David Heinemeier Hansson
0:00 ↗
There are decades where nothing happens and weeks where decades happen. And we have seen decades of progress happen in the last nine months. If you're not recognizing the gravity of the moment, that's the delusion. That's the psychosis. Who would not get delirious if suddenly a genie pops out of the bottle and says, "You can have whatever you want. Every feature you've ever dreamed of in an operating system, I can deliver them to you, most of them in five minutes, a few in 20, and if we really go hog wild, it's gonna take me two hours." I want the fastest car breaking the speed limits. I want the diver watch that can go down the Mariana Trench. I want the operating system that can install in less than 60 seconds. One of the things I always loved about race cars was when I would stumble out of the car absolutely smashed and barely able to hold my head up, and I'd lay down on the garage floor and just think, "Holy fuck, I'm alive." The Overton window does not open itself. It opens one nudge at a time by people risking a little. A little reputation, a little pushback, a little criticism, or maybe sometimes a lot of reputation or a lot of criticism or a lot of pushback. Creating life with another human that you love is literally the peak experience of being on the planet.
The following is a conversation with David Heinemeier Hansson, also known as DHH, creator of Ruby on Rails, CTO of 37signals, creator of the new Omarchy Linux operating system, best-selling author, race car driver, and one of the most outspoken and prolific programmers in the world. For the past 20-plus years, for him, programming meant meticulously handcrafting beautiful Ruby code. But recently, since late 2025, DHH has embraced the AI revolution and has openly transformed himself into, once again, one of the most outspoken and prolific agentic engineers, even though he hates that term. So let's say practitioners of whatever programming is becoming where AI is doing most of the actual programming and the human steers the ship with high-level design, vision, and taste. Yes, it's true. DHH is at times controversial, but he's always fearless, brilliant, and fun to talk to. This was, once again, an intense, eye-opening, and wild rollercoaster ride of a conversation. This is a Lex Fridman podcast. To support it, please check out our sponsors in the description where you can also find links to contact me, ask questions, give feedback, and so on. And now, dear friends, here's DHH.
13 months ago, we sat down right here to talk about programming, and then everything changed. At that time, you were a bit skeptical about the role of AI in the process of programming, and then we went through this rapid evolution of agentic engineering. So, simple question to start: How has your view on AI's role in programming changed? Are you excited? Are you terrified? Are you on an emotional rollercoaster ride?
David Heinemeier Hansson
3:26 ↗
I am incredibly excited.
You've been talking about fun a lot.
David Heinemeier Hansson
3:31 ↗
There is none of the existential threat. That doesn't exist for me as an emotional component. It is only there as an intellectual component, and the emotional component for me is 100% pure, unadulterated joy and optimism and amazement that we've made computers do this. And I find it so interesting that we talked just 13 months ago because it's like we talked in different universes, different eras. I love this quote. I think it's Lenin. There are decades where nothing happens and weeks where decades happen. And we have seen decades of progress happen in the last nine months. I mean, imagine you're there when the Wright brothers take flight. Yesterday, the New York Times would write, "It's gonna be 10,000 years before we fly," and then the day after, we're up in the skies, and just a few years after that, there were cross-Atlantic planes. The 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 humans got to live within the same epoch that they were born in? They never saw that complete change of the world and of society. And to have been blessed with two of those feels just such a privilege. I got to see the internet from pre-internet to post-internet, and then now pre-AI, post-AI. What an amazing run. How fortunate.
Yeah, but the thing is, this feels like a thing that happened faster than anything else in human history.
David Heinemeier Hansson
5:18 ↗
Correct.
So I would say somewhere around December, maybe late November.
David Heinemeier Hansson
5:22 ↗
November 24.
David Heinemeier Hansson
5:26 ↗
That's the exact moment.
You know, when they talk about when one nation invades another in a world war or something like this, this is how we talk about AI changing everything. Yeah, and it really just shifted for many great developers. It shifted to where AI is doing some basic autocomplete, maybe writing 5, 10, 15, 20% of code to writing 80% of code.
David Heinemeier Hansson
5:49 ↗
Or 100.
Or 100, yeah, especially if it's not a public-facing product.
David Heinemeier Hansson
5:54 ↗
To me, that's why even when I look back upon our conversation a year ago, I don't actually have different 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, 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 way of looking things up on the internet, not what was gonna replace me chiseling code. But then we get the agents. And the agents start out being curiosities for about five minutes, and then they get amazing, and then they get, "Oh my God, is this AGI?" And all of that happened since last year, even just within the last nine months. We have basically these few phases here. We have AI in the pre-agentic era. I was excited about that, but it was not fundamentally rewriting the rules of the game for me. It was not completely changing how I worked. I was still chiseling code, I just had a little helper, a little sidekick 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 already doing, and it didn't change the emotional connection I had to the computer. Then we get to November 24th, 2025. Opus 4.5, to me, was the dividing line, where suddenly I didn't even try it on the 24th. I think I tried it on the 26th. I give it a couple of tasks, and I realize that the quality of the output is uncannily close to what I would've written. And I remember just leaning back and thinking, "What just happened?" "How did we go from this autocomplete mess that 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 intelligence and then also the increase in usability?" This agent harness question where I don't know if Opus 4.5 was that much smarter than Opus 4, which is what we had in the summer, but its ability to instrument your computer, to use tools, to check its own work, 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 then the big change that happens for almost anyone who paid attention and started playing with it over the Christmas break. This is what I heard from Shopify and other places with lots of employees who suddenly had a breather, suddenly had a couple of weeks to lean back and just look at what was going on, gave it a try, and had the same mind-blowing experience that these agents were of a different genre than what we had before. And then what we get to is, I'm already excited at this point. Like, by December, I'm already revisiting sort of all my priors and going like, "Wow, if it can do this, can it also do that?" "Oh, yeah, it can." And again, Opus 4.5 now looks like a retarded model. And this is the magic of this progress, is you think you've reached something. I remember thinking at the moment, "If this is the last model we get, I'll be set. I'll be happy." 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 amazing to see an agent do all this work in the way that I wanted it done. Because it was not just about it being able to solve a task, it was also that I could look at its path there and go, "Yep. Yep. Maybe not there, but almost," and it would give you two notes. You'll get to where I wanted to go in the way I wanted to go there. You could produce code I wanted to merge. You could produce code that actually looked beautiful if it was written in Ruby. Rust, different question. But this ability for the agents to truly become an extension of how I wanted to work was very novel. But then we wait just until early spring, and suddenly we get sub-agents. We get harnesses that can subdivide the task, and something that would take Opus quite a while suddenly took a fifth of the time, a tenth of the time, because it could get chopped up and suddenly you got eight sub-agents working for you. But both of those two first phases of the agentic age, 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 when it went off, and then we'll get there, and I would go much faster. But I had to be in the driver's seat. I had to tell it what I wanted, and I had to be the reviewer, the auditor of what was coming out. And then finally now, this summer, with Opus 5, Fable, and Sol, GPT Sol, and 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 going. I'm telling it the problem I have. I'm telling it the fuzzy, 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 because I'm a curious person and I like computers and I like the outcome of it, but I really kinda don't have to. I've become optional in the part that produces the code that picks the route. Remember when early GPS systems came out? They were amazing compared to looking at a map, but you still wanted to pay attention. Is it gonna drive you in the harbor? I remember these newspaper articles. "Ah, GPS, they're terrible because people don't pay attention, they drive in the harbor." When was the last time GPS drove anyone in the harbor? Like, that just doesn't happen anymore. In fact, the cars now just drive themselves, right? And this is where we've arrived at now, that I can trust. For the domain I'm working in right now, I can trust it, 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 able to recover.
Mm-hmm. We should mention that there's all kinds of domains that programmers operate in. There's I actually don't know, but I think the most common domain in development is, like, web dev CRUD. You have a database. You have a user-facing interface, and it does something back and forth, and it could be internal to just one person, to multiple people, to a small number of people, or to the world. And I think for that, it's I mean, you could really get to the 100% of code written by AI and really if you're a good programmer and you have a good intuition about what's happening behind the scenes, you can legitimately not look at the code, at least that's, that's been my experience. I don't know how much prior experience with programming you need to have to kinda know that things are working correctly behind the scenes, like intuitively by observing the ripple effects, the symptoms of the system. But I think for that domain, it's close to 100%, and then there's domains like you're also operating in, which is writing a Linux distribution. There, maybe there's more because you're obsessed with speed and all that kind of stuff. There, maybe you need to look at the code a little bit more. Then there's maybe safety critical systems all the way down to operating a nuclear power plant or a self-driving car. Maybe you need to look at the code more carefully.
Yes, but that said, AI is insanely capable at both finding and fixing security vulnerabilities. This was the whole blowup about Fable. This model was so capable of finding holes that an attacker could exploit that it was simply not safe to release. So the irony here is that when you look at that field, it seems like we've reached levels of intelligence that virtually no human can match because many of these security holes are about stringing combo moves together. You find one little vulnerability here that by itself might not be the worst thing in the world, but then you combine it with four others, and suddenly you have RCE, remote command execution. Humans who are able to do that are very rare. They usually work inside state-sponsored organizations or other clandestine operations. They're not just out and about finding things. So they've gotten so good at that. The Linux distribution is why I've gotten 100% AI pilled. Because I've been working on Omarchy for the last three months, this version that just dropped a few days ago called Quatro. And almost right from the beginning I'm working on that version, the agent acceleration neared 100%, and in the last two months it has been 100%. I have not written any of the code that's shipped in Quatro by hand. I've reviewed the shape of all of it. I've reviewed the individual lines of anything that's critical in the model layer of the system, and I've not looked at a bunch of the UI code. 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 the web part, actually evolving Basecamp and Hey, our professional products that have lots of users and are relatively large code bases, have proven surprisingly tricky to fully accelerate with agents. We just released Basecamp 5 not too long ago. That was the first product at 37signals that was really agent accelerated. Because we were in this final sprint phase from around February. By then, agents were already good. And we had this early surge of, "It's solved. We can just have the designers do the programming. They know what features they want. They know what shape they want it to take. Let's just let them vibe." And we let them vibe. And we ended up with a lot of PRs that individually perhaps could have been justified for a hot moment, taken all together, destroyed the architecture of the system. And we actually had to clean up manually, mop it up by hand, by human hand, to get back to an architecture that felt cohesive and coherent. So we still have a bit of that that was February, by the way. Things are quite different now.
Well, hold on a second. What's the lesson from that? Is one of the lessons from that that you have to be a programmer at this stage to be able to vibe code?
To be able to vibe code on existing substantial code bases, even if they're crud, if you wanna retain the element of architecture that got that system to where it was. Now, that's also a point where I've stressed many times that when people accuse vibe coders of being slob generators, I go right back at them and say, "Have you looked at the average programmer's output?" That is some other slob too. If you've looked at behind the scenes of many great companies and what their code bases look like after there's been 3,000 humans through them, they're awful. Absolutely awful.
Can you tell me the intuition you have? I seem to just, stepping back and observing the different apps that we all rely on, I don't know, Adobe Photoshop, all this kind of stuff. There seems to be the progress on development there has not accelerated. So why is it, what lessons can you draw from Basecamp, like well-established, huge user base? Why aren't we seeing, like, super rapid increase in, like, new updates, features, all this kind of stuff in these well-established apps?
Multiple reasons. I'll start with the first one that's the most critical. As soon as you're having human teams work together on something, the bottleneck is rarely 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 CTO above them, and everyone wants to be part of the shaping process because we're all justifying why we're here, that's where all the productivity goes to die. The revelation I've had working on Omarchy the last three months is that to get that magical 10X, 100X, in a few rare cases, 1000X productivity boost, you have to interact with the agents directly, and you cannot intermediate that 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 humans, and it's great to work together. But it also means we need to temper our expectations with what these agents can do if it's humans driving it, and you have three layers of approval and all the other machinery of a large corporation. The implementation part is only a small segment of it. The other thing I'd say is that most organizations don't know what they want. They don't know how to make it better. They're not bottlenecked on implementation. They're bottlenecked on ideas. They're bottlenecked on vision. They're bottlenecked on taste. And if you don't have those element in excess of your implementational capacity, it doesn't help. So you can make a lot of shitty ideas come true, then what? 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 replicate here. That process is not actually Because we already had this. If you step back for a moment and think of the towering organizations we have who have had tens of 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 them in this case because they have had endless resources, endless programming capacity for decades, right? This is what's showing us that just being able to write a lot of code does not 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 life cycle capacity of internalizing what's going on. And I think it's actually funny that the critique of AI is, 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 as AI, and you're impatient because in the last three months we haven't rewritten the whole world and made it a utopia of software goodness?
You're basically saying everybody should be switching to Linux now. One of the arguments is, like, we can rewrite all the software that's not available in Linux in Linux. Linux has been the love of my life for many years, but one of the reasons I'm still attached to Windows and now, uh, Mac is because of video editing, Premiere.
We're gonna fix that. But.
So the question is, who's gonna build Premiere and Photoshop for Linux? And it feels like one person can now.
And so, and I'm allowed to be impatient. In fact, me or anyone else, impatience is the first step of doing it yourself, right? Like and there's a community of tens of thousands of people that use Adobe Premiere or DaVinci, all these non-linear video editors, that know the frustrations. Everybody shares them. You can look at Reddit. You can look at the forums. Everybody knows it. Uh, I don't know what it is. It may be the management. It may be the meetings. All the things you've always discussed is the The bureaucracy that's inside companies that's slowing everything down in the development of features. But it feels like there if you just let some of the developers loose with a bunch of agents, you can fix all that. But again, based on everything you're saying, maybe the way to do that is to start over from scratch, maybe through open source, or somebody launches a new company that within large companies now it's hard to to really accelerate into the agentic era by letting developers just build.
This is the classic innovator's dilemma. These companies have gotten so good, so established at the old way, and therefore their entire structure, management layers, processes are tuned for a time that no longer exists. But you can't pivot that. These are super tankers. It just doesn't happen. This is why we're finally getting this upset in the technology industry. For a while, I was very upset about the duopoly between Apple and Google on mobile, because it felt like mobile was the most important computing platform that we had, and I could not see a path to unseat either Google or Apple sitting on top of that, controlling everything, taking their toll booth money. But the game has changed. This is no longer the most important platform. Mobile phones are important, but there's also a lot of other form factors coming, whether it's glasses or it's earpieces or whatever else have you. It's all in play now, but the computing platforms themselves are in play for the first time in probably 40 years, if you look at the desktop. Linux has been around since '91. 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 fridge, your toaster, everything runs Linux, except for your computer. Now, funnily enough, your Android is actually Linux, but wrapped so sufficiently that you can't really recognize it. But now there's an opening, and the opening is exactly as you say it. If there's a piece of software you depended on that bound you to Windows, it is completely within reach for you personally to start rewriting it. And maybe you won't get to 100% coverage, but this is the old joke about Microsoft Office. "I only use 5%." Yeah, well, we all use a different 5%. Well, what if we all just build our own 5%? What if I just took the functionality that I need and just did that? That is a completely different challenge and one agents are incredibly capable of doing right now, 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 become a polyglot programmer, something I was absolutely not before. I was a Ruby programmer first and foremost, and then I dabbled a little bit in Bash when I had to. And in the last two months, I've written C++. I've written three applications that have shipped in Omarchy Quattro. I wrote a writing app. I was using this app called Typora, which is a very nice app which in itself is based on another app called iA Writer, which was my real love on the Mac for a clean, simple Markdown writing environment. That's where I write all my essays. 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 shareware, and it had a lot of features I just didn't need. And about six weeks, seven weeks ago, I thought, "Do you know what? I only need 5% of Typora," which is already a basic app. I literally told the agent to get going, that I wanted it written in C++ and Qt, because that would fit well with the aesthetic of what I was building with Quattro. And in, I think, about 20 minutes, it had the first version. I started using it, and it wasn't quite right. Within two days, I'd given up Typora, and I wrote and have written all of my essays since that moment in Omawrite.
Can I ask for your advice and your vision about something? So what I find with agents, you can actually write something like a replacement for 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 starts with one user in mind but expands to a larger audience. So I've actually written a large number of software just for me, and I find it 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.
I've never, unlike you, built something that's used by a large number of people, and I find that a bit scary and intimidating. So, like, at this, with this agentic age, how do you take a step to actually build something that solves your problem, which I think is a beautiful way to build, but also expand it to actually useful for other people? Like, is there advice you can give, and do you also see the problem of how easy it is to just build a tool for yourself?
Yes, but it is simpler than you think. By the time you're done building the tool for yourself, you simply tell your agent to put it on GitHub. You don't even have to do anything else. It will figure out how to put that repo on GitHub. It'll write a nice README. It'll start using GitHub releases so that you can track things. It'll actually be a better software maintainer than you could ever be because it is far more patient, it is far more diligent in doing the drudgery of managing open source software than you are. And you can simply get on with the joy of using the software that you built and evolving it as you see fit. Now, this gets into the big argument we've been having in open source for a while. Is AI contributions good or bad for the maintainer? There's a lot of maintainers right now who are quite upset about the fact that they're suddenly getting a huge influx of pull requests, of contributions from people who may not be the best programmers or programmers at all towards their software. I look at that argument and go, "Are you kidding me?" Here is a vein of free contributions that you can take or don't take, but you're complaining about the fact that they're there? It sounds straight out of My steak is too juicy and my lobster too buttery. What are you complaining about? This is the most amazing thing. This is what we proclamated for so long was going to be the magic of open source, that we could all contribute to it, but it was, of course, never true. There was only a small subset of people who contribute to open source, and that was the very highly skilled wizards. So this is a bit of a reformation moment. We had this disintermediation between us and the computer, and it was this class of clerics and priests called programmers, and suddenly they're being disintermediated here by was it the 95 theses by Luther in 1500s, nailing them to the door? And one of those was very clearly about the intermediation, that you should have direct access to the higher intelligence.
Mm-hmm. But don't you need the wizards to keep a high bar of excellence in the code base?
100% you do, which is also something that's always been true about open source, which is the second argument that really grinds me. I've been running open source projects for 25 years. I have literally looked at the output of thousands, if not tens of thousands of programmers. I feel like I have the statistical basis to assert that most programmers, they suck. And I don't mean that in the way it sounds. That was why I paused for a hot second here. I mean, they suck in the sense that they don't write the code I want to have written. They don't prepare their bug reports with all the relevant information. They don't detail their pull requests with the why. They don't bother to fill in needed code comments. They don't double-check their work. They don't write unit tests. They don't do all of these things it takes to create good software. But do you know who'll do all that stuff? Agents, if you tell them to. They're very diligent at following your instructions, most of the time, sometimes to a fault. But it actually means that if you take the median programmer and their pull requests towards an average open source project, they're already getting outclassed by agents. I would rather get an agent-written pull request to one of my projects than I'd get one written by a human, and it's not just because the quality's better. It's also because I feel a lot less bad if I just reject it. You didn't even write it, so I can simply look at what your agent wrote on your behalf and 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 open source maintainers. They have way too much ball of neuroticism and anxiety that requires them to look upon any contribution as an obligation on their part to do 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." And if you get into the mode of realizing that your project is allowed to exist and evolve according to your vision, according to your roadmap, it becomes so much easier in the agentic age to decline the contributions you don't want because you don't even inconvenience or hurt the feelings of a human. It's just a clanker, and the clanker won't mind. In fact, the producers of clankers, the labs, love when you spend tokens in vain. They get paid just the same. So, in my opinion, this is the absolute best time to ever have been an open source software maintainer. Not only do we get this wealth of glorious pull requests made by 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. I've been running the Omarchy project now for a little over a year, and in the last three months working on Quattro, I have merged over 1,000 pull requests. Quite a lot of those pull requests were written by people who were not classical programmers, or were programmers in other domains, not Linux operating system or distribution development. They were able to contribute their good ideas because of agents, and I got to cherry-pick the very best of that because it was there. Otherwise, those ideas would just have lived inside their heads. Again, is that is not the purpose of open source, that we tap into the collective intelligence and creativity of the whole goddamn planet, and we channel that towards a commons where we all benefit from the combined efforts of everyone. Now, there's currently, I think, about 400 unmerged pull requests on Omarchy, about double what it was a week ago. So the acceleration is real, but so are the tools. I'm not reviewing every pull request anymore. I haven't been reviewing them for quite some time now. I have the agents review them for me, and then they will give me a summary about whether something is ready for the decision, the human decision. Should we merge or should we not merge? I don't need to look at all the pull requests that are either wrong, duplicated, bad. An agent can sort that chaff away from me. I just get to look at the pearls the good ideas that are ready to go, the bug fixes that the agent has validated in a VM on my behalf. All the drudgery of replicating and managing open source projects is evaporating at lightning speed, and we are left with the with the golden juicy parts, the bone marrow of software development deciding what should this thing do and where should it go.
Do you think the good ideas, ultimately the kernel of the good ideas originated in the human mind, so the individual contributors Like, can it be done by agents, or is the PR ultimately requires a human idea? Like how to improve Omarchy, all those incoming Or can it all just be a pool of agents?
That was the mode and the way I was thinking about agents in the first agentic age from November 24 to February 28. I thought all the ideas would originate with the humans. They would tell 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 believe, ideas coming out of models so great that it makes me humble as a person who otherwise prides himself on having good ideas. The agents are incredibly capable of creative thought, and folks who are still stuck in the analysis that agents are parrots just regurgitating the ideas that are already there are delusional about the progress that's been made in the last six to nine months.
So some people listening to you right now will say DHH is suffering from the old case of AI psychosis. Can you steel man the case that you are in fact experiencing in a state of delusion, like One Flew Over the Cuckoo's Nest, and can you argue against it?
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, and I've never seen the things that I've seen in just the last two months. A whole career, a whole life dedicating to the love of computers as my main focus for the working hours and then some, suddenly completely upended, suddenly completely rewritten. Of course I'm delirious. Who looking at this reality is not? If you are not delirious or at the very least very excited Actually, I guess that's not even true. If you're not recognizing the gravity of the moment, that's the delusion. That's the psychosis. The psychosis is believing that the world is barely different. We just have some electronic parrots reciting things to us. Now, in my opinion, the reason this argument is actually brought up is because people are not able to see the fruits of the progress. That goes to your point earlier. Where is the amazing software? Why isn't American GDP running at 12% year over year? First of all, give it a minute. It hasn't even been a year, goddammit. But second of all, I feel on my own account, I have actually brought the pudding. Omarchy Quattro launched on Friday. It's been downloaded by tens of thousands of people. And they like it. They like it a lot.
We should mention that going to Perplexity here, Omarchy is a highly opinionated Arch Linux-based desktop distribution centered on the Hyprland Wayland tiling compositor. It was created by David Heinemeier Hansson, DHH, as a polished developer workstation set up with a cohesive visual style and many everyday tools already configured.
Let me put it. Omarchy is a beautiful, modern, and opinionated Linux system. It's an alternative operating system to macOS and Windows, and it's freaking amazing.
And the Quattro thing is you're pushing towards the age of agents.
That's right. That's the latest version that we just dropped out, which is also funny. 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 where I had watched one too many YouTube videos on Linux Rising and got bitten by that bug. Started working on this second iteration of my attempt at creating a better Linux operating system. The first iteration we talked about last time was called Omakub, was built on top of an existing system called Ubuntu, and it was fine. But the ambition I was able to pour into the new version, Omarchy, because I started seven layers deeper down the stack, was completely different, but it was still done in a pre-agentic world. I wrote all those batch scripts by hand in the beginning, and then I got to see the midway point where things changed over. I did a couple of versions of Omarchy that had partial agent acceleration, and then three months ago, it was full throttle, 100%, everything is written by agents but steered by me. And that just ended up being a very different experience and a very different operating system because I was suddenly granted a limitless ceiling on my ambition. I could look at any feature in Windows, on Mac, on other Linux systems and say, "I want that," and the agents would deliver. So everything I wanted was suddenly within reach, which meant that I got a little delirious. Who would not get delirious if suddenly a genie pops out of the bottle and says, "You can have whatever you want. Every feature you've ever dreamed of in an operating system I can deliver them to you, most of them in five minutes, a few in 20, and if we really go hog wild, it's gonna take me two hours."
Mm-hmm. Have you seen Requiem for a Dream, the movie? So the There's a drug-like effect here. I find it personally We'll talk about many levels of this, but it is overwhelming 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 with it, and I find myself truly overwhelmed with the multitasking, On the verge of almost burnt out. Super excited about so many things, but ultimately looking back, you want to focus on one thing and really ship.
I don't have that sense. I don't have the sense of overwhelm. I don't have the sense of dread. I don't have the sense of distortion because I have a mission. I'm going somewhere, and therefore, I can channel all this new power towards a singular goal and outcome, creating the perfect computer. And therefore, all my investigations with AI are focused towards that end. Now, I also have a day job. And we also use agents there, and that's also targeted towards sort of specific outcomes. But with the Omarchy experience, I got to tap in straight in the back of my head and increase the bandwidth between ideas arriving in my brain and software emerging on the screen. It was like going from dial-up to fiber. I think on your podcast, Elon talked about the actual human bandwidth that Neuralink and other projects are trying to accelerate. Like, the bandwidth we're communicating over right now is quite low because we're limited by our cognition. We're limited by the rate of speech. You're not limited by those things when you're working with a swarm of agents. Now, let me caveat this by saying, if I'd heard myself talk like this nine months ago, I would probably have used the label AI, AI psychosis. I think it would be fitting because nine months ago when people were talking like this, they weren't shipping, and I think that's the ultimate difference, that there were people who saw this early. I was not actually that early. As we've talked about, I was rather skeptical, and I was using AI in certain ways, but I was not the first person who downloaded Claude Code. I think Boris released that in end of February. I don't think I installed Claude Code until September. So there was about six months there where pioneers saw these glimmers of what the future was gonna look like. It wasn't there yet because the intelligence wasn't there. The harnesses weren't there, but they could see the glimmers. I couldn't. Tobi, Tobi Lütke, the CEO of Shopify and now my partner in crime on this Omarchy project in part, he's gotten pilled with Omarchy as well. 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.
Yes. I was still on Earth. And he was He had already boarded the rocket and was heading towards space. I remember actually reading his memo to the company, the AI memo, and thinking, "Ah, seems a little much." Like, you're making a big deal out of something that I cannot yet feel is a tangible thing. Because where's the output of this? I try using it. It writes code I don't like. It wants to interrupt me all the time. Where are you getting this from? Uh, so I get it. I totally get it. If you've not seen what the current quality of intelligence can produce, of course you're gonna look at someone and go like, "They sound a little nutty," because all your experience up until this point would tell you that they are nutty. Because if you look at the history of computer programming, for the last 40, 50 years, people have been promising that regular people could write code just by talking. We're gonna have fourth generation languages. Lisp was once upon a time presented, Smalltalk was presented as these environments that would let regular people write their own applications, and none of it came true in the sense beyond, like, Microsoft Access databases or Excel spreadsheets. That, those are end user programming environments, but it's nothing like what we're able to do now. And then, I mean, I'm not even talking about end user. I'm talking as me, someone who's been programming for 25 years, seeing the output and the acceleration and the quality and then shipping it, and I think that's what's gonna change the entire conversation quite quickly. The few laggards who are still on the fence about whether this is actually useful, who are still trapped in a meme from early '25 about mechanical parrots, they're simply going to be overwhelmed by the evidence that's about to flood over them.
So I hope to see projects like Omarchy, several of them that, like, really show agent first.
It's coming. All of it is coming.
In the spaces, in the silly spaces that I already mentioned, like video editing, all these kinds of apps that are like Typora this kind of stuff.
It's gonna come, and I've already seen the early inklings of this. One of the really neat features about Omarchy is that it has a far more robust plugin system that allows you to extend your operating system and rewrite it really in terms of its user interface and its panel and its features by creating your own plugins that can be cloned off of tools we ship in the box. For example, say you want a different calendar. Omarchy ships with a calendar. You click the little clock. It pops up a calendar, and it doesn't have iCal support, for example. It doesn't consume your appointments. There's about 17 implementations of that already, and in three days, we had 330 plugins on the Omarchy plugin marketplace. I have never seen growth like that with any project I've ever been involved in. I've never seen participation that broadly. I've never seen that many people be able to create software that's meaningful for them Also be usable for others so quickly. All of it is driven by the fact that Omarchy ships a set of skills that tells any agent you bring to it how to create extensions to the operating system, and therefore affording you the vision of the true malleable 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 slob speak at this point. Like, it's just slapped onto everything. I wish we had a different word that just meant AI doing stuff.
Vibe coding feels wrong, too.
It does, because vibe coding to me smells exactly like script kiddies did in the early 2000s. People applying PHP scripts they just downloaded offline they don't understand anything of. And while, I mean, that's true in the sense that much of the vibe coding, including my own vibe coded projects, and as I just mentioned, I have several. I've written things in C++. I've written things in Rust. Rust. I hate Rust with a passion. Rust to me is like pouring acid in my eyes when I have to look at the code.
Wait, wait, wait, wait. Why do, why do you hate Rust?
It is, in my opinion, the ugliest programming language that has been invented in probably the last 40 years.
So it's the opposite of Rails and Ruby.
It's the opposite. But I've written applications in Rust because the output of Rust is actually amazing. The memory safety, the fact that it's a system language, the fact that it's highly efficient and so forth is incredible. So Rust, to me, can both be the most repugnant programming language ever devised for human consumption and a wonderful platform for agentic engineering. 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 programming? Can we just change what programming means? 'Cause who's actually going to be programming the old-school way in a few months?
I don't think we should reuse the term because for me, programming implies an understanding of certain primitives, loops, conditions, variables. All the constructs of programming languages are what constitutes programming itself, not the creation of programs. 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 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 the vibe coder a programmer either. And vibe coding, if we define it here, is you tell an agent to build software for you. You do not look at the implementation. That, to me, is what separates vibe coding from programming or, let's say, agent-accelerated development.
But don't you think, so to push back a little bit, don't you think a programmer, old-school programmer, you, doing agentic engineering is a different kind of agentic engineering than a non-programmer doing agentic engineering? What I mean is, if you know what a for loop is, if you know what function programming is, if you know some of the basic good principles of software engineering, the way you do natural language-based agentic engineering will be different and more systematic, and this, the scale and the variety of things you can actually build is much bigger than a vibe coder.
I'm just gonna extend the pro case here a little bit just for the sake of the argument. I actually think for a while it was to my deficit to know as much as I know about programming because I was instructing the agents to do things as I prescribed them to do, and they were very good at that. And in the first agentic moment, let's not call it an era, it was something that lasted three months. Moment. In the first agentic moment, that felt very productive. I could get more productivity out of telling the agents what to do in the way I would've done it, and I used all of my experience as a programmer to do just that. Then I was a little late on the next moment, and the next moment allowed me, allowed anyone, to describe outcomes to describe problems to the agents and get better solutions than if you had a programmer prescribe the path.
See, okay, so we're having fun arguing here. Well so are you suggesting that there's problems for which programmers are worse than non-programmers at agentic engineering?
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 very good product managers. Software is product management. What should it do? Who should it do it for? How should it do it? How should it look? What are our priorities? What do we start with first? What does version one include? All of those skills are not easily or equally distributed across all programmers. And in the agentic era where you are letting an agent, an AI, do the implementation, these are the skills you need. And I happen to sit in both camps, and this is why I love this moment. I still get to live a little bit in the old world where I look at the full implementation because there are domains and areas where I feel that still brings value, and then I also live in the future. And as I said, when I created Omawrite, I've not looked at a 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. I will treat it as though I was any other user who had opinions about how their digital typewriter should work, and therefore felt like it was a fair experiment as if a writer of any other sort who had opinions about how software should be done and how it should work. As you mentioned, video editors who know exactly where they don't like what Adobe Premiere's doing and how it could be better, but don't have the capacities to improve it how that could be.
The pushback, though, I think what programmers do well, good programmers do well, is systematic design, like think in terms of systems and rigor. And I think still, and probably for a long time to come, even natural language-based control of agentic systems will require a certain kind of rigor. Not overly structured rigor, but really explain very specifically the goals of the system, how to do the verification, what kind of security testing. You wanna do all those kinds of things. Although then to push back on that, the system will should be smart longer term.
I would've said the same thing six months ago. And I think the same thing would've been true six months ago. What I have found since, especially even just the last few weeks of the agentically accelerated development of Mach-E, is that more often than not, I have the humility to recognize that the agent knows best.
You can legitimately say, "Make sure it's secure."
Correct. And it would know more about what that entails. Now, you're laughing... but it's true. And in fact, a good parallel to this is the AGENTS MD/CLAUDE MD files. There was a hot moment where it was all the rage about micro-optimizing that. Oh, you tell your agent this, you tell your agent that, and you ended up with this huge file of instructions for your agent. One of the things that Boris, working on Claude Code, shared in an interview recently about Opus 5 was that the system prompt that they ship for Opus 5, and presumably also Fable, shrunk by 80% because the agent not only needed far less human instruction, it was actually being damaged by overly prescriptive humans. And any programmer who's had a pointy-haired boss knows exactly what that is like. When the boss walks into the room, doesn't know anything, starts telling you how to program, how to code. What do you do? You sulk. You write shittier code if you're mandated to do things that are against your better judgment. Why would an agent not be the same?
Actually, to push back against myself, I do find the actual skill I'm currently developing, which I along those lines is probably what everybody's trying to develop, is not to overspecify things, is to get out of the system's way. Is to specify enough to provide a kinda high-level vision and guidance.
It's even better because you don't even need that. The fundamental insight of modern software development came from the agile software development movement. In the late '90s, early 2000s, a group of smart, honest, and brave people came together and said, "The way we've been trying to do software development for the last 40 to 50 years has not worked, will not work. We've been trying to specify upfront what software should look like. We thought that we could ask humans what they wanted, write it all down, apply our systems thinking and our rigor to a spec sheet, hand that spec sheet over to a group of programmers who would then implement just that, and everyone would be happy." Well, guess what? No one was happy. Because no one knows what they want until they receive it. You don't know what a program should do until you play with it. So in the agentic age, you should resist the temptation to be overly specific upfront. Be as vague as you can to manifest something, then interact with the something. The way you arrive at good software is you write a little bit of software, and then you try to use it. It is in the process of using software that you discover what you really want. It's where you discover what's important and what's not important. And agents are incredible at allowing you to bumble into this experience of realizing what you want because you don't know what you want.
Mm-hmm. And I've actually done something similar to what you suggested, which is implement different designs and totally different implementations, and then I create HTML pages for myself, PHP pages, where I vote on what I like more and so on. And it does this iterative thing, and it's very nice and fun. The interaction with the agentic system, if you create a nice interface for yourself, it can be super fun, and you're focusing on the design part and the most fun parts of the design part.
Correct. And this is what humans are really good at. Differential evaluation. You give me three options, I pick one of them. Now, humans actually fall off a cliff if you give them 22 options. That's the paradox of choice. But you give them three options, they very quickly within a split second will tell you what they like. There are all these ways that exploit the fact that humans will make these snap judgments that are actually pre-intellectual, right? Like, they're coming from the gut, and then they arrive up in the brain, and the brain tries to rationalize why the gut said what it said. And if you're willing to let go of some of that intellectual pre-processing rationalization and simply let your gut drive, the agentic age is a revelation.
So this is a fascinating thing to ask you because you have sort of famously for a long time talked about, like, detailed chiseling of beautiful Rails Ruby code, and now you have switched in a matter of months to not doing that. So, like, what is your current setup look like? 'Cause you're, I assume, are still looking for the beauty, for the chiseling, for the crafting, but you're just crafting in a different medium.
And at different levels of abstraction. When I'm working in Ruby code, and we have a lot of Ruby code because it's in our entire business, and I'm asking agents to make changes to that code, I still sweat the details. What I'm coming to realize is that the economic payoff of that sweat is diminishing rapidly. The reason why I, for 25 years, was sweating every line of code so judiciously was because I knew the payoff of keeping an architecture coherent and malleable was software that could change and evolve 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 the driving economic argument for why you should write beautiful code, 'cause beautiful code is easier to understand, it is simpler, it is more malleable. That was premised on humans doing the modifications. I think it is an open question to which degree this still matters. Now, it does matter, and the reason I say that, at least for the moment, is that tokens are still scarce. At this moment in time, we are all token limited. Well, not all, but anyone who doesn't have endless budgets are token limited. So therefore, there is great payoff to writing systems that agents have an easier time dealing with and evolving without having to relearn the entire context. Just like humans, if they can make iterations and changes to the code base without wrecking the architecture, they can make the next 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 because it's just put together in a way where nothing is connected and it's just a real mess, right? I've seen that with agents, and I've seen it in our own code bases where, again, the first PR is, like, mediocre of quality, and then if you add another PR on top of that, and then five more down the line, it's not very good, right? So there's still a payoff to that, but that payoff is premised on our current moment. I am now able to... And this is, by the way, where the AI psychosis really comes in, when you try to extrapolate what nine months from now it's gonna look like. What two years from now it's gonna look like. But as an intellectual experiment, I think the first computer I used was a Commodore 64.
It had one megahertz CPU and 64K of memory. Everyone who wrote software for that machine internalized certain constraints. Well, all the constraints. That machine was nothing but constraints. If you put that programmer in a time machine and teleported him or her to 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 because all their techniques and all their heuristics would simply be wrong. Or not wrong, because efficient software is still beautiful. They would be out of date with the value that they could create. The amount of optimization you have to apply for a one megahertz computer to produce a video game is just very different from what you have to apply today.
Just imagine that guy that was programming back then on a Commodore. Just remember all those, like what programming felt like and what it meant, highly constrained resources, how slow everything is. It's chiseling. But imagine that person coming now.
Beautiful in many ways. Beautiful. It's romantic too, though. Let's also imagine that. This is one of the things, whenever we bemoan modern car culture, for example, right? And then we think like, "Oh, wasn't it better when we all had the horse-drawn carriages and so on?" No, it wasn't. Do you know what New York smelled like when we used horses for transportation? It was an open sewer. They just shat everywhere.
Yeah, but have you played Red Dead Redemption? You know how cool those cowboys were on the horse? So basically programming, old school programming by hand is kinda like the cowboy culture that we romanticize and make movies about.
That is the way to experience the nostalgic past, when you don't get all the furry impulses and senses. It's already happening, the romanticization of handwritten code. And I have some of it because it was a very romantic era. I'm grateful to have been alive for 20 years of economically valuable handwritten code. That was a, that was a good time.
Where handwritten beautiful code was also valuable.
Correct. Because handwritten beautiful code exists as much today as it did yesterday. There are people who willfully write new video games for the Commodore 64, or for the second Mega Drive, or for other vintage consoles. In fact, one of my favorite video games of late is ModRetro's. Do you know Palmer Luckey's side business, which is recreating old consoles. So he's recreated the Game Boy. It's amazing. Look it up. It's really cool. But what's even cooler than the fact that he recreated this hardware and the original 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. Super cool. Wild attention to detail, input lag, and all the other things. But what's also very cool, that one, I have to wave. That one right there, and hover over it, 'cause then you get to see what I'm talking about. That is the Chromatic Tetris re-implementation. I played a lot of Tetris on the original Game Boy. 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. And that speeds up the game by about 400%. It is an incredible game, and it's a new Game Boy game. 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 games is awesome, and the reason they do it in part, now here they're doing it because they want Tetris, but there are people who do it just for the love of the constraints, for the love of the romantic notion of writing that. Just like there are people still riding around on horses, not to get places fast, but because they like the mode of transportation where it's a literal biological being below you propelling you forward.
And also, we should say that that beautiful code is great training data. So in some sense, the beauty that we've built over a few decades continues.
We gave birth to this moment. What a privilege, and I feel that very personally because I... Almost all of the code I've ever written is public code. The vast majority of my career has been dedicated to writing open source code, and therefore, some of that code, some of those beautiful lines are in the training set. And in fact, I've heard people do this when they write Ruby code. They ask the agent, "Write it like DHH would." And they're pleased with the result, and that does warm my heart a little bit, that I helped give birth to this moment in my tiny little small part. And anyone else who contributed open source code or any code at all that the agents have had access to over the last, well, entire time programming has been around as a discipline, is now part of these new agent overlords.
So you don't have a little bit of sadness that writing beautiful code by hand is now less and less useful?
Am I sad that if I had to get a job tomorrow, it's unlikely that I could apply my skills in such a way that I would be paid to write these manual lines of code? No. I did it for 25 years. Like, I wanna see something new here, especially if we're all gonna live forever, which I hope we don't. Maybe we'll get to that. But, like, I've done that. That's enough. That's fine. I don't feel any more nostalgic about that than I do about the fact that I don't have to spend my time in the field with a hoe or at the assembly line.
Well, hold on a second. You are one of the best programmers in the world at writing handcrafted beautiful code, appreciators of beauty with great taste. That has been replaced. Yeah. That has been replaced. Like, you had now... You're just fast evolving, and you're able to... One of your really great qualities is you're able to change your mind and evolve very quickly. But, like, you had to kill that other person.
Correct. I say humbly. I am grateful for every moment that brought me to where I am right now, and therefore, I have also tried, and maybe this did take some practice. I will concede that. Maybe I wasn't always like this, but stoic philosophy, amor fati, loving your fate, is an incredibly liberating way to live. And you could attack that and say, "Well, that's a point of privilege." True, but also I had the same directional point when I didn't have all this, quote unquote, "privilege." So I think there's a way to live and interact with the world where you accept the things you can change and the things you cannot change and fall in love with all of it. Fall in love with the fact that the world today has jumped two decades forward into technology industry from where it was last year.
So can we talk about the general anxiety that programmers feel, going through the same transformation? They've maybe gone to university, majored in computer science, dreamed of being programmers and building stuff, high salary, and now everything's changing, and there's deep anxiety about, "What do I do with my life?" Can you empathize with that, and what the hell are they supposed to do?
Hugely. Because just because I have this disposition, I'm keenly aware that that's not evenly distributed, that not everyone sits with this position and this optimism for the future. But I do think I wanna separate things a little bit. I think you're gonna have a hard time coping with the new reality if the only thing you loved about programming was the mechanical bits of putting the right logical constructs together to produce something other people told you to produce. Because that mechanical process is under threat. If you are, as you just mentioned, excited about building things, I don't think you're under threat at all. In fact, I think there's a great argument for us needing far more builders than what we have now, and if you look at the employment stats, it's fuzzy. It's not clear what's gonna happen at all. Some stats actually show an increase in openings because the advent of AI is so dramatically lowering the price of programs that people want a lot more programs. This is the classic, the Jevons paradox, that says when the price of something goes down, there's gonna be more demand for it. And this is the example of the ATMs too. When the ATM originally came, a lot of bank tellers were very afraid for their job because suddenly there was a machine that could dispense money from people's account. Well, what the ATMs did was lower the price of a branch, and suddenly banks could afford to open a lot more branches, and we ended up with more bank tellers than we did before. Now, none of this is guaranteed. There are also moments where things do change. The vast majority of people worked in the fields up until the late 1800s, and then suddenly we got mechanized harvesting tools, and we did not need human labor with a hoe and an ox out there. We had mechanical tools to do that for us. In that moment, are there people who were threatened that their job and their livelihood was at stake? Yes. Was that a reasonable thing to feel threatened about? Yes. The Luddites smashing the weaving machines in England had the same thing. And I think maybe that's the better parallel than people working the fields because these were actually highly skilled professionals doing a job that they liked on rather favorable conditions, not slaving outside in the sun all day. But where we live now, would we still like for clothing to be this heavily constrained resource? Like, what if this T-shirt was $400? I mean, okay, then I'd have two. Well, my wardrobe probably looks like I have two. But I think the general progress of mankind depends on productivity improvements. And what are productivity improvements? This is very important because in the abstract, I think everyone says, "Yay, productivity. That's good." What did you think, this was all just vibes? No. Productivity means fewer 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, but it also may not. There may be pockets where a company just needs a certain set of fixed tasks done, and suddenly they can do them with a tenth the number of people. This, by the way, could... is tragic and difficult in the moment for the individual being laid off. It's also amazing for the economy at large. Suddenly you've freed up these resources who can now go do more productive things. This is how the whole economy evolves and improves and how we get growth, and growth is good. I think this is the other thing to make a strong defense for, that growth in general is a good thing. We do not want to wind the clock back to 1920 or 1950. We should be excited about the fact that we've discovered new technology that allows us to do vastly more. And while some of the tasks that may be taken over by AI were tasks that humans were excited to do, and they're just no longer economically viable, a lot of the tasks were just drudgery. This is certainly true of my programming. There were moments of programming I thought were amazing. These were the moments that produced that state of flow we talked about last time, and they were ecstatic. How often did they happen out of the course of a year? If you took your average programmer, how much time out of 2,000 hours at the job did that person spend in a flow state? 100? 200? 25? I think all those answers could be plausible for someone. And taking that drudgery, handing it over to machines, that is the history of civilization.
Yeah, I mean, the hours spent debugging. So programming has so many painful, painful hours in it. And now they're gone. For the... I mean, for the most part, 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. So that's where some of the romanticism also comes in because pain in the moment is difficult. It's hard. It's annoying. Pain in the rear view mirror is accomplishment, proud, learning, steps forward, all the things, right? So we encompass all of this as part of the human experience, that we hate pain in the moment, but we cherish it when it's five minutes past and we don't remember the hardship. We just remember the progress.
So certainly at the societal level, at the macroeconomic level, growth is exciting. But the individual level, there's going to be a lot of anxiety, potentially a lot of suffering. So by way of advice, if you're like a young DHH or young developer programmer now, what would you advise they do?
Don't try to anticipate anything. You will literally go crazy. Because even the smartest brains in the business cannot anticipate what two model hops from here is going to look like. It's an absolute waste of time, and you will develop an AI psychosis trying to deduce what two years from now is gonna look like. Focus on right now, and right now is the most incredible time to be into computers. You can make them do the most amazing things if you lean in. If you maybe for a hot moment force yourself to learn where the state of the art is, I double dog dare you not to get excited about what's possible.
And would you recommend building publicly?
I think it's optional actually to build publicly. I think it's nice to do because then you get to be part of a community, and there's some camaraderie in open source. Not some camaraderie. There's an enormous amount of camaraderie. Open source is an incredible source of community and camaraderie, and it's also a great way to stem some of that personal existential dread. If you're amongst other humans, you can 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, and the future looks bright, and if we build a bunch of things together, we can do things none of us ever dreamed we could do. I was just talking to Ryan Hughes, who's been my main partner on Omarchy almost since day one, and we were literally talking about yesterday how when we started on this journey together, our ambitions were very modest. We're just like, "Can we get this distro to not crash at 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 we're leaning back and thinking, "How can we take over the world? How can we make Omarchy go to Mars? How can we do everything we've ever dreamed of?" And that level of ambition inflation is partly because things have gotten better with agents, but hugely also because we've surrounded ourselves with others excited about the same journey. And if you're just sitting in your little isolated cave worrying about the future, yeah, you're gonna go a little nuts. I think if we learned anything during COVID is that people will go nuts sitting in their little cave inside by themselves worrying about what's gonna happen. That is indistinguishable from depression, ruminating endlessly about things you can't control. I'm sorry to channel some Jensen Huang here, but that's just loser talk. You don't have to be a loser. You can choose to lean in and win. And win is very broadly defined as learning more, making more, contributing more, being part of more. And at the end of the day, what are your choices? You don't have a choice, mate. The future's coming whether you like it or not, so you might as well choose to be excited about it.
But I mean, we should emphasize the fact that planning is nearly 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 reason you go to college. There's a plan underlying that. But it is kinda nuts that it's impossible to really plan. Because, like, you really don't know 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 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 anymore. You'll just... There'll be this glowing orb that just you just talk to because if natural language is enough to accomplish all tasks in life, like, what is life about? Like, what are the major tasks you need to accomplish in life? If you can just do that with voice, and it writes all the programs for you, and all is systematized, and what, what are the jobs needed then? Maybe it's more about the service industry. Maybe there won't be programmers, like agentic engineers at all. And so yes, your advice is wise to focus on the moment, but that doesn't mean that it's just a crazy rollercoaster ride, 'cause we don't know 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 gonna pan out. That is my best advice towards fighting anxiety, is that you have a choice, to a large degree, of how to lean. Do you wanna lean into PDoom? PDoom is gonna happen whether you lean into it or not. You don't have control.
But there's also actual daily schedule of how much learning are you doing, how much building are you doing, how much diversification you're doing. And for example, I traveled across rural China recently, and I did it quite deliberately because you're talking to people in China, in rural anywhere, where time moves much slower, where there's no discussions about agentic engineering. And I started by going on the Great Wall of China. 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. But then I very deliberately wanted to find the timelessness of human existence, really plug into the fact that, okay, it always feels like everything's changing throughout human history but, you know, the universals are still true, of what matters in life. And I just wanted to make sure I keep connected to that.
You're not gonna miss anything. This is the part that actually grinded my gears in the early phases of this transition. Everyone was so up in arms that if you didn't... It's loops now. Oh, no, no, we're done with loops. It's graphs now. Oh, no, no, we're done with that. It's harnesses this, right? They're constantly 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 new portal, and it's all spilling out. That's exciting. But it also means that if I had just been backpacking for the last year, hadn't touched a computer, hadn't witnessed this agentic moment, and I just showed up yesterday, do you know what? I would've been caught up in two weeks. There's not any accumulation, which is in some ways a great relief. If you missed the past year, you can catch up to the frontier in two weeks. If you're a programmer who was out hiking the Himalayas for a year and you come back, you can catch up in two weeks. And this is actually the great credit to the progress we're experiencing. We have so many experiments running simultaneously right now that are constantly and ruthlessly sorting what works, what doesn't work. You don't have to remember or even be part of that entire journey. You can just show up for the results.
But the key step there, when you come back from the backpacking journey, is to be willing to become a totally new, different human because things are changing so fast. I mean, literally, if you left for the backpacking journey in October last year and came back in April or May, it's like, "What, what do you mean? We're not programming anymore?" This, it's a huge transformation.
That's right. You might as well have been in the cryo chamber for 100 years. It's startling, and I actually... I mean, I'm putting some of this on slightly for effect. I also recognize that it's okay to grieve for a hot moment. That's not my disposition. It's not how I do it, but I accept that that's a time-tested mechanism for coping with a sense of loss. And it's okay to have a sense of loss that the world that you knew and maybe you were very fond of is no longer the same.
I maybe I should say out loud and explicitly that I do have that, the Native American with a tear rolling down my... Like, there's a sadness. It's a goodbye. It's a goodbye to the old world of programming. I've spent so many years...
That's, it's so fascinating to me. But it all amounted to this moment. That is, to me, what makes the entire journey so meaningful. AI did not arrive from the sky. It was not alien technology that had blasted through the universe and suddenly showed up on our shores. AI has been around for as long as computer science has been around. In the '50s, they thought the problem was gonna be cracked in about a decade, the smartest people at the time. It took a little longer, but the reason it took a little longer was in part because we took some blind turns on the road to AI and neural networks and thought that there were other symbolic representations that were gonna carry the day, and they didn't. And that probably set us back about 15 years. So some of it was also we had to go through the gaming revolution, 3D games. If we had not had Quake, if we had not had Duke Nukem, if we had not had Unreal Tournament, we'd never have gotten AI because we would never have gotten the GPUs, and therefore we would never have been able to get AI in the shape it is now. So everything we did up until this moment was required to get there. You were part of that. Holy... What a blessing. What a blessing to have contributed in your small part, either by training or by playing video games. Literally, you could just choose to think of it that way. All those times you spent in, for my case, in the '90s just playing video games, they were...
Yeah, I was contributing to the AI revolution right there and then moment.
To the progress of human civilization. No, for sure. Civilization progresses with the death of the old and the birth of the new, and it continues in this way, but it's still the grieving process is a part of it, I think.
Take a moment. It's okay. Take a... And also, you don't even have to accept that it's all dead. Like, if you are still very attached to the hand chiseling, which if you'd asked me a year ago, I would have predicted that I would've been more attached. I'm surprised a little that I've not been more attached. But the reason I've not been more attached, I think, is that I found something more fun. And if it hadn't been like that, if AI had just replaced the thing I loved and gave me something I hated, I'd probably be a little bitter.
I mean, this really feels... It's hilarious and awesome to watch and inspiring to watch. It's like Picasso all of a sudden like, getting access to image generation and getting super excited and switching in a matter of months. That's literally what happened. So, like, it's an artist that appreciated beautiful 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 work where he learned his craft, it was painting realistic paintings, right? Like, he was training to be a master of the old ways of depicting reality the best way we knew how, and then along comes all these other forms of expression. Cubism comes along. And all these forms of abstract expression, he's not decrying that he's not doing Renaissance paintings anymore, that he's not just painting the perfect depiction of an apple. He's reimagining the apple to be a freaking square, and he's getting excited about that. And I feel that transition, that we've gone from this one mode of expression, and we've suddenly opened the gates where the color spectrum is so much wider. I'm seeing more. And it's perhaps because the way I became a programmer. I did not become a programmer because a deep love of if statements. I became a programmer because I wanted programs. I wanted things to exist that did not exist. And at first, I had to learn how to program to make that happen. Then I happened to fall in love with programming as a craft, and then I spent two decades really diving deep on that. But now I'm back. I'm back to having an idea and 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 of a rediscovery. Now, I understand that that's not true for everyone. There are plenty of programmers who just loved programming right from the beginning, and they were attracted to the logical constructs and so forth. I mean, I've been there for two decades, so I get it. But I also think that you're able to discover new sides of yourself if you let it in. Again, I keep coming back to this almost regret minimization framework, to borrow Jeff Bezos's terms here. Are you gonna look at yourself two years from now and think, "Oh, I spent my time well being pissy about the present, about the fact that my industry changed"? Or would I two years from now look back upon this moment and be proud 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 learn how the world spins now, and I'm gonna make not just the best of it, I'm gonna make more of it."
I mean, it is true, a lot of programmers, a lot of friends of mine, me, obviously you, are having fun.
I have had more fun with computers in the last three months than at any time previously.
Quick bathroom break, and then I gotta ask you about how you actually interact with the systems now. I gotta ask you about how has your programming setup changed? So keyboard, voice, what's the IDE?
It's crazy to think about now, but yeah, I used TextMate for almost 20 years. I used TextMate starting in 2005, I think. I helped get the first version out, and then I just wasn't interested. I wasn't in the market for an alternative. And it wasn't until the switch to Linux that I was forced out of my habitat. And now with the switch to, what are we calling it? Agentic engineering? Oh, I fucking hate that term. We gotta come up with something that sounds as plain as programming, but encapsulates the fact that it's with agents. But...
I still think it should be called programming at this point.
All right, let's just call it programming. Programming with agents requires a different tool set. It really does. And the main change here is that you're going from single thread programming in your head to parallel processing. When I was writing code, chiseling it by hand in TextMate or even Neovim, not that long ago, I would just focus on one problem at the time, and I would methodically work my way through it, and that was actually the portal to flow. The portal to flow was deep immersion into a single problem, see it through to the end. 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 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 for a bit, and therefore you realize, well, if I just sit around waiting for them, first of all, that doesn't feel productive. Even if the agent, just one of them, can be highly productive, it does not feel productive. It does not feel good. It feels actually like you're a little bit useless. And maybe I had a moment, when the first agentic moment was there and I was running mostly one agent at a time where I felt like, "I don't know about this." But you can solve a lot of hard problems by simply throwing more resources at it. This is the whole scaling law of AI of itself, right? That if you parallelize these things and you're not running one agent, but you're running a handful, you can feel like you're in a flow state because you're constantly doing programming work in the sense that you're making decisions, and you're helping either unblock an agent because it has a question about which direction to take, or you're ready for a new task. And to do that, you need a different setup. I started first doing it in tmux and just having separate panes and having separate splits. Basically, a terminal with tabs is a good way to think about it. You open a bunch of tabs. I think most humans know exactly how that works. They don't work in just one tab.
So still sticking to the terminal, CLI. So you're not using Claude Code app or the Codex app or the...
Absolutely. No. I love the fact that this agent revolution was kicked off in the terminal because I was already a huge fan of TUIs, terminal user interfaces, and the terminal in general. That feels like a really nice place to be. It's a beautiful place to be. The modern terminal is just a good-looking place to work. So I like that. And then this fact of having multiple agents, especially once it's not just multiple agents running on your own machine, but you start running multiple 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 called Herdr. And Herdr is essentially tmux plus agent notifications. So whenever your agent is done and needs something for you, it goes ding, a little bell telling you it's ready for its human, uh... Is it master or servant? I'm not quite sure always. 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 multiple Herdr setups actually running on individual machines. I went on this crazy phase just about a month ago realizing that doing this work on a single machine is not fast enough. It's like I've discovered multi-core programming, but I only have two cores. I'm like, "What if I had 16 cores? What if I had 32 cores? What if I had 64 cores?" So I instantly went out and I bought these amazing KVMs called GL.iNet Comets. And what they do is it's this little box. You plug in HDMI, you 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 what's special about this is just how easy it was. You connect this thing in, you go to a webpage, log in once, set one password. Now this thing can hop on your tailnet. This has been the other revolution of the last year for me, is discovering these WireGuard networks. Tailscale is essentially turning all the computers you have into a local network wherever you are. Like right now on my phone, I have direct access to all the computers in my Malibu office. I also have access to all my computers in my Copenhagen office. And I can treat them as though I sat right next to them without having to punch holes in a firewall or set up complicated VPNs. And what this does is it just decreases the friction it 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 of mini PCs from prior experiments, and I just said, "What if I just connected all of them?" And I just connected four of the computers in a closet. They all had their little Comet, and suddenly I could run agents on more computers at the same time, and I could control them all with Herdr. And it did get to a point where I maxed out my own processing power. That I think at about... What do we want to put it? About four to five machines running, I don't know, three agents. I have about 16 threads. That's what I can run. And the faster the agents run, of course, the fewer threads I can run, but at the current pace, I can run about 16 threads...
Mm-hmm, at full acceleration.
And that's part of why I've gone from just being excited about the agent moment to being delirious. Because I think as you say, some of it is a little assaulting. Again, we're on a dial-up bandwidth-wise with our computer. I mean, it's actually hilarious. So you think of I'm sitting here a year ago, right? I'm chiseling my code. Yeah. I'm writing my lines. And over the course of, like, an hour, I will have written one beautiful controller, one beautiful model. Like, that's one file. And I've really worked on it, right? So maybe there's 60 lines left. So 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 running 16 threads. I'm producing at some times hundreds of lines of code 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 in general to measure these things, and I think it's right from some in the programming community to ridicule the AI psychosis when everyone talks about the number of lines of code they're writing, and then you ask them, "What did you build?" And they're like, "Well, I..." And then no good answer comes out, right? It is a nice shorthand. It is a representation or encapsulation of how much output that's coming out. Now, whether the output is good or bad is not a referendum on that. But I was producing and producing so much more now, right? And therefore I'm able to keep all these threads going. 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 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 if GitHub was a little faster at showing you your pull request, the web is probably actually a nicer place to do that. I don't know. There's also this other tool I've been playing with a bit called Hunk, which just produces diffs in a really nice way. So you could look at that too. But I find that when I look 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 changed this file, but actually, what do we have in this other file that wasn't touched but maybe should've been touched? So that's why I still like Neovim as a way of doing it. But it's all happening, by the way, of course, in Omarchy. So it's all happening on Linux. And this was the other major breakthrough with agents. Agents love the Unix philosophy. It loves individual tools that it can invoke through the command line, and there is no operating system on Earth of the majors, I'm counting three here, Mac, Windows, Linux, that works as well with that mechanism as Linux. Everything in Linux is either a config file or a CLI tool. Now, that was its main drawback five minutes ago. This was the reason people didn't like 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 drawbacks of Linux five minutes ago are now its major selling points. This was one of the things I thought I was stuck for a weekend with a Mac four months ago. I thought I had a computer at the place I was going that was a Linux machine, so I didn't bring my laptop, and I found out when I arrived I only had a Mac Mini. 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 Linux. And you can actually do a lot now. Homebrew has gotten really quite good. Homebrew is the missing package manager for the Mac, and no one has done more to move the Mac forward in terms of ease of setup. But still something as simple as Raycast, I don't know if you've used that. There's no config file that you can just access. You have to go into the GUI, export a file, then take that file, I don't know, in your freaking backpack on a USB key. And then you can import it somewhere else. You cannot automate the entire setup of your machine. You can't automate at all the configuration of Mac's default key bindings. That has to be a manual process where you're clicking with a mouse like a caveman to set up your machine.
I mean, there's ways around it.
Not good ones. I looked. I tried hard.
So here's my, been my journey. Obviously I'm a Linux person, but because of Adobe Premiere, Adobe products, I'm also a Windows person, so oftentimes I would use WSL, Windows Subsystems for Linux. So Linux inside Windows, which is... in the agentic era is not a good kind of Linux, 'cause it's just...
... it's a sandbox, and you want the Linux to be unleashed to be able to do everything. And so I have now woken up to things like Raycast, and have you know, I'm a keyboard person. Where's the config file? What, can I, can I basically do everything where an agent can set everything up for me and save the configuration so I can replicate it across systems and everything's automated? And so I had to ask a lot of those questions, and a lot of them are missing.
Correct. They don't have good answers. But there are hacks. There are hacks. You don't have to live like this, Lex. There's a better... Actually, this is a great moment. So this is a good moment. Um, I heard it was your birthday. So I brought a little gift. We are going to get you onto the agentic operating system whether you like it or not. And I asked my friends at Dell whether maybe they had a machine that, um, I could get you for your birthday. So here's the machine I use, which is... well, the same one I use, but it's now yours. It's a Dell XPS 14, already set up with Omarchy 4, ready to be configured for you. And once you realize that all your... By the way, we should almost time this. You can be up and running. You have to answer, like, five questions in about one minute. You could literally do it live if you wanted to.
But oftentimes you could actually just build the app yourself. Yeah, they're hacks. Until you have a video editor. Yep. No. This is awesome. Oh, no. Oh, Omarchy, beautiful, modern opinion Linux by DHH. Yeah, let's do it.
Yeah, fill it out while I talk. So exactly what you realized, that pain on the Mac was the exposure I got for about a weekend, and I tried really hard, and I found some decent workarounds, the main one being that Homebrew now has tabs for GUI apps. Because it's hilarious to think upon the recent past where people would go to websites, like, "I need Zoom." They would enter zoom.com, find the download link, download a DMG, double-click on that, see a weird manual interaction telling them to drag an icon from one place of the screen to the other, and that was how you installed apps. And you'd just do that over and over again until you had all the apps that you wanted, and if you were setting up a new machine and you had, like, 12 apps, you'd go to 12 different websites to download them like a total caveman. Homebrew has solved a lot of that, but it has not solved all of it, and has solved nothing in terms of the...
Sorry to interrupt, but look how fast this is going.
Oh, it's gonna be done in about 40 seconds. And then you're gonna be inside of Omarchy.
Super is the Windows or Command key on your keyboard. So which one is the Super key?
Yep, that's the Windows key. Oh, actually, no, no, you have a special edition. There are only about six of these computers in the world. It says Super, and look at where the Copilot key is on the other end of the keyboard. That's the Omarchy logo. This was Spencer Bull at Dell, who I've been working with on a lot of this stuff. They've been incredible. Yes, they did, last year. It was not just dumb that it got rid of the XPS branding. They also got rid of the Escape key and put in capacitive touch buttons at the top. The mistake that Apple realized like three years ago was terrible, and everyone hated them for them. But now they have an incredible team, and this was actually my first Dell laptop for myself. Because I've been using Dell on our servers. Dell was what powered our cloud exit. We've been running Dell since the mid-2000s, I think. But I never ran their laptops, because compared to a MacBook, they just weren't up to the task for me. The XPS' from 2016, or from 2026, with those 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...
Oh, there's a Super key. Nice. Okay, cool. They, for a moment, let go of the XPS brand. And I was like, "This is dumb. This is... XPS is awesome."
Originally, I thought, like, "Do you know what? I should just get a Dell," because we already used them and whatever, and I got a Dell, and I was like, "Man, I can't do it. I can't do it. It's not good enough." And then lo and behold, this year they just fixed everything. They fixed the battery life. They fixed the performance. They fixed the compatibility. They fixed the weight. They fixed the screen. That's a tandem OLED screen—
—which means it's actually better tech than what you'll get in a freaking MacBook. It's lighter than a MacBook, and the chip is finally competitive because Intel spent a good five years developing the 18A process node. It took a long time, but now it's here, and it's freaking incredible. We finally have competition to Apple M chips that can go as long on battery and can have competitive performance.
And there's a beautiful default theme to this thing, and then what is it? Super Space opens up.
Actually, you can just push the Omarchy key. It opens up the menu. So now you control everything from there.
Oh, same thing as the Super. Okay.
Same thing as Super Space. So this has been a huge change for me, that we finally have Linux machines that you don't have to make excuses for. Because I don't mind making excuses, because excuses are really just another word for trade-offs. I love that Framework 13, I think I talked about last time. I still like them. They still make good computers. But I think for a lot of people, they didn't want a DIY experience. They wanted something that just out of the box felt like it was on par in build quality and whatever with Macs, and now we finally have it. So happy birthday.
Thank you so much. An Omarchy Dell laptop from DHH is the best birthday present. This is awesome.
Check the screensaver, by the way.
Yeah, it's kind of amazing. I'm seeing it.
On that tandem OLED screen—
—it is absolutely mind-bogglingly awesome.
Look at that. There's like a retro look to the whole thing.
It's like running a bulletin board system in the '90s, which is exactly what I did when I was 14 years old. So for me, it's a complete throwback and an awesome aesthetic, combining this sense of modern rising culture in Unix with this throwback of the '90s.
Yeah, like a retro look for an agent-first OS.
Which again, pairs... This was why I was so excited that all these agent harnesses arrived as TUIs, arrived as CLIs, arrived as a terminal, because suddenly a whole new generation of both programmers and regular users discovered the glory of the terminal. The terminal was actually one of the most impressive and productive user interfaces we can have. Now, we've built user interfaces since then that were easier to learn. 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 the terminal is for people who want to learn their systems, who want to learn the keyboard commands.
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 these labs now have trillion-dollar valuations, so they need to be able to reach billions of people. And I don't think we're gonna teach billions of people the intricacies of the TUI. So we're gonna have other tools, and we already have other tools, which is, by the way, one of the sweet things about Linux, the ChatGPT Codex tool.
Literally within two hours of that announcement dropping, I had it wrapped up and ready to go in Omakub. Already now, if you go to... You can actually hit the Omakub key and then just write ChatGPT. It should pop up as, like, install. And then just hit return.
Yeah, whatever you put in. And then it's just gonna install it, and it's gonna pop it up, and you have ChatGPT.
And that's how a lot of the apps are, and all the agentic harnesses are built in that way. Tailscale, as I talked about, if you wanna hop on that, is that way. Dropbox. And the reason all these things are in there is because I use them. It's because I built this system for me, and I thought, first and foremost, how do I 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 as I do.
Now, you're obsessed about getting the installation to be under one minute, and a bunch of people asked you online why. So why? Why were you so obsessed with getting the installation to be less than 60 seconds?
First, let me quote Mitchell Hashimoto, who created Ghostty terminal, that's awesome and also available on Omakub, and was also the founder of HashiCorp, who had this great post on X not too long ago that had the killer line, "The pursuit of excellence deserves no explanation." Wanting to make something as good and as fast and as beautiful as possible has no need for justification. We should want beautiful, fast systems that are a delight to use. Now, the funny thing with the install time is it didn't actually start with a one-minute mission. I was being modest when it got going. I thought if I could set up my entire system in 15 minutes, it would already be such a dramatic improvement over what came before it that I would be perfectly happy. And it came off the fact that the benchmarking had been skewed so out of whack. You talked about the Commodore 64 being your first computer, right? When you unwrap that Commodore 64 and you hit the on button, in I think about less than one second, the BASIC interpreter was ready to accept your command. It was literally almost instant. There was virtually no boot time. Do you know how long it took me 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? Brand-new computer, out of the box. I set it up. There's a software update. Before I was ready to install Adobe Lightroom, that I had it fully up to date, it took 42 minutes. That's grotesque. That's absurd. That's an insult to everyone who likes computers, that you get a new computer and before you can use it, you have to spend 42 minutes doing updates on something that was already installed on it out of the box from the factory. That is just preposterous, and it's crazy that that's not even the worst thing. I also got a PC, because all these Linux machines, they come as PCs, right?
And they have Windows pre-installed. So I got one. This was just three weeks ago. Brand-new machine, Intel Panther Lake, the whole shebang. An hour and 35 minutes from unwrapping the thing until it's ready to use. How long did you take before you think you were in Omakub?
I mean, it's less than a minute for sure.
A minute. We can build computers like this. We have the technology. We've had the technology since like 1981 when the Commodore 64 first came out. The fact that we now have a computer that's a trillion times faster and you have to spend 42 minutes or an hour and a half setting it up before it's ready to use is so infuriating to me that it became a passion. And it's funny, just while we were having a little break here, I was talking to Ryan Hughes, my co-conspirator here. We now have, we think, a line where we can create these special turbo images that are pre-made for specific computers because it relies on some knowledge of the hardware, like the Dell XPS. We're gonna make a Dell XPS turbo image that can install in about 12 seconds. A full Linux system installed in about 12 seconds.
That obsession is what gives me this great satisfaction with using computers, that we do not accept things the way they are. And this is why this agent-accelerated reality that I now inhabit is so invigorating. Because I can look at these kinds of constraints and go, do you know what? I could have stopped when we were at five minutes. That would already have been fast enough. This was the argument on X, right? Why do you keep going? Five minutes is fast enough. Okay, maybe for you, but for me, 12 seconds is a lot more fun. And I always try to think these first principles. Okay, in that machine, actually hit the Omarchy key again, and then do speed. You'll see a list come up, and one of them is, like, disk speed.
Speed test, network speed, disk speed.
Disk speed. Do that one, and then run it. So this is testing the performance of the built-in NVMe drive, and what are you getting?
So seven gigabytes a second. That's how fast that drive is. Why can we not install a Linux distribution in one second? The Omarchy distribution is 5.8 gigabytes. If we can run at seven gig... Now, of course, you can't actually do that because the USB key is a lot slower than that, but you should be able to maximize the underlying physics and protocols that we're dealing with, and that's your goal. We gotta get all the way back down to that.
Yeah. I love that you're doing a Linux distribution. 'Cause I mean, you could be building anything. But the fact that you're building a Linux distribution means you're rethinking about what a computer is, 'cause an operating system is the fundamental, like, the foundation of how you interact with a computer.
Yes. And it's been so stale.
The macOS we have today is scarcely different from the macOS we had 10 years ago. In fact, it's worse in many ways because Apple continues to chokehold the control on users. They don't want them to install software that they haven't already said good to go for. They don't want users to reconfigure their hotkeys without doing it manually. They didn't want to change the 500-millisecond animation switching workspaces. It is an infuriatingly locked-down computer. Now, to Apple's credit, it's a pretty good computer for being locked down, but I don't want a locked-down computer. I wanna own my computer. Better yet, I wanna mutate my computer, and this is where the agentic age needs a new operating system. When you can vibe code whatever app comes to your mind, you should be able to vibe code your operating system. You should be able to change anything, how it looks, how it works, what panels are there, what are not there, and that requires Linux. As simple as that. None of the other two operating systems can deliver this.
But can you speak to the, I think somebody asked you, like, "Who the hell are you to think you could do better than Ubuntu?" For example. Or, like, "Why are you doing it when Ubuntu is already pretty good?" And your answer was, "I thought I could do better." So can you speak to, this was on X, can you speak to, like, where the craziness comes from, thinking that you as one person or maybe with Ryan, you could start building something that's going to be better than Ubuntu or Arch or any of this?
Well, I'll say that it didn't start quite as grandiose. When I got going with my Linux journey, I actually just started with Ubuntu, and I just built on top of it. Here's a system that already does a lot of the things I wanted to do, and at least it's Linux. So can I just build on top of it and make it more modern, make it actually look good, and have the programs I wanna use? I got, like, some version of that, and then I discovered that there's seven layers deeper you could go, and as the closer you get to the metal, the closer you get to the kernel, the closer you get to the individual packages and so forth, the more freedom you discover. So as I discovered more and more freedom, I got more and more ambitious, and I realized, oh, surprise, surprise, I have strong opinions about how computers should work, how they should look, what programs should be installed, and whether they should take 42 minutes to install or 45 seconds. That's the current world record, by the way. So if anyone installs Omarchy Quattro after this and can beat 45 seconds, take a screenshot, send it to me, and you'll be the new holder of the world record.
When, how long ago did you cross the 60-second mark?
It's funny because it's like that the old limit on, what was it? It was a four-mile mile or something like that in 1952 that people thought was impossible for humans to beat. Then one guy beats it, and literally within the next few months, three more guys do, right? So we had this goal. I originally had the goal it was two minutes. And then we beat two minutes, and I'm like, "Well, if we beat two minutes, why can't we beat one minute?" And I remember thinking like, "That'd be crazy. You can't install a modern operating system in one minute when they're all..." 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. No, no, it's not the bar because there's no speed limit. Like, none of the things you take for granted are fixed pillars of reality in almost any case. Until you get down to the seven gigabytes per second that drive is able to transfer, you've not reached the limit yet, so you should just keep pushing. And of course, then it becomes a game in itself and a satisfaction itself. Every second I could shave off was an excitement. And when we broke the one-minute barrier, which was really only broken, I think, three weeks ago or something like that, maybe even two weeks ago, I just got crazy. And then I got a swarm of agents to run all these auto research loops of trying all these different theories on like, oh, if you change this thing, if you change that thing, what if you cut this out? What if you do things in parallel? What if while you were typing in your username, we start preloading packages into memory system, we can fire them in faster. And it becomes a fun game. And I think that's how you should think about product development. It should be fun. It should be frivolous. It should be trying to overshoot the target and making it so much better than what it quote unquote needs to be.
Especially now because you can move so quickly and you can try so many different approaches because of the agents.
Yes, that just makes it more satisfying. But I think this was always true. If you look at the greatest Mercedes-Benz of all time, I think it was called the W126. I saw this wonderful documentary on it. It was steered by this guy who for years had worked in Mercedes' department for safety. And suddenly he gets put in charge for the whole S-Class project, and he just puts in all of it. This is where you got the auto-retracting seat belts and washers on the headlights and everything. This car was so overbuilt for the market. It was unbelievable. But you look at that and go like, "Well, I want that." In just the same way that you look at a mechanical watch and go, "It could go down to 100 meters?" Or I think Rolex has one that can go down to 4,000 meters, the deep ocean, the deep diver ocean or whatever it's called. No one outside of three people in the world, whoever buy that watch, will ever put it to the test. But don't you wanna be the kind of person who is a patron of people who wanna push the human race that much forward? I want the fastest car breaking the speed limits. I want the diver watch that can go down the Mariana Trench. I want the operating system that can install in less than 60 seconds.
So from the individual level, I admire those people. I think everybody should strive for that. But also the consequences of that we should mention, like the pursuit of excellence, even when the metrics seem to be frivolous, 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 speed.
It's discovery. It's nice to have the metric that you're 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 Mars. It's like, why? Who cares? Well, actually, the real thing is you'll discover a lot about chemical propulsion. You'll discover a lot about how humans can live for a long time in space. Maybe you'll discover whole new ways of doing solar energy in space or data centers in space or whatever the hell.
All kinds of... Maybe you'll meet—
—aliens, but you never know. But setting a goal and chasing it, especially when it's numerical, is really nice.
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 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, this specific goal was motivated by seeing others do this. I mean, Elon is the singular individual in this conversation. The guy is so out there in his goals. They seem so preposterous at first glance. I've certainly considered them preposterous on many glances. And then you realize that they come true sometimes. Maybe not always on schedule, and so what? So what? So we didn't get the self-driving car on schedule. I gave Elon a lot of shit for that back in '17. But now it's here, and who would have gotten us here except for the person who thought it was possible in '17, who is so impatient with reality that the sheer force is just gonna propel that reality to the place he wants it to be. I think that's deeply inspiring, and we need these kinds of goalposts to be moved. We need someone to run the four-minute mile, so the rest of us realize, "Well, okay, but I guess I can run it quicker." That's what I try to strive for, and I really strive for with Omarchy because it's just more fun to live that way, to have high aspirations and ambitions. And I will say I've actually changed my mind on that a little bit. I spent quite a lot of time speaking the other book that there are a lot of people who don't actually need those 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. But I've come to realize that for entrepreneurs, it's healthy to have suitable chunky goals. The idea of just running the Italian restaurant can't just be that. It should be maybe I just run the one Italian restaurant, but I make the best damn pizza possible. I make the most delicious Parmesan. We have to strive for excellence to muster the motivation to keep going.
And yeah, to dare to think big like you could build a Linux—
—distribution. Is there some stuff you remember about what it took to get us to be so fast? You mentioned a few things that were, the agents were discovering.
One of the things was to treat the lag of human input as an opportunity to preload. So there's five questions you answer when you set up an Omarchy machine. 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 even consider, which is, I mean, the oldest trick in the book. All sorts of video games for a long time have done this, where you let the player interact with something, and then you do preloading and so forth in the background. So it's an age-old technique, but I just hadn't thought about it for an installer. I thought, first you ask the user some questions about their password and their username and their time zone and so forth, and then you do the work, when, in fact, you could preload that. Now, the other things were just a bunch of other tweaks to how the installer itself was doing things in a certain order. Some of them was also just shrinking it. The last version of Omarchy was 7.5 gigabytes, and a lot of the time we're down to now is literally decompressing compressed package files. That's the bulk of what is on the ISO, and that's the bulk of the install time. So by shrinking it down to, I think we're at 5.85 gigabytes now, a pretty chunky reduction, it almost literally translated one-to-one, and I did some really interesting things there. So Omarchy uses the JetBrains font, which is an awesome font, beautiful font, and actually a font I didn't like at first glance, but then it was the only font I could get to look perfect in Mitchell Hashimoto's Ghostty. Ghostty, for whatever reason, did not wanna render my beloved Bitstream Vera Sans just the way I was used to it. But it rendered this JetBrains font perfectly, so I switched over. But the standard package for JetBrains font on Arch is 200 megabytes because it includes all these variations and all these subtypes and whatever. And I realized, like, we're not using any of that. We're using the monospace version in this one Nerd font patched edition. Do you know how big that is? I think it's 16 megabytes. So I simply just came up with a new package, the slim version of the JetBrains package. And there, right there, I saved 180 megabytes. I just went through and did that a bunch of times. We did that with the NVIDIA drivers too. So the NVIDIA drivers, when they came straight off the, I think the Arch repo, were not compressed with the most extreme form of compression, which I think called ZSTD, which is a really slow form of compression. It's slow to create the archive, but if you're building a package, you have all the time in the world if that means you can then shrink the ISO. So on just the two NVIDIA packages, I think we saved 200 megabytes. So what I loved about this was I imagined myself being a car designer at McLaren. So McLaren makes some of the lightest super sports cars in the world right now, and they are absolutely obsessed with shaving every damn last gram off their cars. That's why even a new 750S McLaren is just way lighter than the competition. They use a carbon fiber monocoque, and that helps them over like Ferrari, for example, that uses aluminum. But they also just have this obsession with weight, and I would see this interview with the McLaren folks talking to some journalist just going over, "Yeah, we saved 370 grams off that one." And I just went, "Wait, what? There's someone on a car that weighs maybe 1,040 kilos or 1,380 or something. They're worrying about 370 grams?" What obsessed maniacs. I love them. I want that in my life. I want a McLaren just for that reason. So I thought in that moment as I was shaving individual megabytes off the packages, man, I'm kinda like a McLaren car developer here, even though I mean, we haven't even completed that work. The package could shrink even more. But this is actually this other conflict that we've had with Omarchy and the traditional Linux community is bloat. This idea that it's actually not kosher in certain circles to ship with pre-installed software because who are you to make decisions 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 short for omakase, which literally means chef's choice. I'm the chef. I'm making the choices, and this is the collection of applications I think is awesome," which by the way has a video editor. It ships with OBS to do your recording. It's just in there. OBS is pre-installed. Kdenlive is a video editor, timeline editor, which is great. I edit all my videos with that. And then I made my own clip editor. You talked about the thing with syncing up. I had this thing with clips, so I would do a podcast or whatever. I need a clip for X. And it's kind of a hassle to use Kdenlive or other timeline editors, so I made Omacut. Omacut is included too. It's just for making clips, and you can drive the whole thing with keyboard. So you can go to just the right time mark and then you can hit Control + Space. It moves the clip line to that line. Then you go to the end of the clip, and you hit Alt + Space, and it moves the end line. And then you hit Control + S, and it saves. I can make clips so damn fast now with Omacut. It's amazing. But it has all the software on it. It has Neovim on it. It has Herder 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 out. That takes up some space. It should because this is not supposed to be this just barren landscape where you have to reconstruct everything. This is supposed to be a productive system the minute you unpack it. What is it built in? So it's actually mostly built in Bash for the Linux configuration part. Bash is a great language for that. It is made for that really, for system administration. It also has its limitations. You should not build everything in Bash.
How are agents with generating Bash? Pretty good?
Oh, amazing. Except for one thing, and I have to smack the agents over the back of the head every single time I catch it, and remind it to look at the agent's MD file, because I will have an instruction in there not to make early exits. The agents love to have preconditions instead of fully expanded conditionals. I like something that says, "If this, then do that." Agents like to have precondition or exit, another precondition or exit. And then it falls down to the final thing it actually wants to do. I hate that style. So except for that, it's gotten shockingly, dramatically, awesomely good, to the point where when we talked last time, we talked about how I would ask the agent for a Bash thing, and then I would type it in myself, and I kinda wouldn't remember. And then I took the effort to make sure that all that competence wasn't draining out of my fingers, so I actually learned how to write Bash myself, and now I'm not writing any Bash myself. I have not written any Bash myself for probably a couple months, because the agents have just gotten so good, that if you give them the few pointers on style, and occasionally put pushback on complexity... This is... I tweeted about this a couple of days ago. It blows my mind how often an agent can do something, say it's done, have the work reviewed by another agent, they say it's done, and then I go like, "Eh, looks a little too complicated for me." And then it'll go like, "Oh, yeah, you're right. You're right. I totally overcomplicated this," and it'll cut it in half. I'm like, "Couldn't it just do this, like, from the outset?" And then someone else reminded me, 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 able to go like, "Ah, you're right."
See, that, that just... Those little indications to me show that the programmer is still needed, that intuition, 'cause I think, what, what did you say? I have a sense of the landscape of the code.
When I build something like Omarchy, I do look at the code. I look at the shape. And I look at the proportions. Like, I don't have to chisel everything myself anymore. This is, again, when I'm flattering myself for... Not one moment I'm a McLaren auto engineer, and the next moment I'm a da Vinci, working with his whole studio of students who are doing all the chiseling on the fine marble, right? And he goes like, "Ah, no, the proportions are not quite right." That's how it feels working with 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 when the argument doesn't land. I can tell you when the shape isn't proportional to the problem, or when it just feels like it's too complicated. And shockingly often, it just works. It's like the early days, where the jokes was, "Don't make mistakes." I don't think you have to tell the agents that anymore, because now the harnesses are so good, they run automated testing. They get that feedback. But 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? How much are you—
I type the whole thing. I actually, in Omarchy's built-in Voxtype, audio transcription, and it's actually really good. I just find that I don't think that way, and it surprised myself that I don't like to talk to the computer. I like when I'm in the car, but when I'm in front of the keyboard, I just wanna type. I also just love typing. Are you using the voice?
I'm an introvert who doesn't like social interaction or speaking. And—
I think that's part of mine too.
But, like, for me, for some reason, voice is really powerful. I want to give this... So I run with this thing. It's called Plaud. So 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, in terms of flow, much better experience than behind a computer. And I save, well, usually it's agent-generated problems that I wanna really talk through. And so I write, I speak a long prompt, like, and a lot of it, it's like stream of consciousness reasoning. So I'll sometimes just change my mind about a design, but all of that is in there, and then I have an LLM that processes that. With Plaud, there's obviously an API. You can grab it, transcribe it, clean up the transcript, 'cause you want it to be... 'Cause I'm referring to files. I'm referring to technical concepts, like files and functions and so on, that you wanna make sure the transcription, the speech-to-text gets correctly. So you wanna have a dictionary. You want it to be aware of the code base when you process the files. And then you have this gigantic thing that's often 10, 15, 20 minute me discussing the full design, and I find that to be really, really powerful. Because it doesn't have the problem of overspecification because of how much stream of consciousness thinking the system has about what you're imagining. And this is really good for the early steps of a design. Like, if I wanted to build Omarchy from 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 speaking. I highly recommend people try a voice-based speaking for prolonged periods of time. 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. I try to just type or speak two sentences, and then I get a little frustrated that I could've typed that just as fast maybe. But I have not tried this whole thing of just talking continuously for 20 minutes.
And especially you because you speak very fast.
Yeah. No, I could see it. I think because the reason I've been trying to do it for things like replying to emails, and it got to a point where I thought it could help. And it sort of did, but then it would make a small mistake, and I'd have to correct the mistake. But if you're doing stream of conscious to an LLM who doesn't care about all this other stuff anyway, that probably works way better.
There's two things you have to do. One, okay, you have the raw audio, transcribes it. I use ElevenLabs for transcription. They're probably the best speech-to-text model. There's still mistakes there, so you need to have a prompt that has access to a dictionary and to a lot of—
How do you build the dictionary?
First of all, there's a dictionary of terms that are relevant to you. Then have it aware of... So you have an agent that looks through, when you're talking about code, looks through your entire code base for key terms that you use often. Also, when we're talking about messaging, Gmail, and WhatsApp, and so on, it's aware of all the people and common terms that you get wrong.
You've really gone deep in this.
That, to me, is a problem that needs to be solved because one of the issues you probably have speech to text is that it makes mistakes sometimes.
Correct. Then I can't use it for dictating an email.
Right. If it never makes mistakes, you—
Then it would be more appealing.
Yeah, and you are okay being not perfectly real time, actually. You're okay waiting five seconds potentially. I don't know what your use case is. I'm okay waiting five seconds, 10 seconds for it to process everything.
You gotta vibe code this into an app that does all of this such that it just works.
I think I mentioned to you offline that it's very surprising to me that translation is not done perfectly. Speech-to-speech translation.
It seems like you've gotten much closer to the best of what's possible by putting all these things together.
But, you know, when you're shipping that kind of stuff, you do have to start thinking about latency.
Don't let the perfect be the enemy of good. I mean, if you're suddenly enabling something people wouldn't be doing before at all, it's okay to take five seconds longer. I think this is one of the areas that Steve Jobs was so good, that he realized, like, the first iPhone was terrible in all sorts of ways, right? Absolutely horrendously slow data connection, very underpowered, all the things. But the product was so compelling that it just didn't matter. You'd be shipping too late if you waited until everything was just right, just perfect. If you're solving a problem that people currently don't have a solution for, they're 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 don't need to solve it.
Mm-hmm. Well, I do think there's a few... So, like, Whisper Flow is really good. A few people have stepped up that speech-to-text thing.
Yeah. Yeah, we're using Voxtype. But others have—
Uh, Voxtype is open source?
It is. It's just using one of the open models. I forget which parrot model it's using or something like that. It works quite well for short commands, so I have used it for that. I think you just hold down F9 and it starts a dictation.
Oh, so it's already in there?
You just have to get online. It'll offer you to install the Voxtype package.
Got it. Yeah, yeah, yeah.
Because the Voxtype package includes a model that's 150 megabytes, and I was like, "Ah." "I don't know if I can carry that." So we don't have that in there by default, but we have the option set up. And I actually just yesterday was thinking we really need to combine all of these things. So Omarchy Quattro is truly amazing as a malleable operating system where you can add functionality to it just by talking to your agent. But that's exactly what we should be doing. There's a lot of people who would find it very natural just to talk to their computer and say, "Hey, can you make me a stock panel? I wanna track Apple and Dell," and just see the agent go off through, entirely through voice, both in terms of the input and in terms of the output, and then your operating system just changes. This was the vision of Iron Man and Jarvis and bespoke applications that just appear 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 AI's inventing literally the next frame, but you can actually play the game. And I think if they're able to do that, why can't we do that with our operating system? Why can't we just talk to it and ask it to be a different way or change, and it'll all just change? I think this, by the way, is one of the other breakthroughs that is going to lead to the total domination of Linux. The agents have taken all the hardship out of diagnosing Linux systems and turned the fact that Linux produces these overly specific, totally arcane error messages into its greatest advantage. When Linux has an issue, the agent can take that very specific error message that makes no sense to a normal human and correlate it with the fact that the agent was pre-trained on 40 million lines of Linux code, so it knows exactly where to look and dial it down.
I have not had a single problem on my Linux machine since the beginning of this year that an agent could not diagnose. That was not true a year and a half ago. A year and a half ago, I was still searching on forums to find answers to esoteric questions. Now the agents know the source code of not just the Linux operating system, but every single piece of software I have on that box. The agent has access to the source code of all of it. In fact, this was one thing I built in just before we shipped. So once you set up your default agent, Omarchy Quattro has a crash watcher. If any app on your machine crashes, it'll pop up a little thing, ask you whether you want your AI to diagnose the problem.
Unbelievable. I've seen things where there's an error in some subsystem. Some, some... Not even Linux, just some app I installed. The agents start digging through the logs and looks up systemd log. Then it checks out the damn source code of the application that crashed, pins down that it's in this Rust file line 472 that there's an unbounded, unwrapped variable that overflowed or whatever it is. Then offers you whether you wanna file a bug report with all this detail. I'll give you one amazing anecdote. So I was working with this guy JDX, who's working on mise. This, this package manager for fast-moving development tools. This is how we manage all the agent software and so forth. Normal package managers are not built to be updated seven times a day. And all of these agent harnesses are updated about seven times a day, so we needed an out-of-band package manager, and mise just turns out to be perfect for this. Anyway, it had a small bug. There was some race condition which agents are very good at finding because when you start running multiple agents at the same time, they will suss out all these race conditions you have in your underlying infrastructure that was never triggered by a human trying to do manual thing. So it finds this issue, right? I tell the agent, "Hey, can you post this as a bug report to JDX's GitHub?" 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 that it needed to file. Well, it went to GitHub and tried to file all 28 issues at the same time, which it did in about, I don't know, 12 seconds. GitHub, not unreasonably, marked that as probable spam and banned my Omarchy bot. 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 what? Just email JDX. Here's his email address." I had already set it up with hey.com an email address. We have a CLI that's in beta right now. So I set it up with that so it can send to email. That's how it gives me reports about outstanding issues and PRs. So it sends JDX this email about the bug it found, and he was like, "Wait, this is the first time I've gotten a bug report in unreleased software." Because it had downloaded the source code to mise, saw that he had already fixed the bug, but that it would not fix the problem entirely in software they had not shipped yet. Pinpointed the problem, and he was like, "Damn it, I got a bug report before we even cut a release." Unbelievable AGI levels of mind-blowing stuff.
That's amazing. That's amazing.
And that was actually, if you remember, that was an open question whether AI systems could debug well. That was kinda the thought that, yeah, sure, they can write code, but can they find issues and—
They are so incredibly good at this, and this is actually interesting because I was on the perhaps same side of that, a little skeptical about can it reason about all these things. And Mikhail, who's the CTO at Shopify, ran a scientific study on this. They had... And this was, I think, late last year or early this year, where he had agents go back through all the incidents, both outages and other problems that Shopify had had in production, trace that back to the PR that was merged, and find out whether PRs that had been reviewed by human or PRs that had been reviewed by agents were of higher quality. Well, lo and surprise, the PRs that have been reviewed by agents caused far fewer issues in production. And this was with models we had six months ago. At this point, 100% in the majority of domains we work in today, agents are better at finding bugs.
So you now run agents all day. What stands out to you as the better model? Who's currently winning? It seems to be changing constantly. Sol, Fable, Grok 4.6, Gemini, the Chinese OpenWay models.
What's amazing to me is that you could rattle off so many different contenders. That this market is so wide open, that it does actually change back and forth, that we have real competition, that there are so many labs that are able to get either to the frontier or close to it. That, by the way, is remarkable. I still don't fully understand that. But to answer your question, the best model in general right now is Fable. The second-best model, in my opinion, is Opus 5. But the tier just below Opus 5, and it's not even that they're always below, sometimes they're ahead. I'll get to that in a second. I would rank GPT Sol very good. Grok 4.6 I just started testing a few days ago. I had this wonderful test that I've set up by accident where I translated this Python library into Rust. The screensaver you just saw with all the cool animation? That's powered by a Python library called Terminal Text Effects. Really cool library. We've been using it since the first day of Omarchy. The problem with that is it's written in Python, so when it starts up, especially on a 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 of energy, and it spins up your fans, and it drains your battery. Doesn't really matter on a local computer, but it does matter on a laptop. So I thought, "Do you know what? This sounds like a problem for Rust." So first I gave Fable the challenge, 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. "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. I want it to be pixel perfect, frame by frame, do a full analysis, don't stop until you're finished.
I kid you not, in just under 45 minutes, it was like, "Am I heard all? I'm finished. I've checked everything. I have reduced the startup time from 86 milliseconds to two 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?"
—and I don't know why I'm surprised, because this translation job is something we've known for a while that AI is pretty good at, but it was still staggering to me that I could one-shot a full translation of a Python library I had been using for a year, that others had been using for much longer, and turn it into a Rust executable without knowing any Rust, 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." It packaged it up as a new package. It told me, "What do you want it to call?" "Uh, let's call it TTFX. Let's create a new Git repo." It sets up the Git repo. "Let's create a new package, build package for our build system." It puts that 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 TTFX." It does all of it, and I'm just sitting there. And again, I'm already at this point fully delirious with agent acceleration, and I still had to lean back and I go like, "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."
Now, as I understand, you also repeated the same experiment not just with Fable.
I did it on all the models. First, I did it... Actually, funny thing. 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 the job. Part of the reason why I think it was able to do that was that the first thing 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, and here, how you analyze it, and here you run the effects," and so on. I didn't review the plan at all. I didn't change the plan. It just made the plan so Opus could take over. And then I thought, "Well, if Opus can finish the job, maybe some of the other agents could finish it, too." So the first thing I did was 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 and a half. But here's the kicker. The per token cost, I didn't pay per token. I have a max subscription to Claude, right? So it did it within the subscription. It used all my tokens, so I had to switch over to Opus 5. But if I had paid per token, it would've been 550 bucks, I think, to do the whole thing. And I thought for a second, "Holy shit. What a steal." If I personally had to learn Rust well enough to be able to do this 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 10x execution speed up. This is ama- I would totally pay 500 bucks for this. 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, gave it to Sol. Sol, in an hour and a half, and I think $46 worth of token, repeated the task. Did the same thing. And I thought, "Well, blimey, that's amazing." Then I got greedy. So I asked GPT Luna, which is this crazy cheap model that OpenAI has as well, "Can you do it?" Absolutely not. First of all, it didn't even wanna start the task. I think something happened I don't know. I think 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 told it not to stop and so forth. So Sol could do that. Fable could do that. But Luna couldn't. So I had... I think I did 12 prompts. Kept telling it to do it, and eventually I sorta got it started. The first thing it did was to cheat. So the first thing it looked outside its own directory, saw that there was already another implementation, and just did a short wrapper around that and said, "I'm done." Hilarious. But it couldn't finish because it made just a couple thing. But I mean, 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, it's cool that there's others trying, but, like, I'm not gonna use it," because it 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. Absolutely unbelievable. Then I repeated, too, with Kimi K3, which took forever. I forget how long Kimi actually spent on 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, 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 fastest. It was the one that wrote the plan, but 550 bucks, and the output the same. The others, Sol, Grok, about the same 1/10 the cost. DeepSeek, 1/20 the cost, but you have to wait a little longer. Absolutely gobsmackingly incredible. And now, by the way, by the way, so I had Fable finish the first job, and then Opus finished it. That was the first one shot, right?
It's 10 times faster. I did two auto research runs, which isn't even auto research anymore. You don't have to do the slash. You just tell it to keep going until you tell it to stop. It ended up... I think we ended up with a 46 time execution improvement over the original.
Um, I do, you know, I just don't think there's anything that compares to Fable in terms of planning. So I usually do Fable for planning and for reviewing, 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 not run out of tokens too much.
What I found, it's really interesting because Fable is, in my opinion, the best model right now, but it also makes mistakes. And the best way to get the best software, I would actually rather have two differently sourced sort of... I mean, they're not mid-tier, they're all frontier. But have, let's say Opus 5 and Codex... and have one check the other's job. This is my standard operating procedure now. I'll have Opus or Fable do the work, and then I always end it, review 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. And then that's my workflow when I'm having my agents on my own machine do it, and then I push to GitHub. And then Copilot, I kid you not, has actually gotten good. Copilot keeps finding stuff that's legitimately broken, which is also incredible acceleration because the first version of Copilot that started doing this was literally retarded. It would just constantly flag things that were nonsense. It would constantly flag the same problem over and over again as you would push. It was really annoying, so I think a lot of people actually ended up turning that off. And if they did, they should turn it back 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 surprised? Even if you're a good programmer, if you finish a job and you ask your also very good peer to review it, you're gonna end up with better code. Of course you're gonna end up with better code. So build that into your process. Pick one of the agents to drive with. I've mainly been driving with Claude, which is actually interesting because I have some other reservations about Anthropic. But the reason I'm sticking with Claude is, in my opinion, they actually have the best harness.
And one of the reasons it's the best harness is it's multi-agent running. So if you wanna run multiple agents at the same 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. So if you wanna do this thing where you have multiple threads going on... the Claude Code is just the nicest setup. They also just, they keep being a little further ahead, which probably shouldn't be surprising. I mean, Boris is one of the guys that's working on that, he was the first one and basically came up with the thing. It's just interesting that that has been an enduring advantage. I've also used OpenCode a lot.
I think OpenCode is great. I use OpenCode mainly as my main harness for all the open models, so Kimi K3 and so forth. And I inference not on the Chinese servers. So I inference on Fireworks... which is a really nice service where you can just pay by the token, and if you're using these open weight models, they're not that expensive by token. So that's a great way to do that. But I don't think I would be using... Well, maybe I would. But if... Using Claude with a subscription does feel like a bargain, like a crazy bargain.
Yeah, I knew. And I actually have four.
I just signed up for my second one this morning... because I got up at 4:00 with jet lag, and I started working with the agents right away, and now I'm three days away from limits resetting on Fable, and I ran out of tokens.
So and they should really— they—
I mean, I don't... Why is that so complicated? Can't you just stack one subscription? Why do I need to log in multiple times?
And obviously you'll have to develop some kind of tooling...
... to move from one... to the other. And then—
Oh, we're building that into Omarchy, by the way. So the next version of Omarchy is gonna ship with multi-sub support. I wish that the labs would just make you buy a max 100 times.
It's kind of actually fascinating though for me because there's... It's very challenging for me to use Grok 46 and then a Claude Code because Grok is so fast.
Yes. That's actually the deal. The deal with Grok is that their fast 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 it almost gets you to this single thread mode... where the agent is able to keep up with you.
Yeah, but I don't know how... I can't... Once I get used to Claude Code, uh, I can't handle the speed of Grok 'cause I'm used to now multitasking. And so now I'm back in this... anxious mode of, uh...
... switching. So I mean, all of this is a learning a learning process of what programming actually is supposed to look like.
By the way, one of the reasons why I'm not sure what the final form of these harnesses is gonna take, because one of the things we've started experimenting with at Base Camp is that we put the agents inside of Base Camp, and we treat them as coworkers.
And then we assign them tasks...
... inside of Base Camp, so they will do work on a to-do in Base Camp, or they'll work, do work on a card. And it turns out that a collaboration tool that's optimized for asynchronous communication is actually the right format versus these harnesses are a little more like chat. And chat, which was, I mean, the original format for these things, is not the right thing because you're sitting around waiting. It entices you to sit around and wait, versus when you're dealing with something like Base Camp and you can just give an agent a to-do item, you don't expect to hear back immediately.
So I don't know what the final form is gonna take, but I've been using all of the harnesses. So I use them inside of Herdr, and Herdr has panes. So oftentimes I'll run Claude up top, and then I'll run Codex down below, and then maybe I also have an OpenCode set up. But I will say just very recently, I found that the human in the loop is the limit here. So I've started setting up more automated systems. I've been building an Omarchy bot that can do more autonomous development on Omarchy that is just On a regular schedule process, all depending to do or PRs and 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 go or I think you should close. And then I just make the final determination there.
Do you find the multitasks, the task- switching mentally exhausting?
Yes, but in an exhilarating way. Like, one of the things I always loved about race cars was when I would stumble out of the car absolutely smashed and barely able to hold my head up, and I'd lay down on the garage floor and just think, "Holy fuck, I'm alive." That's the kind of exhaustion, mental exhaustion I'm feeling at moments with the agents right now.
By the way, when you're racing, where, what's the source of the exhaustion? Like—
It's physical. It's just... I mean, I bet it's the same thing with jujitsu, right? Like, it's actually satisfying to feel exhausted. It feels like you've applied yourself. And I feel that with the mental exhaustion you can get from agents.
Are you also mentally drained from racing?
Yes. If you drive in difficult conditions, especially if you drive in the rain, where you're constantly managing the thing right at the knife's edge. But most of the time, I just enjoyed the physical exhaustion.
And here, with the multitask switching, I do find that, like, I get really exhausted, and I think it's because I'm thinking a lot. 'Cause, like, you're basically—
... there's no, like passive low energy thinking.
Solve problem, next problem, solve problem, and you're really thinking.
It's funny. It's actually very similar with racing. There's some tracks, like Le Mans, where you get time to relax. There's the Mulsanne straights where it's kinda long. You're going fast, but in a straight line, you can take a breath. And then there are other tracks where it's just coming full on all the time, and you don't have any moment to relax, and you stumble out of the car after an hour in each, and it's very different how exhausted you are. And this is exactly what I'm finding with the agents too, that when you're running at max human capacity... you're just constantly in a corner.
Do you think people are in danger of burnout?
I think the moment we're in right now is gonna pass. That this need to... It's not a babysitting motion, but constant interaction with the agents is gonna fade. From what I've seen on Omarchy, I think we can automate a ton of the development and debugging of the system... to the point where I can just review that email once a day, and then make the decisions once a day. Yep, goes, no, goes, in, out, and whatever. We're not there yet, but I think we're gonna get there.
Well, can you sort of, like, empathize with the feeling that, say, a young person sitting in San Francisco, everybody around them is not sleeping. They're obsessed. They're programming nonstop. And there's this kinda sense, that AGI is gonna be here any minute. There's, everybody knows somebody who's made millions of dollars because they sold a startup. And so there's a kinda feeling like you shouldn't be sleeping. You should be constantly managing agents and programming and building and, 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.
It's always exhausting. There's no other way around that. And by the way, hasn't San Francisco always been like this? I remember the dotcom boom years, and everyone was just the same, and then it was all mobile, and before that it was the gold rush. So I think that's probably... just San Francisco. It attracts... the kind of individuals...
... who would, who would think, "I don't have time to sleep." I've not been on that track, generally speaking, but I will admit that the last three months have felt more exhausting than any other project I've done in the last five years, since the Hey launch. That was the last time we had a crazy exhausting launch.
Do you think you can keep going at this pace?
No, no, no, no, no. This is not sustainable at all.
But I also, I can see the end of the tunnel. Because the automation, which by the way, everyone is building. This is so hilarious. Like, everyone is building their little gas town, their little agent coordination, their setup. This is all gonna be solved. We're not all gonna have to build our own coordination harnesses. Of course we're not. And in fact, I'm a little surprised that it's gone this long, that there's not more of it has been sucked up by the major labs. You'd think that they'd just built this stuff in. But some of it is also this is what a new domain looks like. I remember for a hot moment when 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. 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 end of the tunnel. So I don't mind a sprint. In fact, I welcome it now. I like this notion that most time is calm, but then occasionally you gotta climb a mountain.
If we look back to the DHH a year ago, what advice would you give to that guy?
I wouldn't wanna spoil it.
What a... thriller that this has been. I mean, if you would've scripted this, I'd be like, "This is so far-fetched. Get out of here. Unbelievable," right? I wouldn't wanna know anything. The fact that this rollercoaster and this acceleration has been absorbed in real time by everyone, no one knew, right? Like, some had premonitions that were a little better than others, but no one knew, not exactly the way... 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... absolutely applies...
Uh, that said, what do you think DHH from a year from now will be like? I mean, what are the possible versions?
That all our dreams are coming true. Like, this is the AI maximalist abundance argument. That once Elon's robots have the level of intelligence and AGI that I'm experiencing on some of this development stuff, the world is gonna be so unrecognizably different that we can't even imagine it. Now, I don't spend any time thinking about that because I think that is the way to the AI psychosis. So therefore, I just immerse myself in the moment and get absolutely the maximum out of it and...
Just have fun. I mean, I get it why people are finding this exhausting. Just keeping up. Not even using agents, just, like, being like, "What's the latest model? What did, what tool are we using now? Is it TMox? Is it a Herder? Is it none of it? Is it agents?" Whatever. I get it, but I also again think we should be so blessed. You are alive in this moment where decades are happening... in weeks.
Um, quick bathroom break. I've been generating a lot of video recently with Higgsfield, and so they, they became a sponsor. Created a racing video. Wanted to get your opinion on it.
Oh, yeah? Oh, let me see. Let me see.
If they downshift on a straight like they do in the movies, I'll call it out. That is the worst. Let me see.
... this is fully AI generated. So first it goes—
Shit, that's my suit. That's my car.
Lights go green on a street that never sleeps.
