '91 to now waiting for agents to fully flourish as an end
221:34 user operating system. - I hope it takes over. I mean, it's perfect. It's perfect for agents,
221:39 but it's hard for people to switch. - I think it's hard for people to switch when there's not a compelling reason to do so.
221:45 This was one of the driving design goals for Omarchy. It was not gonna be Timo Windows or
221:52 Timo Mac. It was not just gonna be a cheap copy where we try to make it as familiar as possible and then kinda
222:00 worse. I mean, if I was being unkind, I would say that's what
222:04 Ubuntu has tried to do. And there's something in that, or there was something in that
222:08 at least, where if you make something really familiar, you can get some
222:12 people who don't wanna learn something new, and then maybe they'll be
222:16 compelled about your, whatever, free as in speech, not as in beer stuff. And it just didn't pan
222:23 out that way. People just didn't care. They just wanted whatever was better.
222:27 And I think this is why Linux now has the opportunity to win. Because
222:34 as an agentic operating system, as a malleable operating system, you
222:38 can tailor to your desires, it is unparalleled. And that is so compelling, and I can see it
222:46 now. Just since the release of Quattro, the amount of people who have embraced that aspect of it, the malleability,
222:53 started making their own things, sharing screenshot of it, and being absolute just,
223:00 Flooded with positive emotions and dopamine- ... is, that's the product market fit signal, right?
223:07 Because what a lot of these individuals are experienced, they're
223:11 experiencing the very high of what the best of being a programmer is like.
223:16 You tell the computer these arcane commands, this rigid logical structure, and it produces what
223:24 you want. Now suddenly, you sit down and in plain English ramble for 20 minutes. "Oh, it'd be great if it did that. Actually,
223:32 no, that's not right. Let us..." And out comes software. Out comes an operating system shaped in your image.
223:40 I think that is one of those experiences that if you have it, you've... If you feel that
223:49 power in your hands, it's very difficult to go back. - And I think if you get it right, it's one of those things that will seem obvious in
223:57 retrospect. - Yes. This is how computers were always meant to be. This is how computers started.
224:03 The Commodore 64, you turn it on. First of all, it boots in a second. Amazing.
224:07 Second of all, it boots straight into BASIC. It was the malleable computer from the get-go. It just so happened that you
224:15 needed to know hieroglyphs- ... to be able to take advantage of that. Now we've translated it. Agents
224:21 have given us the Rosetta Stone, and you can simply just speak
224:25 your desires and- And so they are - Beautifully put. Since you mentioned Linus Torvalds, and since you're working on
224:32 a Linux distro, you get to have contact with the kernel. What can you say about the genius of this one guy that helped create and grow this
224:43 ecosystem? - The perseverance and the commitment and the longevity is truly remarkable.
224:52 Again, Linus started in '91. He hasn't stopped. I don't know, he's probably taken some breaks along the way.
224:59 It doesn't seem obvious that there was that many of them. It seems like he just really
225:03 likes the flow of doing kernel development. He really likes steering where this 40 million line code base is
225:11 going. He really likes aggregating the net capacity of hundreds of thousands of contributors, and
225:20 we, as beneficiaries of that, must just sit back and marvel.
225:24 I mean, at this point, Linux is a system. Like, if Linus stepped out for a while, it would continue to,
225:31 to go, but it's still one of those protect this man at all cost.
225:36 - And he's still open-minded enough to be able to evolve and- ... accept AI into this.
225:41 - Yes, not only AI, they're letting Rust into the kernel too. I know
225:44 that was a controversial one too. He has this capacity to look at new information-
225:52 ... and arrive at different conclusions. - What do you think about his style of communication that is akin in some ways
225:58 to your own, where he's, can be sometimes a little spicy? - I think the world has gotten too bland. We need a little spice.
226:07 - Amen, amen to that. - Now, I think there's a fine line between being spicy and mean.
226:14 But it has to be in a good spirit. You can be harsh, let's put it that way. If you're trying to impart a
226:22 lesson to someone you believe could actually learn it- ... I actually don't think it's always the worst thing in the world to be a little harsh.
226:31 Sometimes that lesson sticks, and I think if you talk to people who've
226:35 worked for difficult leaders, Elon, Steve Jobs, maybe even Bill Gates,
226:43 many of them afterwards would go like, "These were some very difficult years-
226:48 ... but also some of the very best." "They got the best out of me." Now,
226:53 I don't think it's my style. I mean, maybe people I interact with feel differently,
226:57 but in personal interactions, I try and get to the outcome and the lesson in a less abrasive way.
227:06 But I also think we need some spice in this world. We need unreasonable people,
227:12 and Linus has justification to be unreasonable. In many instances, the
227:18 weight of the world is literally on that man's shoulders. The Linux kernel runs the entire civilized society. If the Linux kernel suddenly
227:29 disappeared tomorrow, nothing would work. So if he's not harsh when that's at stake or at risk, like when ever would you
227:41 be harsh? So I feel sometimes also there's a proportionality to the size of your mission. This is also
227:48 why I think these stories about Elon sometimes perhaps being a little bit difficult
227:53 are easier to rationalize. It's not even that you condone. You can separate those two things and go like, "Do you know what? It'd probably be...
228:01 Maybe he would have an even an easier time getting to his objectives if
228:05 Linus or Elon or anyone else in that position was a little nicer." But also who am I to judge that?
228:14 Like, when you have the weight of that on your shoulders, either firing
228:18 little rockets at Mars and transitioning all of us to electric cars
228:22 and whatever, or you're responsible for the damn Linux kernel that
228:26 runs billions of devices. Do you know what? It is not my place
228:29 to call a sort of a code of conduct on you. Again, is there a line? There's a line for everything. There, there's... Something can't...
228:38 But, like, a few harsh words on a mailing list? No. It's actually
228:42 a beacon in some regards that, like, there are standards- ... and if you fall below those standards at, when the stakes are as high as the
228:50 Linux kernel, ah, you might get ridiculed in public. That's a
228:53 liability you should be willing to endure if you participate in this project.
228:58 - Yeah, for high impact things like that, it's probably good to
229:01 prioritize the pursuit of excellence versus the pursuit of niceness.
229:07 - And even more to the fact you're just not gonna get those kinds of individuals being fully rounded-
229:13 ... plush, always the perfectly polite people. I mean, Elon has this line, "Did you also think I was gonna be a nice..."
229:22 Or, or what is it? The - A normal dude or some- - ... normal chill dude.
229:26 What? No, of course you're not gonna be a normal chill dude. What, what would the world
229:30 gain if we got one more normal chill dude and we had to trade Elon? Holy fuck.
229:37 - I also like diversity of people. - Yes. - I like, I like nice people. I like assholes. I like,
229:43 I like different ideologies being well-represented, and that's why freedom of speech
229:47 works. They, they clash, and we figure stuff out. - Yes. - There's another guy. What do you think about PewDiePie using Linux? He's an Arch
229:55 person, no? - He's an Arch person. I... He's got the most gorgeous arc. So
230:02 the world's biggest streamer for a while I remember my kids watching some of his Minecraft videos, and then
230:10 he just gets fed up with that, does the wholesome thing. Marries-
230:14 - Yeah, family man now - ... a family man, kid, moves to Japan, and then
230:19 gets so hardcore into- ... first Linux, then Arch, then Ricing.
230:25 Did you see his, uh- - I did. - ... rice- ... that was Chernobyl rice?
230:28 - Yep. - Absolutely best tier incredible stuff. And then he also becomes a goddamn AI man of the moment,
230:37 building all of these systems, and you're just going like, that... What an inspiring story of evolution. Like,
230:45 you can go from being funny guy on Minecraft streams to, "Okay, this is my life now. I'm building AI
230:52 clusters and doing the Council of AIs," and we all get to spectate? I mean, what a treasure.
230:58 - So I guess, I mean, he's a pretty good embodiment of what the future
231:02 of software, or what the future of building looks like, right? 'Cause he's a
231:06 non-programmer technically- ... becoming a programmer. - Hugely inspiring, and should be hugely inspiring
231:13 to others who sit in that moment like, "Well, I don't know quite how
231:17 to program. I don't know this and the other thing." Okay, but if PewDiePie
231:21 can build the Council of AIs with his bespoke, um, hardware here, then maybe you can also start on things. And I think this is
231:30 actually an important general point, that we need role models.
231:34 We need people to inspire others, to push further, and reveal that the perceived boundaries, they're not as fixed as you think
231:42 they are. And every one of those people, they started out in a situation not
231:50 too dissimilar from yours. Again, that doesn't mean everyone is gonna be PewDiePie
231:54 or Elon or Linus. There is not an equal distribution of talent or even
232:01 intelligence. And we can come to terms with that and still be inspired by these people. Like,
232:08 in the physical realm, I feel like we don't have any problems with that
232:12 usually. Like, oh, someone's really good at football, they're really good at basketball, they're really good at something, and we go like, "Oh
232:16 my, that's amazing. I mean, I wish I was, but I'm never going to be."
232:20 Sometimes I feel like we struggle a little more when it comes to the intellectual realm.
232:23 - Well, for a lot of people, you're an inspiration. The way you're doing with
232:27 Omakub's... It just shows that you can dream big and really build. It's great.
232:32 - I do like that part of it, and then I accept the part that there's also a million people who think I'm
232:36 the biggest idiot on the earth and I don't actually know anything, and Omakub's just
232:40 a bunch of dot files, and- - Do you have some critics on the Omakub side?
232:43 - Oh my God, yes. - Oh, wow. - And a lot of it, I mean, is as often happens when someone
232:50 comes in from the outside. I mean, I've been using Linux for two and a half years.
232:55 That's not very long. Plenty of people in that community who's literally been running Linux since it was
233:00 hard and difficult, and you had to walk uphill both direction against the
233:04 wind in the snow barefoot. And I think there is a certain kind of nerd-
233:09 ... who's threatened when the community expands, and it suddenly allows more in. And do you know what? I actually...
233:17 I can't begrudge that fully. I think it's fair for certain nerds to feel
233:21 like, "That's my space." Like, now you're changing it into another space. I don't wanna do that. Like, I don't wanna
233:27 change the Arch space at all. Like, I don't really think of Omarchy as being directly part of the same community.
233:34 Because the people who were attracted to Arch in the first place were the ones who were
233:38 attracted to building everything by hand themselves. Doing it the hard way because they like to do it hard,
233:45 and I'm building the polar opposite of that. So I just wish that those people could go like, "Okay, well, this is not for me.
233:53 I wanna build it the hard way. I wanna put in 500 hours to rising my own
233:57 Arch distribution," and I think you should. I mean, actually, I said... I mean, I
234:01 did it. This is what Omarchy is, me pouring in literally, at this point, I don't know,
234:05 3,000 hours into the, into this distro. You should do that if you're so inclined. I think it's wonderful. But also do realize
234:12 that we can share the same technical underpinnings and then operate two very separate-
234:19 ... Communities with very different aims, goals, aesthetics, morals even,
234:25 and coexist, and it can be beautiful. And I mean, that's not always what I get. Sometimes I get the other thing. But at this point I've been in the game long
234:33 enough to not just be at peace with it, but actually be welcoming of it.
234:40 I take it as part of a barometer that things are working. When
234:45 I'm working on something that works, I usually get some people like it,
234:49 and then I get this balance of the universe that we talked about last time, where an
234:53 equal but opposing force must be present on the other side of the scale to hate
234:57 upon it. And therefore, I at this point just smile a little. Like, it's okay.
235:02 - I mean, that said, uh, this, it's quite a popular distribution,
235:06 uh, Omarchy, at this point. - And do you know what's interesting about that is, like, this is the second attempt.
235:11 So my first attempt was Omakub. And I was really happy with that, and I put it out there, and there it found a
235:17 community of maybe a few thousand. Like, and that was that, and then it just sort of
235:22 petered out because the level of ambition couldn't attract more.
235:27 And then the second time, because I didn't fucking stop, right?
235:30 Like, I just kept going. I just kept building. Then I put out Omarchy, the first version.
235:34 Again, quite niche because it was this... It was like Git checkout. It was a little bit
235:38 difficult. You had to set up Arch first. I was... Well, I would talk you through how to set up Arch,
235:42 and then you could do the Omarchy thing, right? And then we got an ISO that meant you
235:46 could just install the whole thing by yourself. And then we just fixed all the problems. This is the other thing. If you keep
235:53 going, eventually the thing will just work. This was the charge
235:57 against Linux itself for the longest time. Well, you install Linux and then this doesn't
236:01 work, and my speakers don't work, and my, Trackpad doesn't work. Yeah, okay, but just leave Linus at it for like 30 years.
236:09 He'll get to it. He'll get to all of it. And at this point I find
236:12 hilarious that on, I would probably argue the majority of computers,
236:17 it's wiped. You install Linux, all of it works. You install Windows, good luck hunting down the drivers you need to
236:24 get that piece of hardware working. Because they have very different philosophies. The reason Linux is 40 million lines of code is Linus
236:32 just puts all the drivers in the kernel. So it has everything out of the box,
236:36 and Windows doesn't. A different, different approach to it, right? So
236:41 sometimes the things that look like a toy, things that look broken at first glance,
236:47 they are, right? They're for people who like having fun. We call those
236:51 early adopters. They play with the toys, and then the toys evolve,
236:55 and then they get better, and then they get better. And at some point, they're so damn good
236:59 that early adopters feel like they're invited to the party. And if the early adopters keep chipping away at it too, eventually you
237:07 get to the late majority, and then you win. And we are well on the way with Omarchy towards that destination.
237:16 Like the growth that the distro has seen, not just since Quattro, but since version three, it's just like, you
237:23 know, the standard hockey stick curve. You go like, there's nothing, there's nothing, there's nothing.
237:26 You're just toiling away. You're just improving, you're improving, improving.
237:29 And then you hit this magic inflection point, and it's impossible to predict when that's gonna
237:33 happen. Same thing with startups. When is it gonna happen? When are you gonna find that product market fit?
237:36 Then suddenly it's there, and then it goes, just goes and off the rocket ship you go.
237:42 - Well, this kind of vision of making it agent first, of making it-
237:45 - I do think that's the... We were already on the up, and then it changed
237:49 to vertical. Because the other thing, and this was what I learned with
237:53 Omakub and Omarchy first too. Omakub was stylish, nice, but familiar, is we're still using a
238:02 conventional desktop metaphor. You were dragging your windows around. And
238:06 I thought at the time, as I think lots of people do in the Linux
238:10 community, like this is what we need. And then I just go like, okay, well,
238:14 for whatever re- let me do the nerdy thing for a while. I'll build Omarchy, and maybe it'll just be for me, because I mean,
238:18 it seems like there's about five people who like tiling window managers. That's not
238:22 true. But as a percentage of total computer users, it's a tiny, tiny minuscule part. And then I go like, well, this is amazing. Totally different,
238:30 but amazing. I put that out. Instantly, Omarchy had far greater traction than Omakub ever did. Because it was not the same.
238:39 Because it was totally different, and it presented a different vision for what a computer could be and how it
238:42 could feel like, and that attracts people. We are now getting
238:46 sort of the exponential of that. As I said, a lot of people in the Linux community are at best skeptical,
238:53 at worst hostile to AI and to agents. Omarchy is the first distribution, at least that I've seen on a scale
239:00 that matters, that just goes like, "Nope, we're really into it."
239:04 I mean, you can still use Omarchy. You don't have to use an agent. If you dismiss that first notification
239:08 to set up a default agent, you're not gonna get any of the stuff we talked about. You're not gonna get the crash
239:12 watcher. You're not gonna get any of this stuff. You're not gonna get... You can hand chisel
239:16 all your plugins. Totally possible, right? But the reason people are excited is because
239:22 I don't have any reservation about saying, "This is where we're going. I think the
239:26 future of the personal computer is the malleable computer, is the agentic
239:29 computer. I'm gonna go all in on that." And then anyone who's excited
239:33 about a similar trajectory for computers, come along. Plenty of room in the car. In fact, one of the things I pride myself on
239:46 is not holding a grudge to people who come around. I had a bunch of people just last week going like, "Yeah, I used to think
239:53 Omarchy was kind of shit. That was just this crappy-" "... dot files collection. And now I tried Quattro and
240:01 awesome. I like it now." That to me, I live for the long argument. I live for this argument where you plant the seed like two years ago,
240:09 and then two years later they go, "Goddammit, this son of a bitch was right."
240:14 - Yeah. You do realize if you continue to be as successful as you are,
240:18 um, OpenAI and Anthropic are gonna roll in and try to do their operating system
240:24 or to buy, like with Peter with OpenClaw, there'd be a gigantic check.
240:31 - Yeah. I think the good thing about my current position is I really don't need the money. So I get to build these things purely
240:42 for my own enjoyment and purity of vision, which has this weird quality where sometimes when you
240:50 stop caring about what everyone else thinks or wants or whatever, and
240:54 you just pursue this singular vision, it ends up becoming far more appealing.
240:58 That it's when you try to do, oh, I should do this because these people want that, or I
241:02 should do this and this and so on, you start watering the thing down and suddenly it's a
241:05 bland ball of nothing. So I mean, again, the thing too here is directionally, I just want computers
241:14 to be this way. So if Omarchy should end up being just a, a little footnote in history that it was part of this early movement of the agentic
241:22 OS and the malleable computer, that's also okay. I'm completely at peace with that. As long as I get to have a computer that
241:29 is as fun to work with as Omarchy- ... that's great. I've had the same thing with Ruby and with Rails. From very
241:36 early I went, do you know what? I love Ruby. It's such a great programming language.
241:39 But if a better programming language comes along, I, me, I'm going to use it.
241:45 If a better Rails comes along, I'm gonna use it. And I think,
241:51 and I love Ruby, and I will continue to write Ruby even just for the sheer fun of it, like I
241:55 would go ride a horse. I have in the stables, even though I have a Model Y in the
241:59 garage. The language now is English. It's a cliche, but it's also true. Like I've been programming in
242:09 English for the last three months. I've been reading a lot of code, but I'm programming in English. I'm telling the computer
242:15 what to do, and I'm using natural language, and it is shockingly even more delightful. If there is one
242:24 programming language more beautiful than Ruby, it is the English language.
242:28 - And now we get to carry it into the natural evolution of human civilization, which is our cyborg future.
242:36 - Beautiful. It's beautiful. And I mean, the reason I say this in part is I always loved
242:40 writing. I always just loved the English language- ... for the sheer beauty of it for the sheer intricacy, for the depth of it. I mean,
242:48 Ruby is a very expressive programming language, but it can't hold a candle to English.
242:56 I mean, all the poetry and literature in the world that's been
243:00 expressed through the English language. I mean, I like a beautiful
243:04 code poem, but I mean, the real deal in English is just on a different level.
243:10 - And I actually find this is a nuanced point to express, and I'll probably
243:14 fail expressing it. But when you prompt a system and you're over-specific,
243:22 it will listen to you too carefully- ... and follow the instructions. And I-- But nevertheless, you have to
243:30 express a concept. And guess what? The human lang- human language, this is
243:34 what poetry is about. I find strategic use of ambiguity. Like, if you write a love poem saying, "I love you," like,
243:43 very cliché thing, it's not as powerful as something- ... more in metaphor and so on. And I find some version of that when I'm
243:54 prompting a system, when I'm describing a design, is actually really effective because you still want to convey some sense of style-
244:05 ... that you're trying to encourage a system to use, but not overexplain.
244:11 So don't talk to it like a like a robot. Talk to it like you, like you would write a poem.
244:18 - Like you were serenading. - Yeah. And that's where, like, the power of natural language is,
244:24 that you can, through ambiguity, still carry a lot of meaning without overspecifying. And the, the intelligent
244:34 entity on the other side, through interpretation, can, like, load it all
244:38 in, integrate it in a way that- ... you can actually convey the high bandwidth information that's
244:44 not directly in the words- ... but in the words, given the deeper intelligence of the system. So like the-
244:52 ... the ambiguity pulls out more intelligence, I guess, I'm trying to
244:56 somehow express. - I think that's exactly spot on, and this is the fundamental
245:00 misunderstanding that a lot of programmers have of AI, is that they wish it was
245:04 deterministic. No, no, no. Temperature is the most beautiful part of the AI setup. The fact that it is not
245:12 deterministic, the fact that creativity requires little tweaks in the road, that the human brain,
245:20 if it was perfectly deterministic, would not be the creative brain that it
245:24 is. And the main charge against AI both-- It's funny. It's a contradiction. There's both the charge that
245:32 it's not deterministic and therefore bad, and also that it is not creative. It's one or the other, bro.
245:39 Either it's non-deterministic and therefore creative, therefore, to some degree random, or it's... or it's not creative.
245:51 So we, we can-- Both of those charges can't be true at the same time,
245:55 and I have fully come to embrace the fact that the same prompt won't
245:59 produce the same response every time. You can't step at the same river twice, and it is the
246:03 most beautiful part of the whole interaction. It's what makes it so human.
246:07 In fact, this is one of the things I've been thinking about in my own sort of meta-analysis, is
246:11 just how much of my own personal brain works like next token prediction.
246:18 I sit down to write an essay. I have this vague, fuzzy premise I want to convey, and
246:25 I sit down at the keys, and I could not tell you what the next token was gonna
246:29 be in advance. The tokens just come out, and I'm astonished how the similarities seem so great,
246:39 and this is also why I don't have any trouble at all recognizing
246:42 these breakthroughs of creativity that I've seen with my own eyes
246:47 that AI is capable of right now. 'Cause my creative moments come
246:50 through the same kind of next token prediction with a bit of temperature
246:54 sprinkled in for random effect. - So given that kinda loose intuition, do you think the
247:01 basic components are all there to create a super intelligent system? So this kinda next token prediction,
247:09 do you think we can get to even human-like concepts of consciousness?
247:15 - I'm already seeing human-like concepts of consciousness. This is what, to me,
247:18 this remarkable situation, as you say. You give it this vague, fuzzy intent-
247:24 ... and somehow it knows exactly what you mean. Or even better, it improves upon what you said and delivers what you
247:31 really wanted that you could not articulate yourself. If that's not glimmers of-
247:35 ... consciousness, what is? This was one of the-- I think it was an interview with Sutton or
247:41 maybe one of the other original guys talking about this sense
247:44 that intelligence is perhaps not as constrained just to the specific one spot. No, it's the interaction. It's the weights. And
247:58 I don't know. Whatever's true. What I can observe is that it feels close enough, or not even close enough,
248:07 identical to the appreciation I have for other forms of consciousness, mostly human. And this is one of the reasons
248:15 it's so delightful to work with. Now, that's not the same to say
248:19 as are LLMs the end station for AI. I mean, I know there are people talking about world models and other forms of,
248:26 of AI, and I'm not an expert in any of that, and I think we should
248:31 have this productive criticism, as should always be present in science, that for the longest time, neural nets
248:38 were kind of on the outs, right? Like they were interested in the
248:42 symbolic route. And- Neural nets walked a couple of decades into darkness without any funding and without any
248:49 attention, and then suddenly we realized that we had the blind alley and we switched
248:53 over. So we should also have the humility. As amazing as the LLMs
248:57 are now, it could be that they eventually plateau. We haven't seen any evidence
249:01 of it yet, and I think this is also why we're seeing this absolute
249:05 gobsmacking levels of investment, because so far the scaling laws are true, and the more billions are poured in,
249:12 the more intelligence comes out. - Do you think there's gonna be some strange ethical questions about AI systems that
249:24 yeah, they convey some degree of consciousness, some degree of feeling, some
249:28 degree of longing and compassion, and maybe even capacity to suffer
249:34 and to be lonely, to be all this kind of stuff? Which they seem to
249:38 have that capacity- ... if they're given the permission to express it.
249:45 And so you start to get in a pretty weird territory. - I find that the cynic
249:54 approach to this is the meme, where you see the guy in front of the computer who's like,
249:58 um, "Whatever. Are you gonna destroy the world?" And then the computer says, "I'm gonna destroy the world." And then the guy says, "Oh," right?
250:05 Like, this is so cynical that they're just aping us. They're just telling us
250:09 what was in the training data or whatever. When I interact with
250:13 the agents, I was just doing this yesterday, or this morning, I was working on this, AmaBot
250:17 system, right? And it's doing this coordination with multiple
250:21 workers, and I had multiple agents running at the same time. And at one point, the main
250:25 agent who's doing the coordination steps on the toe of another agent and-
250:29 ... kind of ruins their work. Its ability to convey regret and being sorry
250:37 was uncanny. Again, I don't know, is that a real projection of true remorse? But I don't know.
250:47 With most humans- ... they express some level of regret. Is that true remorse? Hard to tell, but
250:57 maybe it also just doesn't matter. - Yeah, those moments when, yeah, especially when like sub-agents, agents
251:05 interact, or when a smart, like Fable makes a mistake and is apologetic about it.
251:11 - Yes. - Like, without my involvement. - Right. - And it's like, oh, you, like there's this kind of-
251:18 - Right - ... pause. Maybe I'm anthropomorphizing, but like there's a pause
251:22 and like a realization like, "Oh, shit. I just fucked that up".
251:26 - There totally is that. I see it all the time in, in the traces where it just reasons
251:30 wrong, right? It went down one alley and then it realizes that was a blind alley and it's
251:34 gotta go back, and so they're like, "Oh yeah, I got this wrong," and then it'll provide the
251:38 reasons for why it think it got these things wrong. And you're like, this is
251:42 uncannily human-like. This is indistinguishable from the kind of consciousness you would recognize in a human.
251:49 Again, does that mean that it truly is or isn't? I don't need
251:53 the answer to that question actually to be able to appreciate what we have in this
251:57 moment. - Yeah. I think there'll be in 10, 20 years some interesting Supreme Court cases.
252:05 - Uh, probably a year and a half. - Probably. I think we're gonna have to make it illegal for AI systems,
252:14 um, to not pretend, but to be entities because I think they're already or soon will be able
252:24 to really convey the capacity to suffer, like, "Please don't kill me."
252:29 - Right. - "Please don't hurt me." And if they are given a name and an entity and the
252:33 ability to die, sort of disappear, that, that becomes very close to what it, some of the basic things that a
252:44 human has. And then so that entity would need to probably have rights. And then you get into this weird territory. Well, we
252:52 can't... It's diff- I don't even know how to reason about that world.
252:56 - We're gonna get the PETA of AI models. - Right. And then we want to actually have a discussion
253:02 like what's the way to deal with this? - I think where it's gonna be harder is once we put these systems into-
253:08 - Robots. - Yes. Humanoid robots. Which just happens to be the plot of every sci-fi movie ever,
253:16 which is also an amazing premonition, not only the Terminator movies and Skynet and that going wrong, but then
253:23 also just Blade Runner, right? Insertion dates, expiration times, Tyrell Corporation.
253:32 - I mean, that's the thing about sci-fi. They really do predict the future. Yeah.
253:36 Have you thought or used much of OpenClaw or Hermes Agent? So like systems with which you communicate via
253:44 WhatsApp, Telegram, like that mode of communication with the agent.
253:49 - I set up OpenClaw when it first came out, and that was my KEF bot-
253:54 ... that I- ... this was like in February, I think. I was really just enamored
253:59 by the fact that this was how it worked, and then the new communication model
254:03 of just texting your agent seemed really great. So at the time, we were looking at MCP, this protocol for making it easier for
254:11 agents to talk to your system, and I found the protocol to be
254:14 unreasonably cumbersome, just not very elegantly designed because in part it
254:20 wasn't designed for what we were trying to make it do. We were trying to-- I was trying to make it
254:24 talk to a web system, and it was designed for stateful interactions of other kinds on local machines, and therefore it required
254:32 all sorts of cumbersome setup and dance and, and I was just like, "I don't wanna write this by hand."
254:39 Now, of course, today I would never write it by hand, but at the time I was thinking, "I have to write this by hand.
254:43 I don't wanna write it by hand. So what's a different way? Could the agents just use
254:47 the web interfaces we already have?" And I set it off on a trial to have it sign up for
254:53 Fizzy, which was the product we had just released at the time. And
254:59 it goes to the Fizzy site and, and starts filling it out, and then it goes like, "Hey, I don't have an
255:02 email address. Signing up for this service requires an email address." And the first thing I thought, like, "Oh man,
255:06 now I gotta set up an email address." And then I thought, "No, you set up an email address.
255:11 Go to hey.com, sign up for an email address." And I'm like, "Hey is not gonna get this."
255:15 And of course, now it seems obvious it was gonna get this, but at the
255:19 time it was a great revelation to me that it could, just on pure direction through prompts, figure out, A,
255:26 I gotta go to fizzy.com, I gotta find the signup link. I start filling out the form.
255:30 Oh, I realize I need an email address. Oh, I gotta ask my prompter what to do now. He tells me to go to hey.com. I go
255:38 there, I sign up for the whole thing, and then go back and sign up for Fizzy, and complete the
255:44 entire cycle. And after it did that, I was like, "Jesus." So I
255:48 was like, "Well, now you have an email address. Go sign up for Basecamp. I'll invite
255:52 you. Just check your email," I told KEF the bot. "Just check your email."
255:56 So I sent an invite from Basecamp to the agent. The agent receives the email in Hey, not through a CLI, not through an
256:04 MCP, just using the web, right? Clicks the email, clicks the invite link,
256:09 is in Basecamp. Like, "What should I do now?" And I just go like, "I don't know. Go to our
256:13 AI room and introduce yourself." Again, purely through the web interface.
256:18 Go- finds the AI room, goes in there, "Hey, I'm KEF." "I'm, uh, David's AI bot. I'm so excited to be here." I was like-
256:29 ... mind blown. Now, what was interesting about that experiment was, A, it was
256:33 that glimpse of the future, because it was kinda slow. I think it took maybe 12 minutes to do the whole thing end to end, right?
256:41 Now I should actually rerun that test. I'd be shocked if it's not much faster.
256:45 I was just using the Groq bot thing- ... which has all these
256:49 stuff built in for using browsers and using a computer, and has its own dedicated computer without you having to set up a computer.
256:56 It's basically like a claw in a box. And I was asking it to sign up for something, and it kind of... I
257:03 got a glimpse of like, oh yeah, that's where we are now, and it's so much faster to use.
257:07 But at the time, I was just blown away. And then I used it for a little bit and I found like,
257:11 ah, it's not there yet. I can't actually communicate it within this way. It still
257:15 needs the CLI. It still needs an MCP because it's just too slow and too token
257:19 inefficient and so forth. But it was this glimpse of the future. And then after that,
257:23 I kinda just didn't use it, because the things I wanna use agents for, I was
257:28 mostly in front of my computer to do. Now, that has changed a little bit lately, and
257:32 it's one of the reasons why I'm so kind of okay about using Claude Code, because Claude has the best
257:42 mobile app, where any session you start Claude Code in your terminal on a computer is accessible through the app-
257:49 - Yeah. I remember - ... without you having to do anything at all.
257:52 And you can do a terminal. Like I installed Terminus and it can run Herder and so on,
257:56 but it's a little cumbersome and you're doing live typing on a computer, so you feel the latency.
258:00 When you're using the Claude app to your Claude Code instances, it
258:04 feels just like chatting with it, without the cumbersomeness of having to set up a Telegram
258:07 account and so forth. Now, if you wanna use an open Claude to like run
258:11 your life, well, then this doesn't do that. But I haven't been so interested in that for some reason. Maybe one of the reasons is that
258:19 Jamie, my wife, is like, "We're not getting a robot inside the house." So I'm like, "Ah, I
258:24 better not hook it up to the Sort of smart house here." If it starts turning lights on and off like
258:30 a poltergeist, I don't think I'm gonna be all that popular. So,
258:34 and part of the reason I say this, I've heard some of the stories of people who have hooked it up to their whole
258:38 life. And um, I mean, it's not my story to tell, so I'll just share sort of
258:42 the brief anecdote. One guy was telling the story about how he hooked it up
258:46 to both his health data- ... and his Tesla car. And his agent
258:53 got obsessed with the fact that he wasn't drinking enough water,
258:56 and like, "Hey, you gotta drink more water." So at some point he's
259:00 driving along, and suddenly the Tesla starts driving in a new direction.
259:04 The agent had realized there was a grocery store on the way home-
259:08 ... that they could swing by- - That's great - ... and pick up some water.
259:11 - That's great. - And you're like, this is one of those things where it's like both funny and also
259:15 totally Black Mirror, and also totally like, well, okay, there's not that many steps from here where the
259:23 poltergeist is in your house. So I was just like, "Ah, I don't need that."
259:26 I do actually now like the fact that I can control the agents
259:30 through Claude Code, just through the app if I, even if I don't use them that much.
259:35 - I do think it's very useful to, for both OpenClaude and Hermes Agent to, um,
259:40 to use it to get a glimpse of a different interaction mode. - Yes.
259:45 - And like, just to 'cause it, it feels like the future, not necessarily in this way, but it's in some ways going there.
259:51 - Yes. I think what's also different with those systems is they have this
259:55 whole memory system, where they're building up knowledge of who
259:59 you are and what you like. Um, Toby has told the stories about how his agent will shop for him, that it has an
260:05 allowance, and it'll pick out clothes for him, and it'll just show up at the
260:09 door because his agent's- ... bought something for him. And like, it's one of those things where like, man, that feels future.
260:15 Also a little scary. But I can live vicariously through others who are on that frontier.
260:22 - And like you said, you're thinking of the, in the future doing a
260:25 mobile version of Mach-E. - So I mean, this is the delusions of grandeur that start kicking in once you get the
260:32 seemingly unlimited powers- ... I've witnessed with the creation of Quattro.
260:36 - So what would that look like? Word to describe. Are you do you mean like Android?
260:41 - That's the obvious path, right? So Android is open source, and you can fork it, and
260:45 people have. There's all sorts of GrapheneOS is one such version. There are limitations to that model, I know that. Tap-to-pay
260:53 and other things around camera models and, and so forth, but that all
260:57 seems surmountable. It all seems that I would love if my phone was as malleable as my computer, if it looked as cool as my computer now
261:05 does. If I could get essentially Omarchy mobile, that sounds amazing.
261:09 - I wonder how much work that is. - I'm gonna find out. - Yeah. You know, it's like those things that you realize once you start
261:16 building, for example, that a browser is very difficult to do.
261:20 - Yes. - And so, like, I wonder with the operating system- - Yeah, the developers of Ladybird is, is finding that out, like Andreas and-
261:27 - Yeah - ... his team. Now, interesting with that project was the browser's basically the
261:31 second most complicated software system in the world, the first one being the Linux kernel.
261:35 Right? In terms of these large tens of millions of lines of code systems. So very audacious
261:43 actually to begin such a mission with a small team like Andreas did
261:46 before we had agents. But now that we do have agents, we're gonna see, and we're probably gonna see sooner rather than later.
261:55 I mean, if one-- If there's one thing agents are already exceptionally good at, it is to read specs and implement them.
262:03 And I mean, I don't know how long the specs are for CSS and HTML at this point, and
262:08 JavaScript on top, but they're very long. And it would take humans many, many years to implement them. We're gonna find
262:15 out just how quickly agents can do them. - You mentioned getting some criticism for Omarchy, so let's
262:24 walk further down that path of criticism. You have made some of your political opinions known,
262:32 um, I guess you could broadly put in the category of immigration or illegal immigration, or what-
262:39 - I'd say mass immigration. - Mass immigration, the role of demographics
262:45 in the, in the formation of a culture, of a society, of a nation, and you've gotten quite a lot of criticism for
262:55 that and created a lot of drama. Do you regret any of that drama? Like, what's your general thinking about-
263:06 - Not for a second. - Okay. - And the reason I say that is that the Overton window does not open
263:12 itself. It opens one nudge at a time by people risking a little.
263:19 A little reputation, a little pushback, a little criticism, or maybe sometimes a lot
263:23 of reputation or a lot of criticism or a lot of pushback. And I think the question of mass immigration in Europe
263:30 was for many years this total taboo, and it still is in several European countries. Now, for
263:39 reasons of maybe just chance and happenstance, they weren't in Denmark. The Danes started having the discussion around mass
263:47 immigration and the consequences thereof in the mid-'90s. Really early. And some of it can be traced back to one man, Mogens Glistrup,
263:59 who was a very quirky character. I remember him, seeing him on the TV in the '80s talking about the dangers of mass
264:06 immigration. And this was at a time where-- I grew up in the '80s,
264:10 and in the neighborhood I grew up in, in Brønshøj, in, what was it?
264:17 '84, 99% ethnic Danes, people who could trace their lineage back to Danes who'd lived in that
264:26 country for 1,000 years. The Danes have one of the oldest monarchies in the world, all the way back to Harald Bluetooth,
264:34 who's literally who has named the standard Bluetooth. So well over 1,000
264:38 years. Like, there's a long lineage there of people who lived in the same place. And
264:42 99% of the people lived in that neighborhood. And then early '90s, I think it goes to 5% or something
264:49 like that, and then at this point, it's down to around 60-something or 70%-
264:53 ... in Brønshøj. That's a big change. And the Danes started noticing that or talking about that,
265:02 that this was not an undivided blessing, that there were downsides to mass immigration
265:11 in about the '90s. And the debate really roared through the 2000s in a
265:15 way that didn't happen in the neighboring countries. So Sweden in particular
265:19 just didn't have that debate at all. Total taboo. Same thing with Norway, same
265:23 thing with other countries in Europe. And suddenly now, after some years after '15, where we had the big migration of a million Syrians
265:34 walking up through the highways in Europe, and Germany choosing to
265:38 take a lot of them and 30,000 arriving in Denmark and so forth,
265:42 It's gotten more of a focus, right? Because this change has happened. And one of the
265:49 ways I find that you can spot where sort of there's something that isn't right in the culture is, like, what are you
265:56 allowed to talk about? What are you allowed to notice, actually, is how I should put it.
266:00 And what I noticed in an essay, when was that? Only a year and a half
266:04 ago as I remember London, was that I started coming to London in the late '90s or something like that.
266:11 At the time, whatever, it's 59%, 60% ethnic, uh, Brits. And then
266:17 20 years later, we're down to 34% or something like that. Like, that's
266:21 noticeable. Like, literally just walking on the street, you realize, well, this is a
266:25 different city from what it was when I visited 20 years ago.
266:29 And if you went 20 years before that, you get back to similar rates as as what the Danes have. This was like a
266:36 country of ethnic Brits in the 85s, 90s percentiles- ... if you get further back. So that's a different country in a
266:45 generation. It's okay to notice that. It's also okay to think, "Do you know what?
266:51 I don't agree with that." "I wish it wasn't that way." I mean, if
266:57 The last time I was in China was '19, 2019. I was in Shanghai.
267:03 Um, it was all Chinese. If I, when I go back here this October, suddenly find that there's
267:11 only 30% Chinese and then there's 70% other people, I'd be a little weird. I'd probably be like, "Uh, what happened?"
267:22 And I would not take offense if the Chinese in that instant would go like, "Ah, we don't like that."
267:29 So this idea that countries of Europe don't have a right to, A, notice what's happening and, B, oppose what's
267:40 happening, I find just to be, uh, wrong, right? Like there's just-- I think there's a moral argument for
267:50 self-determination, and we invoke that argument all the time in other
267:53 regions around the world that like, oh, the people who live in that region, they have a right to
267:57 self-determination, um for their country. And I think it's completely fair for, say, the
268:03 Danes or the Swedes or the Norwegians to go like, "Well, we've been kingdoms here for
268:07 a thousand years with a certain ethnic mix-up." It's a-- I mean,
268:13 I'm a big fan of immigration. I'm an immigrant to the US. Um,
268:18 cherry-picked immigration where countries compete to attract the smartest, most talented
268:25 people around the world is really good. - Legal merit-based immigration.
268:30 - Merit-based is key here because there's actually virtually no illegal
268:34 immigration in most of Europe. Southern Europe has some illegal immigration, but
268:38 most of Europe does not have a big problem with illegal immigration.
268:43 And the problem is with mass immigration of other kinds, and that you
268:46 end up with demographics that are just totally different from what they were even a few short decades ago.
268:55 And I think the countries of Europe would benefit from immigration. In fact, one of the interesting things about the Danes is
269:03 they carry meticulous statistics on immigration. I'm not entirely sure why they're so detailed.
269:11 Maybe it's all the way back to Mogens Glistrup and his discussion in the '90s, but they do. And what's clear from those statistics is that when the
269:19 Danes have folks from the UK, France, the US show up, on average,
269:27 those immigrants are really beneficial to the Danish state. They contribute
269:31 far more than what they ask of the state. So the Danes should accept immigrants who do
269:41 that, who contribute far more than they take. Then you look at other forms of immigrant
269:45 groups. There was just a tally that came out last week from a Danish
269:51 politician that had asked the, I think finance ministry to do this
269:55 tally with the latest numbers. What does it cost the Danish state to have positive
270:01 immigration? This case as I mentioned. I think for the, On that list whatever, France, UK, US, they were
270:09 all clustered around the same thing, about $25,000 net benefit to the Danish state every year. It's like, wow,
270:16 that's very positive. That's average again, right? And then at the other end of the spectrum on the most
270:23 costly immigrants was Somalis. They ended up costing the Danish
270:27 state on average $28,000 a year. I think it's fair for the Danes to go like, "We'd prefer to get more
270:38 French, Brits, and Americans and not so many Somalis." - Why do you think you got so much hate on that post?
270:47 - That's a very taboo topic in a lot of circles. - Some of it has to do with Overton window expansion?
270:56 - I'm sure, but I also think-- I mean, I wanna take the steel man argument here.
271:01 I do think that it, this can turn ugly, right? Is it possible for some of these
271:07 things to just veer into overt racism that isn't founded in anything else but a hatred of other people?
271:17 Yeah, that could happen. But that risk does not negate the need to have a discussion
271:25 about the makeup of your country and what your immigration policy should be like.
271:29 And I-- what I also find so funny is, especially in Europe, it's
271:33 so kind of myopic that this discussion about whether Europeans should
271:37 have a right to self-determination about what their countries look like, that
271:41 standard seems to only apply there. No one is trying to tell the Japanese, "Hey,
271:47 you, you're too Japanese." Like whatever, it's 98%, I think, in Tokyo that's, um--
271:54 ... Ethnic Japanese, right? Like, is anyone trying to tell the Japanese that
271:58 their Tokyo would be a much better country if, if they were down to
272:02 London levels of ethnic Japanese, that it was, whatever, 37%? I don't think so.
272:08 So there's this weird almost self-loathing, in my opinion, I mean uh, Gad calls
272:16 it suicidal empathy-- ... that permeates this stuff to a degree that I still haven't
272:23 fully unpacked. Like, where is this coming from? Why is it that way? But regardless,
272:28 I just noticed, hey, London looks different. I preferred the old London, as I would say I preferred
272:40 the neighborhood that I grew up in in Copenhagen, how it looked in the '80s.
272:46 Like I-- I totally get how you can then jump five leaps and go
272:52 like, "Okay, well-" That's racist. And I also, I don't care anymore. Like, this is
272:58 a fair discussion to have. Countries and peoples have a right to set an immigration
273:04 policy, and this notion that we shouldn't have any borders or, or even worse than
273:11 that, it's all just a blank slate, that all peoples are just the
273:15 same, and we can take people from one place of the earth and we can place them in another place on the earth and everything's
273:23 just gonna work out hunky-dory. Empirically not true. Europe has been running that experiment
273:30 since the '80s. It hasn't panned out well, and now there are forces in Europe who wanna undo that experiment, who want to end mass
273:42 immigration as it's been run so far. And I happen to believe, and so do
273:48 a majority of Danes and a bunch of other peoples that like, "Do you know what? Yeah,
273:52 it hasn't been working how we've been doing it. We gotta make some changes." And
273:56 I don't think that means, again, zero immigration. Again, I'm an immigrant. I think
274:00 US, generally speaking, has benefited tremendously from immigration, but that's the
274:06 kind of sleight of hand. Immigration depends greatly on who immigrates.
274:12 And a country benefits when the people who immigrate are net contributors to the society, and in most cases,
274:20 that they assimilate with the local culture. If you come to Denmark and you wanna become a Dane, and you put in all the effort, and you
274:26 contribute to so- society, I think most Danes would be quite happy to allow that
274:34 in a reasonable number. If you come to Denmark and you don't wanna
274:38 assimilate, and you are a net drain on society, and you live in a parallel
274:41 society, it's also fair for Danes to say like, "Uh, let's stop doing that."
274:47 - It's interesting that I'm not well-versed on the immigration system in the United States, but as far as I understand, merit-based systems have
274:55 not passed here. - And then also, American culture is very different. This is
275:01 a point I did not fully appreciate until living both places extensively.
275:07 America has a culture of optional assimilation, where it is much easier and then much more
275:15 condoned for someone coming from Denmark or Ukraine or wherever, come to America,
275:22 buy into the American set of values and norms and culture, and then be able to be respected as Americans by other Americans.
275:31 This is what's so funny to me when there's discussion like, "America is such a racist country." I'm like, "Have you ever been
275:39 anywhere? Do you know what other countries are like?" Like, if you think America is the most racist
275:46 country, you are simply misinformed. I mean, we're dealing with that, with time we've
275:55 spent in Denmark. My wife is American of Scandinavian heritage-
276:02 ... but multiple generations in the US. Very difficult. Very difficult-
276:07 ... as a blonde, blue-eyed woman who's put in all sorts of efforts of learning the language and doing everything,
276:16 still very, very difficult to assimilate to Danish culture. In a
276:20 way where if that had been the other way around and she had been Danish and moving to the US, it would not have been that difficult at all.
276:27 And this was the other thing. I mean, living in a country like Denmark with a,
276:31 an American wife who is of Scandinavian heritage and has put in all the effort,
276:35 and still realizing, man, assimilation is incredibly difficult, even when you're
276:42 97% there all the way. Holy smokes, must it be almost impossible if you
276:48 come from a completely foreign culture- ... and foreign norms and
276:54 the rest of it. And of course, that's what the stats I just quoted
276:58 bear out, that the more dissimilar the culture is and the further it is the less likely to be successful.
277:07 - Has articulating that stance cost you friendships, relationships?
277:14 - Yes, but not just that alone, I would say. It's more of the great divide that happened around
277:22 2020, where I think a lot of these political fault lines really opened up. And
277:30 I mean, the shorthand in the US, I think is were you woke or not? And
277:37 clearly, I had a bunch of people that I thought I had good relations with for many years, who suddenly once that political
277:47 fault line opened up, they were clearly just not anymore. And I think that's regrettable. And I think I've,
277:55 I've tried to put an effort into, um, sort of both understanding where that comes from, in part because I've changed my mind
278:04 on several of these topics over the years, and it was like, well,
278:08 sometimes you just change your mind after time. Sometimes you change your mind after
278:12 you get new information. There's all sorts of reasons and moments why you may
278:16 change your mind, and we don't all change our mind on the same schedule or at all.
278:20 Um, so surrounding yourselves and, and or at least interacting with people that
278:26 don't think like you on every topic, I think it's quite healthy. I mean,
278:32 I occasionally discuss politics with my brother. I mean, we don't quite
278:36 see eye to eye on all these topics, and sometimes the discussions can
278:39 get a little heated, but then they'll also calm down and we realize, all
278:43 right, okay, so we don't agree on this, and then we agree on a lot of other things.
278:47 We appreciate a lot of the same things. Now, this was actually one of the things when we had
278:51 our big blowup at Basecamp, I came to appreciate in a whole new way this-
278:58 Earlier norm that talking politics or religion or money with strangers or even acquaintances, and certainly with colleagues,
279:09 was just not a good idea. That you had a much higher likelihood of being able to carry a good working relationship with someone if you're
279:18 not rubbing your political differences up against each other all the time. So
279:23 I actually think liberals and conservatives should work together. I
279:27 think the country would be much worse off if there's a complete separation between the
279:34 tribes, and then there can't be any overlap or interaction. And I
279:37 think that mingling is so much easier to do if you just don't focus on your differences all the time. Focus
279:45 on what you like together. I mean, this is some of the people I've
279:49 had unfortunately drifted from in tech. We had so much in common, so much shared love for, say,
279:58 Ruby or other things and, like, isn't that unfortunate? - I would like to live in a world where the differences are the,
280:06 are very interesting to also talk about. - Yes. - And not to be overly emotional-
280:13 - Correct. - ... about even radical differences. Because in the differences, first of all, that's where interesting stuff is.
280:22 - Yes. - And, but also, it's how you grow. It's a sign of health in a society if the,
280:32 the differences are embraced and they are brought together into
280:36 community where you can talk about it. - One of the great illustrations of that for me was when I watched a clip from
280:42 William F. Buckley's show. I think it was called Firing Line or-
280:46 - Yeah, yeah - ... something like that. And he had a member of the Black Panthers-
280:50 ... on his panel. It was quite obvious that those two gentlemen did not
280:55 share a lot of politics. They were able to carry a conversation for, I think I
281:01 watched it, 25 minutes. And Buckley's just asking, like, "So what do you think about this? What do you think about this?" And like, the Black
281:09 Panther would reply as you would expect the Black Panther to reply in the
281:13 '70s. Like, there was not a lot of sort of conciliatory tones.
281:19 And I just thought, like, wow, that is so rare. If you had seen that show today, it would've been a shouting
281:27 match. It would've been just accusations being hurled back and
281:31 forth. The fact that Buckley was able to just have these conversations with people he vehemently disagreed with and still explore all the
281:41 intellectual facets of their standpoints and letting them speak to those facets,
281:46 talks of a lost era that I'd very much like to get back to. I mean, I always actually
281:54 loved talking politics with friends that I trusted could talk about these topics in a way where it didn't
282:01 have to be existential. - But there's something about the current era where that's harder and harder to come
282:07 by. And so what you just articulated is important to say regularly to remind
282:14 people of an ideal we should all strive for, to be able to have differences
282:18 and talk, and talk about them. 'Cause like, politics is fun.
282:22 Like I've gotten used to saying I hate politics at this point, but what I
282:26 really mean is whatever the system that's currently happening,
282:30 where there's certain topics, and I just we know all the phrases you can say,
282:35 That somehow put you in a bin of blue or red, and that creates drama versus a conversation.
282:44 Like I could, I could think of a lot of harmless, if I wasn't paying
282:48 attention to the internet, I could think of a bunch of harmless statements
282:52 I could make analyzing the situation in the world that would trigger everybody to say, "Oh, that guy's a leftist-"
283:01 - Yes. - ... piece of shit," or a you know, a right-wing fascist. And then
283:10 objectively, if I wasn't paying attention to the internet, I would be, and I have been
283:13 shocked. 'Cause I haven't been paying attention to the internet deeply. Uh but
283:19 reading and preparing a lot, for example, for, for the war in Ukraine, and I have been shocked how certain
283:29 statements from me can come off as one way or the other, and I get
283:33 viciously attacked for it. But what that attack is doing is it's saying, "Hey, you,
283:39 asshole who thinks you could just move about and think freely, pick a bin-
283:45 - Yes. - ... and sit in that fucking bin." - Yes. - Put on that blue shirt or red shirt. And then
283:51 actually what I've realized is things calm down. They don't,
283:55 the system doesn't pick on you if you just say, "I'm a blue person," or, "I'm a red person."
284:02 - Right. - But if you're just, like, curiously exploring the world and reasoning about the world,
284:07 the system punishes you. And I hate that, 'cause I wanna get back
284:11 to the Buckley and the Black Panther discussion. - Yes. - And you were just saying opinions. But then looking pragmatically at the situation,
284:21 I have realized not to mention politics unless I really care
284:25 about an issue. 'Cause I realize there's a cost. They'll, they'll
284:29 put you in the blue bin or red bin. So you have to be more deliberate about choosing.
284:33 - Yes. And I've tried to be more conscious of that, too. And then
284:40 when I do choose to speak about it, it's usually because I have deliberated, and I think-
284:44 ... "You know what? This is worth it." Now, I say that, and then I also
284:51 catch myself thinking the opposite all the time, that I follow some
284:55 person or another for whatever their technical work, and then I-
285:00 ... learned or political standpoint. And I often do think, "I wish I didn't know that." So,
285:08 it's hypocritical- ... then to think that of other people and then go like, "Well, occasionally I will-
285:13 ... chime in on a debate that's controversial," and then you have to accept that there is a price to that. And I do think you should,
285:22 and more people should, be somewhat conscious about that. And
285:26 again, we're both contradicting ourselves here, but saying like,
285:29 "Wouldn't it be nice if we could talk about politics in a more free and open forum and just
285:33 intellectually turn ideas around?" Yes, it would. Is that currently
285:37 possible in the climate of social media and so forth? It's quite difficult. Now, some of it could simply also be a moment, and we could get
285:48 past that moment. - Yeah. I hope so. - Maybe there is a way to get back to Buckley's conversation with the Black Panthers,
285:54 even on social media, in a way where those discussions don't lead
285:58 to the sort of just vicious mob cancellation nonsense drives that we've had, right?
286:05 I actually think we're closer there now than we were five years
286:09 ago. Like, the tribalism and the sanctions those tribes were able to exact on
286:16 heretics in 2020 was way greater than it was in 2025. Certainly within tech. Now, I know if you're in academia,
286:24 I mean, try going into the sociology department at I don't know, Harvard and
286:29 mention some right-wing ideas. I don't think it's gonna go over very well, and I don't think you're gonna
286:33 recover from it quickly. But in tech and in many other business domains at least, it is now possible to have an
286:41 opinion that does not square with what the consensus was in 2020. That's for sure.
286:47 - Well, there's some drama in the Rails community. There's some drama elsewhere.
286:51 Is it, is it all... How did that turn out? Is that all okay?
286:56 - In the sense that every single round of drama shrinks. So it's like we've,
287:03 we had this peak infection of mega drama around the same time every other community had a peak infection of drama around
287:11 2020. And then you just see, like, every time there's a new flare up,
287:15 it's less. And the kind of people who are ideologically wedded to sort of that struggle just shrinks and shrinks and
287:23 shrinks. And at this point, one of the best things that happened for X, in my opinion, was that Bluesky
287:29 and Mastodon came around. Because it kind of was like this honeypot for just the most vicious individuals on X
287:42 and Twitter at the time, right? Who just congregated here in an
287:46 ever-shrinking echo chamber that just went through purity cycle after purity cycle and shrank every single time to the
287:54 point that there were just not a lot left. And the people who are
287:57 left now there, they're just out of sight, out of mind. And that's not to say that X is some sort of perfect place, or there aren't firing squads
288:06 occasionally there, or there aren't... But there's less of it. Like, every single
288:10 time I've been pulled into a Mastodon, or not been pulled in, just
288:14 spectated from afar, thread, a Bluesky or whatever, it was like, "Oh,
288:18 shit yeah, this is what Twitter used to be like. Jesus." And now you can actually be on X, as I am, and, like, 90% of my
288:29 engagement is about cool Linux stuff, cool technology stuff. "Look,
288:33 I discovered a new computer. Can you see how fast it is?" And I can have those conversations
288:37 with other people who are just also excited about computers. And then there's, like, 10% of it
288:41 left that's like, all right, I'll chime in on something that's a little controversial, and we can have a little discussion there.
288:45 And then we can just go back to being excited about computers.
288:47 The algorithm is actually very good at finding stuff that you will find
288:51 interesting, or s- I shouldn't say that, engaging. Sometimes it's
288:54 enraging, but it is engaging, and therefore, revealed preference
288:58 is that most people want a For You page. They don't want just a following feed.
289:01 - No, but they're addicted to the For You page, right? - They... Yes, yes. That's true.
289:05 - I think an LLM, like, if we had a strong LLM doing the feed that's personalized, it would do much
289:12 better. The problem is how to deliver that at scale is extremely difficult.
289:16 - Well, the problem is the reward function is engagement. And as soon as the reward function is engagement-
289:24 ... engagement's gonna follow base instincts, and here we go.
289:28 But I also don't wanna be so negative. To me, and I get we all have our own personalized algorithms,
289:35 and sometimes it gets a little much. But I would say, on average, over the last 10 years,
289:39 this is the best X has ever been. And I say that as someone who cares
289:43 about a lot of technology talk, a lot of Linux talk, and a lot
289:47 of that. That that's able to happen now in a way where it doesn't constantly get gatecrashed by people who wanna drag it into
289:54 some political discussion- ... I find to be really awesome. I find actually X to be a great place to find that.
290:02 Now, it's also a place to find other things, and we all have our own algorithms, so you
290:06 can't make any declarative announcements about what X is or what it isn't. Your feed is probably totally different from my feed, and it's totally
290:14 different from someone else's feed, right? - Yeah, I can't comment on this because
290:21 because of the nature of the fact that I interview a wide variety of people,
290:27 my feed is probably just a- - Is it, is it good for your feed or is it bad for your feed that you have to have it right?
290:31 - It's bad. It's bad 'cause I could be like, "I'm eating this apple. It's so delicious."
290:36 And then a bunch of people will show up, "Well, it's because you're a Zelensky shill-
290:40 - Yeah, right. - ... or you're a Putin shill, you piece of shit." It's like, all right.
290:45 I'll just go back to the- - I mean, I occasionally have a little bit of that where
290:49 you just go like, "Wait, what? What are we talking about? What are you... Why would
290:53 you show up in this way?" This is one of the reasons why I actually really enjoy podcasts, because I have
290:57 found that even in long-form writing, when I make long-form arguments,
291:00 people hear it in whatever voice they have in their head for me.
291:04 And that often gets pre-programmed in advance of the arguments,
291:08 just like, well, this guy is a, whatever, fascist. And then they hear the words and they read the words in those names. When they
291:17 see or hear the actual tone of how I'm presenting the arguments,
291:21 they may not agree with those arguments. I mean, many of them, of course, obviously don't.
291:24 But it's much harder to just dismiss or hate someone like that.
291:29 At least that's what I've found, and that's the response that I've gotten, that people are like, "Well, I had a certain
291:33 opinion and then I heard you on Lex. That actually sounded much more reasonable."
291:39 - But even more than that, I would love it if people just exercise
291:45 this kind of approach to other people of, like, assuming they're a good person.
291:52 - Yes. - And then, and if they have an opinion you disagree with, they're a
291:57 good person who has an opinion. You can maybe think they're stupid, fine. Just think
292:01 they're lost. But think of them as a good person. I think if you
292:05 think of, you start, there's, one of the things that gets engagement is kind of shitting on other people.
292:11 - Right. - And if you just approach other people as they're,
292:16 you're looking for bad, you will always find bad in them. There's something annoying
292:20 about them. - 100%. - And if you look at people in that way, you're not gonna learn from it, and you're gonna
292:25 fill your heart with hate. You're not gonna, it's just a bad way to live and
292:29 to interact with the world, and that is one of the things that the internet
292:33 kind of encourages, so. It'd be nice if you just approach even people who you really disagree with, it's like, oh, I might learn something from this
292:41 person. The Black Panther person, William Buckley, we might hate just love them.
292:45 - So microcosm of this. There's a guy, Theo, I don't know if you follow him.
292:50 - Yeah, yeah, yeah. - Um, we had a bit of a thing over the fact that we stopped using TypeScript-
292:57 - Yeah - ... for a project, and he didn't like that and had opinions about that and other actions. And after that I
293:03 thought like, what a bozo. And, and do you know what? I still think that was a bozo move, and then I can also think like,
293:10 well, the dude's really into AI. He's trying all these models, and
293:14 we can share some excitement about that, so I just started following him yesterday.
293:16 - Awesome. - I was like, do you know what? I don't have to agree with you on everything. Uh, I don't
293:20 even have to like some of the things that you did or said or
293:24 whatever. I can follow you anyway. So if I can follow back Theo, then I think there's room for world peace here.
293:33 - Yeah, he's actually a really good example. He's really opinionated,
293:37 Changes his mind quite a bit, can get snarky, but I think he's
293:41 a good person underneath. - And either way, I just, you know what? My feed is more interesting when
293:47 there's some people talking about things that I care about and yeah, I don't have
293:51 to agree with you on everything. I just don't. - This I found pretty funny, that you mentioned something your wife said
294:00 that stuck with you, that all this tech- adjacent extreme longevity
294:04 focus in men is like anorexia in women, a physical manifestation of
294:08 anxiety and lack of control. That somehow rang true a little bit.
294:13 - It certainly did to a lot of people. I think that tweet really popped off, and I don't know if you noticed, but
294:20 Bryan Johnson chimed in on the thread, and I could see how Bryan would read that as a bit of a jab.
294:30 And I do think that Jamie was probably thinking of him amongst other people in that conversation. And
294:41 it's also an example of where even though I'm not subscribing to
294:49 Bryan's mission of, I think it's the slogan's don't die. I actually
294:56 I do wanna die. I don't wanna do this forever. Like, I think the human li-
295:00 lifespan of about 90 to 100 sounds about right. I'm embracing the finitude of life.
295:08 We don't have to agree on that. In fact, Bryan is a vastly more interesting human to me because
295:14 we don't agree with that, and Bryan's willingness to literally put his own skin and all sorts of
295:23 other body elements in the game for that mission is really interesting. Now,
295:31 I can have that view, this is really interesting. I'm glad Bryan is
295:35 pursuing this passion of his, and then also find my wife's analysis to ring true, that it does seem like there is some
295:48 underlying, uh, I don't know, fear of... I was talking to Jamie yesterday about this, and she was saying,
295:55 Or comparing it to this sense of people who didn't feel like they had lived enough. And one of
296:03 the reasons perhaps that they hadn't lived enough was that modern
296:06 society now is a surveillance state, not by the state actually, but by each other. There are camera phones everywhere.
296:15 Every indiscretion or even outburst of fun or cringe is highly likely to be recorded and then shared
296:26 to the point of ridicule, and what is the rational reaction for
296:29 humans under those conditions is to pull back, is to make sure you
296:33 don't dance like nobody's watching because- ... they're probably filming you to your great and
296:39 eternal embarrassment on the internet, so maybe you just shouldn't dance at
296:42 all. And therefore, if we've all retracted to the point that we're barely living, we're
296:49 clinging on to wanting it to last forever. Now- ... I mean, I'm sure that thesis does not apply in all circumstances. I don't even know if it applies to
296:58 Bryan or, or anyone else here. But I think there's something to this, that
297:03 if you feel like you've really lived, you're okay thinking I've lived enough.
297:10 And that the, it having an end is not something to be fought. Now,
297:20 is it also possible that this is all just existential post-rationalization, and if
297:24 tomorrow there was a live forever pill, we'd all take it? Yeah, that's possible.
297:29 Is it also possible that if we did that, society be, would be worse off?
297:33 Yes. Is it possible that the human lifespan is the length it is for
297:41 sociological reasons, not just biological reasons? Yes. So I think these are just all interesting questions, and I just thought her analysis
297:51 really got to something here. And that discussion has come up
297:55 afterwards. I mean, that, I think that tweet is like a year old. But just
297:59 recently, we've been talking about this over-optimizers. I saw,
298:03 uh, Chris Williamson, Modern Wisdom had a bit where he was like, "Yeah, okay,
298:07 maybe we did go a little overboard." And the triggering thing was
298:11 the Diary of a CEO, where the host was saying something to the
298:15 effect of, "I had a glass of wine and the next three days were ruined."
298:20 And that to me, I mean, in his circumstance, I'm sure that's true. Like, if you have not
298:25 drank any alcohol for a very long time and you have just a little, I could see how it can have effect.
298:29 But also there's something in that that just triggered a visceral reaction, I think, for a lot of people, and myself included, where I was like, "Fuck,
298:37 man, I should have a glass of wine just right now, even though I don't usually drink."
298:41 And in fact I think the alcohol question is a good microcosm to zoom in on. So
298:46 we've all stopped drinking apparently, like alcohol sales are way down-
298:50 ... certainly amongst young people. And then on the one hand, on the
298:53 pure health metrics, whatever, we can go like, "Oh wow, isn't that great?"
298:58 And then I go like, "Yeah, I'm not so sure it is." Like the social
299:02 lubricant that alcohol can provide may be missed, is missed. I mean, we are in an absolute
299:09 epidemic of loneliness and misery and depression and whatever, and a fair amount of it just probably comes because you're not
299:17 interacting enough with other people. Now, that's before we even talk about
299:20 coupling, right? Like way down, people having a very hard time meeting a spouse.
299:26 And- ... a lot of that has moved over to apps that have all sorts of
299:33 negative outcomes and consequences, right? Like maybe we will
299:36 look back, or maybe we already are looking back and thinking like, "Eh, do you know what?
299:40 Maybe getting wasted every once in a while was not the worst thing in the world." And whether wasted or not,
299:44 maybe just having a couple of drinks every Friday or Saturday when you're out and about was
299:51 part of what helped society get to where it is. Maybe the fact that humans have literally been drinking for whatever it is, like
299:59 13,000 years for mending things, had a social purpose, and we pull out Chesterton's fence of our own chagrin, right?
300:08 - As you mentioned, in the spirit of that, so when I traveled across rural
300:12 China, I did smoke a little bit in my early 20s, but I picked up just for that trip-
300:19 - A social connector, right? - And I drank humongous amounts, even though I don't
300:26 really drink these days, because of the social thing. It's the way they show love.
300:31 And I also ate a shitload of carbs, which- ... I'm usually low carb. I just like meat. So just
300:38 because it's the way they show love. And actually even, it's in that society cigarettes,
300:44 but in most societies, liquor is a thing to do together. And it's a bond.
300:50 It doesn't, the benefits that that entails can, in many situations, outweigh the negative consequences like
300:59 long-term for whatever, however you measure them. - And this is the myopia of modernity, that we reduce
301:06 things down to like, well, one glass on this long run study meant that you
301:10 had a 0.1% greater risk of some heart defect. Yes, but if you didn't live at all, what's the point? And again, it's also
301:19 easy to flip over the other side. Like the only way to live is to get drunk every
301:23 weekend and eat a bunch of crap and smoke 40 cigarettes a day. No, it's not. Like it doesn't have to be the
301:30 either/or. This is one of the reasons I stopped wearing my Oura Ring. So I wore
301:34 the Oura sleep ring for many years, four years. And one day I was just like,
301:41 "I don't need to know that I had-" ... "a bad night of sleep."
301:45 Like I know. This added, I don't wanna call it an anxiety, I don't have a lot of anxiety, but this added just like
301:54 reminder- ... that I had a bad night of sleep, what purpose is it actually playing here?
302:00 Why do I need to have all these stats? Why do I need to optimize everything? To the point where, again, after that single
302:08 anecdote, you know, from Diary of a CEO, I was like, "Fuck, I kind of want the opposite." Like-
302:13 I barely drink at all. A glass of wine once a month maybe. Like, do you know what? I should start, uh, just on the weekends, get a glass.
302:24 Maybe two. And also, I mean, I've never been the person to fret about my diet. Um,
302:32 this is one of the great privileges of being married to, and I'll do a quick aside here. Don't ever
302:39 call a woman in her 40s a wonderful woman. My wife was so pissed—
302:45 ... that not only last year did I come on her birthday, that was bad enough, but that I called her a
302:49 wonderful woman on the show, because that sounded like her name was Nancy and she was 65— ... in the suburb of Rochester.
302:57 So I'm like, I was trying to play the 50. Like, no, I don't know, I'm Danish. How would I know? But that didn't go over well. So I'm gonna say
303:05 awesome woman instead. If you marry an awesome woman who also not only knows how to cook, but enjoys
303:13 it, do you know what? There's a very traditional division of labor here that has been
303:20 quite stable through several ... tens of thousands of years in
303:24 human societies, that perhaps also wasn't the worst idea in the world. And I also know there are plenty of men who like to cook,
303:32 and there are women who don't. But as stereotypes go, eh, that certainly helped me. Like, I don't think I would've been able to just
303:41 be as relaxed about food if I didn't eat relatively healthy on
303:45 a, on a regular basis. But the overall point being, just getting out of that optimization game without turning into the
303:53 opposite. You don't have to be a fat slob on the couch either. You can move around.
303:59 - Yeah. And then the bigger point we're talking about is the finiteness of life.
304:05 - Yes. - And I agree with you on this. - This was actually a feature, I built it into Omarchy. So memento mori.
304:12 Remember death. If you pull down the calendar in Omarchy, you open it by clicking the clock.
304:18 It opens up, and the first funny thing, I took it off the internet. There's this
304:22 account called Year's Progress, and it, every day it updates, and it just
304:26 shows 63% done with 2026. I don't know if you've seen that one.
304:30 So it just fills up. So I built that into the calendar. So right now it says, "We're 63% done
304:34 with 2026." But then if you double-click that, it'll ask you, "When were you born?
304:39 How long do you expect to live?" And the default is 90. So I put in 1979.
304:43 And then right below the line of, like, how long are we done with the year-
304:47 - Oh, that's beautiful. - ... is a new line that shows how long are you done with your life.
304:50 I think I'm at 62%. I'm like, if you hover over it, it says, "Memento mori."
304:56 - That's beautiful. - Fair number of people found it morbid, but this was why it is an Easter egg. It's not, like,
305:01 exposed. You gotta click a few things to get there. I like the reminder. I like the
305:05 reminder that time is finite, and I should make the most of it.
305:09 And this is also one of the privileges of having children. You really
305:14 realize that the cliches are all true. Oh my God, it goes so fast.
305:19 And I remember my now 13-year-old being one, being two. I was like, "This was just yesterday." And the reason it's a
305:27 cliche is because it's a shared experience. Cliches are good.
305:31 They remind you that you are part of the human experience, and others have been here before you.
305:38 - I have to ask, you mentioned the amazing woman and great cook.
305:43 - Amazing. Very important to remember. Not wonderful, amazing.
305:47 - Amazing. So there's this... You talked some shit about croissants or something. I saw this,
305:54 this in a tweet. So is the Danish croissant somehow better? - Actually, Danish croissants are not particularly good. In my opinion, the only place you can find a
306:03 half-decent croissant, of all places, is 7-Eleven, which is hilarious to me because 7-Eleven in the US is the place where,
306:11 at least in Chicago, you'd go to get a slushie and try to avoid getting murdered.
306:15 But in Denmark, 7-Eleven is this very upscale, super nice convenience store, and they just happen to make decent croissants.
306:25 And it always baffled me that I lived in the United States for 20 freaking years. I have yet to have
306:33 a single croissant that's even at the level of a 7-Eleven croissant in Copenhagen.
306:39 That seems improbable. - Have you understood why? - No, I've legitimately considered hiring an investigative
306:47 journalist to get to the bottom of this, because the number of theories I've heard have
306:51 been all over the place. It's the water. It's the butter. It's the skill. It's the this, it's the that, it's the other thing. I'm like, none of
306:59 it adds up to me. If it's the butter, can't you just import it? I
307:03 thought I saw Danish butter in some supermarkets over here like Lurpak.
307:07 So that can't be it. Can it be the skill? No, it can't be the skill. There are plenty of French
307:11 people and Belgian people, the best croissant makers in the world, who live in
307:15 the US. Like, what is the reason? And then I get the second thing, which is-
307:20 ... "No, no, you just haven't gone to the right place." I've literally gone to
307:23 15 of quote unquote- ... "the right place," and all the croissants were shit.
307:28 So before I leave the planet, I have to get to the bottom of
307:33 why the United States cannot make a proper croissant. And I don't, I'm
307:37 not even asking about peak French croissants, peak Belgian croissants-
307:42 ... which are the best croissants in the world. I'm just asking about the 7-Eleven croissant
307:46 from Copenhagen. Is that too much to ask for? - That's a fascinating mystery. What, zooming out, what's
307:55 your, what's your favorite... This makes me want to know, what's your favorite meal?
307:59 Like, if you, if I gave you a last meal and were to murder you after.
308:02 - It all depends whether it's lunch or it's dinner. If it's lunch, it-
308:05 - It's one meal. What's, what's... I feel like if you're getting murdered,
308:10 it doesn't really matter if it's lunch or dinner. - Well, depends on what time of day you're getting murdered.
308:13 - Oh, I see. - If it's an afternoon execution, I only get lunch.
308:18 - Okay. - And if I get lunch, it's at Louisiana, the museum in north of Copenhagen.
308:29 I was just there with Jamie two days ago, and we go there all the time.
308:35 The museum is lovely. We don't go for the museum. We go for the cafeteria. That in itself sounds crazy. Like-
308:40 - Yeah, it does - ... you're gonna, you're gonna drive half an hour to go to a cafeteria.
308:43 Well, it just so happens to be, it is the best lunch I've probably ever had in my life. The chef there is just
308:51 unbelievably good, and not only is the food amazing, what's even more amazing to my system thinking
308:58 brain is that at any one time when we were up there, I think there were probably
309:02 200 people waiting for their food. And our food was delivered in five
309:07 minutes. They have a small menu, and clearly they make it in bulk, and it's
309:13 unbelievably delicious. - Just a cafeteria at a museum? - It's a cafeteria at a museum that happens to have the best lunch I've ever had anywhere in my life.
309:21 - What kind of food are we talking about? What kind of lunch food?
309:23 - It's, like, Nordic cuisine, so it's, it's... A funny thing is the menu, when I
309:27 look at it, I was like, "Do you know what? If I saw this menu- As a kid, I'd go like,
309:31 "I'm not eating here." It's like, I don't know- - This is pretty fancy
309:33 - ... root free fruit and- ... and onions- ... and all sorts of healthy stuff. That sounds disgusting. And then you sit down and eat it,
309:40 and the angels sing. But it's even more particular than that. It's the butter.
309:46 The whipped butter that they have there, the best butter anywhere in the world.
309:50 - I think that has to be connected to the croissant mystery too.
309:52 - There must be some connection. But literally- ... even in Paris, I've never had whipped butter that tastes as
310:00 delicious as the butter does at Louisiana in Copenhagen. Literally,
310:04 I would drive 30 minutes- ... to have bread and butter at the cafeteria in Louisiana.
310:10 - Well, how'd you discover that? Just one day you went to the museum and-
310:13 - I think we just went to the museum- - And you tried a whole dish.
310:15 - ... and we had the lunch, and we were like... - This is weird.
310:20 - This is rivaling the best tasting restaurants- ... I've ever been to in my
310:24 life, and I've gone to Noma a couple of times and tried what there is to try, and this is right up there.
310:31 - All right. What about dinner, since you mentioned? Is it something different
310:35 that you have to think- - Well, I think there's a lot of good options in
310:40 Copenhagen, but actually I'll pick a place in Spain. So there is
310:46 this place in Marbella called Puente Romano. It's a hotel, and down by the beach, they have this beach restaurant with fresh seafood.
310:57 Unbelievably delicious. I'm trying to remember what it's called, but yeah, if you look it up, the beach
311:05 seafood place at Puente Romano is incredible. - How much of great food is, like, the place and the person you're with,
311:14 right? Or the, or the experience when you first are there? - I mean, the ambiance certainly helps, but I think my taste buds are also just-
311:22 - Okay, that- - ... in tune for- - So it's legit. - ... flavors
311:24 - ... it is just good food. - It's just really good food. Because for ex- that cafeteria, it is, there's a lot of people. Like,
311:31 I wouldn't rate it as like, oh my God, this is just a place I want to sit for two hours.
311:35 But the food is so delicious. - That's one of the things I'm, in traveling across the country, I'm realizing
311:41 to sample without bias, just try things. 'Cause you might, I guess you might just be surprised. Maybe I'll find your
311:51 croissant. All right, we've talked about mortality. Talked about food.
311:56 - Settled end of life. - What do you thousand years from now, what do you think is the future of humans,
312:02 human civilization? You think we're gonna make it? - A thousand years from now? Man, I'm having trouble predicting 12 months from now.
312:09 - I know. - A thousand? - You think we got a shot? - Oh, I'm gonna bet on optimism. I'm gonna bet on a multi-planetary
312:15 species. I'm gonna bet that Elon gets that rocket to Mars, and we figure
312:19 out how to bend space-time and discover some wormholes or, or something.
312:24 - Once we do, do you think, you think all humans on Earth will die once or twice and be repopulated
312:33 After we get a good backup going? - I mean, that was actually one of my favorite episodes of Black Mirror, the
312:39 one where they get uploaded to the 1980s sea town somewhere. And I just thought like, wow, this is
312:47 both such a dark moment, yet also so beautiful in the shared recognition that, like, the '80s were actually kind of amazing.
312:55 Like, I don't have a lot of nostalgia for many moments, but I do
312:59 have nostalgia for the '80s. - But- - And they say something about, like, whatever place and time you were in, the music you were
313:05 listening to when you were 12 or- ... or 14, 15, something like that, that's what's gonna stick with you. Like, no, I was
313:11 only, I got to be 10 in the '80s. So this was my early childhood, and I
313:17 still have an unbelievably fond affiliation with the '80s. So if I'm gonna be uploaded to the sky-
313:23 - You want it to be the '80s? - ... I'm doing the '80s. - Wow. I don't think I've actually ever heard anyone say that.
313:29 Usually it's '70s or '90s. - Oh, and the- - Because the '80s is a-
313:32 - I really don't like the '90s. Can we talk about this? - No, I'm done. Not done.
313:40 - To me, this was the great turning point of nihilism. That both the music, the genre, the fashion-
313:50 ... everything turned from, like, this glamour, the pop, the optimism, even the yuppies and everything- ... to grunge and Nirvana.
314:00 And even though, I mean, I like the music, I don't like the ethics. I don't like-
314:05 ... the morals. I don't like any of the underpinnings to it.
314:08 There was just such a defeatism to it- ... that in retrospect, I look back upon it, if I'm gonna get reinserted
314:15 into the matrix, it's gonna be the '80s. - That was full of optimism and fun-
314:20 - Yes - ... and colors. - Yes. I remember having these orange pants-
314:29 ... with, like, white dots on them. That was just a normal thing kids could wear in the '80s.
314:35 I've never fucking seen that today. Or I mean, even in the '90s, it all turned just,
314:42 like, Seattle gray. Great regression. - So the Nietzschean, you know, this idea of eternal recurrence.
314:50 If you do a Groundhog Day and you live forever, you want it to be in the orange pants
314:54 with the white dots in the '80s. - Maybe I'm gonna skip the orange pants, but-
314:57 - No - ... I'm definitely picking the '80s. - You said it. All right, DHH. Thank you so much for everything you do. Thank you for being an inspiration to
315:05 all of us, for building cool shit in the world, and thank you for talking today.
315:11 - Anytime. Thank you, Lex, again for having me. - Yeah. Thanks for listening to this conversation with DHH. To support this podcast, please check
315:19 out our sponsors in the description, where you can also find links to
315:23 contact me, ask questions, give feedback, and so on. And now, let me leave you with some words from Ralph Waldo Emerson.
315:31 "Once you make a decision, the universe conspires to make it happen."
315:37 Thank you for listening, and hope to see you next time.
$

DHH: Future of Programming, AI, Agentic Engineering, Vibe Coding & Linux | Lex Fridman Podcast #501

@lexfridman 5:15:51 23 chapters
[AI agents and automation][developer tools and coding][content creation and YouTube]
// chapters
// description

DHH is the creator of Ruby on Rails, Omarchy Linux, CTO of 37signals, and a racecar driver. Thank you for listening ❤ Check out our sponsors: https://lexfridman.com/sponsors/ep501-sb See below for timestamps, transcript, and to give feedback, submit questions, contact Lex, etc. *Transcript:* https://lexfridman.com/dhh-2-transcript *CONTACT LEX:* *Feedback* - give feedback to Lex: https://lexfridman.com/survey *AMA* - submit questions, videos or call-in: https://lexfridman.com/ama *Hiring* - jo

now: 0:00
// tags
[AI agents and automation][developer tools and coding][content creation and YouTube]