← All transcripts

Why AI Has DHH Working 16-Hour Days Transcript, AI Summary & Key Points

a16z · 17 hours ago · Science & Technology · 52:16 · EN

Watch on YouTube

Answer

DHH works 16-hour days because agents let him put more of his taste and perfectionism into software — the barrier between spotting a flaw and fixing it has collapsed, and his dopamine payoff comes from the last 2% of polish that agents finally make affordable.

AI Summary

David Heinemeier Hansson argues that AI agents make Linux, not macOS or Windows, the natural operating system for the agent era — the terminal, config files and verbose error messages that once made Linux hostile are exactly what agents need. His agent-first distribution Omarchy is being built almost entirely by agents (he has written no code by hand for five to six months), aims to make the whole OS mutable by prompt, and targets being bigger than macOS within roughly three years. He contends that software development has become cheap enough to break platform lock-in, that taste and the willingness to polish the final 2% matter more than ever, and that open weights are essential to keep frontier labs from forming a duopoly. The conversation also covers the return of pseudonymous hacker and 90s demo culture, and DHH's claim that he is now working 16-hour days because the perfectionist's workflow has been supercharged.

Key Points

  • Every legacy weakness of Linux — a terminal-heavy interface, config files, inscrutable error messages — is one of the three things agents want most, so Linux is now the better OS for the agent era.
  • Omarchy's entire operating system has been developed by agents for the past five to six months; DHH has written zero lines of code by hand but hundreds of thousands through agents.
  • Users can modify any element of the OS — start menu, clock, calendar, new features — just by prompting, with changes landing in seconds to a few minutes.
  • macOS and Windows are unfit for the age of agents; macOS hotkey remapping can only be done by clicking through a settings app, and Apple is adding more popup screens warning against agent file access.
  • DHH says Apple risks becoming the AOL of the agent era: a walled garden with watered-down agents instead of the open internet.
  • The window for a third operating system has been open for the first time in 25 years — the last premiere that mattered was OS X in 2001.
  • An Omarchy community member replicated all of Adobe Lightroom in about two weeks, better than a product Adobe spent a decade and a half and tens of millions of dollars building.
  • Application lock-in collapses because software development cost has dropped to zero: anyone who can steer agents can replicate top apps, so 'first we take Manhattan desktop, then we take Berlin mobile.'

AI in practice

Used for

Agents

  • Acting as the development team for the Omarchy Linux distribution 2 held 05:16
  • Grokbot — Autonomously procuring parts to build itself a physical presence in Martin Casado's house 2 held 23:04

Tools & resources

11 items

3No. 8308
AIAINotes.us Tool

37signals

37signals.com/

37signals is a software company led by David Heinemeier Hansson. It is the maker of Basecamp, Ruby on Rails, and Hey. Its official site presents the company through a catalog of the ideas, or “signals,” that drive it.

Mentioned in
1 video
Kind
Other
BNo. 8305
AIAINotes.us Tool

Basecamp

basecamp.com

Basecamp is a project-management and team-communication SaaS developed by 37signals. It brings tasks, discussions, decisions, files, deadlines, client work, and company-wide communication into shared project spaces rather than scattering them across separate apps, email, and chats. The service supports project planning across activities such as product launches, website redesigns, client onboarding, customer support, and internal operations. Basecamp's website describes Basecamp 5, released in 2026, as its current version and offers a free trial and paid service for teams, companies, and nonprofits.

Mentioned in
1 video
Kind
Other
CNo. 8302
AIAINotes.us AI product

Claude Code

claude.com

Claude Code is Anthropic's AI coding agent, available through Claude subscription plans. It was described in the videos as a frontier coding agent used to write Omarchy's code, with David Heinemeier Hansson running multiple subscriptions. Anthropic's Claude site lists Claude Code as a feature of its paid Pro plan.

Mentioned in
1 video
Kind
AI
DNo. 2126
AIAINotes.us AI product

DeepSeek

deepseek.com

DeepSeek (深度求索) is an AI research project and model provider that develops and open-sources large foundation models (notably named DeepSeek-V4 and DeepSeek-R1) and provides hosted inference via a chat interface (“DeepSeek AI”) and an API. It publishes open-source model releases and offers a hosted API for text-based prompt inference; video reports state DeepSeek models can be exposed as external backends and routed into the Codex model picker via Codex Router.

Mentioned in
7 videos
Kind
AI
FNo. 0417
AIAINotes.us AI product

Fireworks

fireworks.ai

Fireworks is an AI platform that provides training and inference tools to transform open models into domain-specific or specialized intelligence. The service offers capabilities for model training, fine-tuning and serving to support customized AI deployments.

Mentioned in
6 videos
Kind
AI
FNo. 5919
AIAINotes.us AI product

Fireworks AI

fireworks.ai

Fireworks AI is a training and inference platform for transforming open models into specialized, deployable AI systems. Its training offerings range from guided runs, where the platform helps define the task and plan, through configuration-led managed training, to user-written training logic with custom losses, trainers, and reinforcement-learning loops; trained checkpoints can be deployed to production. Its inference service supports open models and customer-trained versions through serverless, dedicated on-demand, and reserved deployments, with an optimized inference engine and OpenAI- and Anthropic-compatible interfaces. Fireworks also provides a model library for running open-source models and supports the continuous train-and-serve workflow described by the videos, including monitoring, hot-loading new weights without downtime, and matching training and serving numerics.

Mentioned in
2 videos
Kind
AI
HNo. 4609
AIAINotes.us Tool

hey

Open source · rakyll/hey

hey is a command-line HTTP load generator and ApacheBench replacement. It sends a specified number of requests, or requests for a specified duration, using configurable concurrent workers and optional per-worker rate limits, then prints request statistics or CSV metrics. It supports common HTTP methods, custom headers and request bodies, authentication, proxies, timeouts, redirects, compression and keep-alive controls, and HTTP/2 endpoints. The project provides binaries for Linux, macOS and Windows and can be installed on macOS with Homebrew.

Mentioned in
1 video
Kind
Other
ONo. 4247
AIAINotes.us Tool

Omarchy

Open source · omacom/omarchy

Omarchy is an opinionated, preconfigured, terminal-oriented and agentic Arch Linux-based desktop distribution and ready-to-use Linux setup by DHH, developed by Omacom. It provides a tiling desktop workflow with navigation, hotkeys, themes, clipboard history, reminders, text extraction and dictation, screenshots and recording, an Omarchy command-line interface, terminal and Neovim tooling, AI and development tools, shell utilities, browsers, web apps, gaming support, networking, dotfiles, updates, and system snapshots. It includes applications such as Chromium, Obsidian, LibreOffice, Kdenlive, and OBS Studio. Its extensive manual covers configuration, hardware, troubleshooting, and installation, including dual-boot and unattended installation options, and is maintained in the repository's `manual/` directory and mirrored to learn.omacom.io. Omarchy is released under the MIT License.

Mentioned in
4 videos
Kind
Other
QNo. 8303
AIAINotes.us Tool

Quickshell

In the AINotes directory

Quickshell is a malleable shell used for desktop interactions in Omarchy. It was created by the pseudonymous developer outf-foxed.

Mentioned in
1 video
Kind
Other
RNo. 7482
AIAINotes.us Tool

Ruby on Rails

Open source · rails/rails

Ruby on Rails is an open-source, full-stack Ruby web application framework for building database-backed applications using the Model–View–Controller pattern. Its Active Record layer represents domain data and business logic, maps objects to database tables, and supports associations, callbacks, encryption, and query scopes. Action Pack, including Action Controllers, Action Dispatch, and Action Views, routes requests, processes parameters, renders HTML or JSON through Ruby and HTML templates, and provides caching support. The framework also includes Action Mailer, Action Mailbox, Active Job, Action Cable, Active Storage, Action Text, Active Model, and Active Support, several of which can be used independently. Rails provides integrated tools for email, WebSockets, background jobs, cloud file storage, security protections, and identifies Hotwire as its default front-end framework. Applications can be installed with RubyGems and created with the `rails new` command. Ruby on Rails is released under the MIT License.

Mentioned in
2 videos
Kind
Other
XNo. 8306
AIAINotes.us Tool

X (Twitter)

In the AINotes directory

X, formerly Twitter, is a social platform referenced in the video as a place to follow people and organizations, including David Heinemeier Hansson, Martin Casado, Anish Acharya, a16z, and the video's host.

Mentioned in
1 video
Kind
Other

Links mentioned

🔒 Full analysis locked

Unlock more videos and the full analysis

A credit unlocks one video's full analysis for good — the build steps, the tools and how each was used, the methods behind every use case. Pro opens the whole library instead, and raises how many videos you can analyse a day.

Unlock full analysis — free

Transcript

Searchable transcript of Why AI Has DHH Working 16-Hour Days — a16z (52:16). Search for a phrase, then click its timestamp to jump straight to that moment in the video.

Captions sourced from the original video on YouTube, published by a16z. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.

00:00 Mac OS and Windows are just unfit for the age of agents. Every single thing that made Linux hard to use in the past. Oh, it's very terminalheavy. It's got a bunch of config files that's hard to figure out and it's got inscrutable error messages. These are the three things agents love. It's really remarkable what you can do. Just tell your agent, modify the OS on which I'm operating right now, and it does it and it works.

00:22 Anyone who still thinks that there is moat in source code are delusional. It's entirely possible that by next year we will have true super intelligence. Everything is just going to mutate live and that's 1500 tokens a second. We can do that today. What's going to happen two years from now when these models are going to run at a 100,000 tokens a second, a million tokens a second.

00:42 How do you think about the Frontier Labs versus open source? Um well David, thank you so much for joining us. So we're going to talk about um all things uh company building AI software. Um maybe listen, everybody knows you. I know you, you know, and are huge fans, but would you mind just like the rough sketch of your background? I mean, you know, Ruby on Rails and Base Camp and Hey, and then now like your current projects.

01:08 >> Sure. So, as you said, I'm David Hansen. I have been building things for the internet software for some good 30 years or so. About 24 of those. Um, no, that's not quite right. 23 of those base camp >> the original project management tools SAS before it was called SAS >> released the day before or the day after Facebook went live so that's about how old >> that's awesome >> that became the um sort of founding application for Ruby and Rails Ruby and Rails was extracted out of that and Ruby and Rails was then used to

01:46 build Twitter it's been used to build GitHub Shopify Airbnb, Coinbase, and a million other applications out there. And I've continued to work on both of those ever since. We still sell Basec Camp. That's version five now. I've worked on that for a long time. Ruby on Rails is up to version eight. And 5 years ago released hey.com, our email service. And then in those 25 years or so, wrote uh four books including Rework and Remote.

02:16 it doesn't have to be crazy at work. A bunch of titles and um kind of just was uh developing my voice around a bunch of these topics, both how to run a company, what I think about technology, and I put a lot of those thoughts onto X every day now. But uh yeah, that's the quick summary. Oh, and then what am I doing now? I'm doing a Linux distribution called Omachi, the first native OS for the Age of Agents.

02:49 >> I love it. >> And we're having a lot of fun with that. Getting hundreds of thousands of people to switch from Windows and Mac to a Linux system to finally make the prophecy come true after 25 years that this is the year of Linux on the desktop. >> When was your first Linux distro? >> So, it was Ubuntu and it was only two and a half years ago. If we're talking about the desktop now, I've been using Linux on the server since the 90s.

03:12 Yeah. Yeah. >> But I did not run uh Linux on the desktop. I was a Mac head since 2001. I switched to beta 4, I believe it was, of OS 10 and then just stayed in that ecosystem right up until Apple drove off the rails and got addicted to easy service revenue where you can just take 30% of other people's business. That is a great business to be in. You don't even have to innovate.

03:40 You can waste 10 years developing a car you never sell to anyone. You can launch a huge flop of a ski mask. No one cares. If you can just take 30% of other people's business, you're golden. Well, what was not golden about that was their relationship to developers. And I was certainly one of those developers who had been advocating for Apple for decades.

03:58 Suddenly, finally, I had enough and I switched to Linux. is that your big why now for Linux desktop happening? I mean, how much of it is Apple and Windows and sort of others, you know, muddying the waters of their ecosystems versus AI and new technology that lets you modify the OS? I was playing with it last night and it's really remarkable what you do what you can do.

04:20 You just tell your agent, modify the OS on which I'm operating right now and it does it and it works. It's it's amazing. >> That's what's making Linux happen in the big way. I mean, all of these other motivating factors like I had about Apple not being so nice to the developers is something a few developers care about. Not even all of them. Plenty of them just continued to suffer the abuse because that's what you got to do.

04:44 If you sell commercial software, you have to be on the app stores. So, I don't think that that was ever going to motivate droves of people to switch to Linux. But the age of agents absolutely positively will. And there was just another drop today that Apple is making it even harder to use Mac OS with agents. Even more scary popup screens about how you shouldn't let Claude have access to your files.

05:06 Well, what is it going to do then if it doesn't have access to your file? So, um is just going in a very different direction. We are leaning all in. The entire operating system is being developed by agents for the past five months or so now, maybe even six months. I've not written a single line of code for Amachi by hand, but I've written hundreds of thousands of lines of code through agents and we're using that to accelerate the development of Amachi itself and we're creating things at a rate I scarcely believe would

05:38 be possible and then we're putting all that power into the hands of users and I think exactly as you had anyone who gets the experience where the entire operating system is malible that they can change anything they can change how any element of it work. They want a new start menu. They want to move the clock around. They want a different calendar. They just want to add a feature to an existing part of the operating system.

05:59 And they can just do that by prompting. And it literally happens in sometimes seconds, maybe at most a few minutes is deeply intoxicating. And I don't think any of those people, if they really soak in it for five minutes, are going back. So, I'm extremely bullish on the idea that both Mac OS and Windows are just unfit for the age of agents. And I think the age of agents is going to be the only thing that matters when it comes to these computers in about 5 minutes.

06:26 So, we're really just leaning hard into the idea that operating systems we've had for 40 years built around mouse and click dynamics, they just don't work for the age of agents. I mean, let me give you just one basic example. If you want to remap the hotkeys to go to a different workspace in Mac OS, there's only one way to do that. Manually go into a settings app, click one field at a time, put in the new hotkey, then click the next field, then put in that hotkey.

06:56 You cannot do that aentically. Well, I guess you can have computer use to it, but that's pretty dumb. That's pretty silly. And what we have on the Linux side is that every single thing that made Linux hard to use in the past, oh, it's very terminalheavy. It's got a bunch of config files that's hard to figure out and it's got inscrutable error messages.

07:15 These are the three things agents love. They want a command line interface for everything. They want uh predictable config files that they can just edit and update. They want all the details in the error messages. So all these historic deficits for Linux have literally this year turned into the reason why you want Linux on your computer. If you're into agents, if you're into AI, and at this point, who that's making anything is not into agents, is not into AI.

07:44 So, I think uh Apple is running the risk right now of becoming the AOL of the agent era. We're going to provide this Walt Garden. We have Ginmoji over here. You do not need to use an actual agent from an actual Frontier Lab. Just give us a chance to give you something watered down and protected. And most people don't want AOL. They want the open internet and this is what we're trying to give them with Amachi.

08:10 >> What what uh what happens to mobile in this world? >> Mobile has to change the same way and this is actually one of the reasons why I am so excited about this age of agents first on the desktop because the window is now open for a third operating system for the first time in 25 years. The last time we had an opening was that OX10 in 2001. That was the last time anyone premiered a new operating system that made any dent in the global market share.

08:39 >> Since then, nothing. Window closed. This year, window open. Suddenly, Linux has a chance again in a way it literally hasn't happened for 25 years. Mobile has been suffering under the same setup. We've had a duopoly and Google and Apple have just divided that juicy 30% revenue from everyone else's business between them. And what were you going to do?

09:02 Microsoft tried. They spent, I don't know, several billion dollars to make Windows OS a thing on mobile and it wasn't a thing in huge part because they did not have the applications. Who wanted to switch to a Windows phone if they couldn't get Instagram? So, that died on the vine. That's what's changing right now. >> We're seeing it already on Linux.

09:21 someone in the uh Omachi community rewrote the only application I still kept a Mac around to run Adobe Lightroom because it had all my plugins and it had all my presets for my cameras and I just wasn't willing to give that up. So, I had like a Mac just for that, right? >> One guy over the course of like two weeks replicated all of Lightroom. not just replicated but made it better than Adobe had literally spent I don't know a decade and a half and tens of millions of dollars if not more developing.

09:53 Now apply that to a phone and operating system running on a mobile device that suddenly can get every app you'd ever want because if it does not exist someone is going to come up with it because they can just vibe code it in 5 minutes. That completely changes the game. The entire lock in on these ecosystems is about application availability. And in the age of agents, software development has just dropped to zero in terms of pricing, in terms of difficulty.

10:23 Anyone, not even software developers, anyone who can steer these agents can replicate something as good as an Adobe, as good as anyone. That's a completely reset of the playing field. And that's why this can happen. So my motto right now is first we take Manhattan desktop then we take Berlin mobile. >> Well it is fascinating because I think like the concept of building a third u mobile ecosystem 5 years ago was it was just I mean it's certain to fail completely daunting and now for some token budget we can replicate

10:57 the top 10 100,000 10,000 apps and for many apps they may not even need to be replicated because the agents can just natively do the work that you're signing the app up to do. >> Correct. It's absolutely on the table. >> I think that's really interesting too because one thing is to replicate say Adobe Adobe Lightroom or Adobe Photoshop. That's always been the historical example, right?

11:16 Linux was never going to succeed because it did not have like three apps. It did not have Microsoft Word. It did not have Adobe Photoshop and a couple of other things. Now, first of all, when was the last time I opened a word processor? I just ask my agent to come up with the contract and then I will review it in markdown and when I'm happy I'll just have it make a PDF.

11:38 It looks amazing. So, I've been setting up the Omicon Foundation which is the nonprofit behind Omachi. We've raised I don't know $23 million in tokens and donations and I've had to make a lot of contracts. Every single one of those contracts made by an agent were no word process was involved. I don't even sign my documents anymore. I just look it over and say, "That's good to go.

12:00 sign it and the agent has my signature and it'll put it on the document and then it will mail it out. So, I don't go through word process anymore. I don't need word anymore. The same thing with photos. You want something changed about your photo, make this look great. What? You're going to spend tens of hours, hundreds of hours learning the intricacies of Photoshop.

12:19 No one's going to do that. They're going to ask their agent to do it. And the agent is going to do a phenomenal job so that your photo comes back looking amazing. or you want to remove someone in that photo, you just say like, "Hey, remove that idiot over there on the left. I just want to look at these two people." Boom. So, I think the agents, they're going to just eradicate entire classes of applications because these applications historically, especially something like Photoshop, was not just an application.

12:43 It was a profession. It was something you had to learn and spend in some cases years getting good at. Now, that's all great, but agents have taken all that on and now they can do it without the app. But then, even if you do want the app, what I've noticed and what I've seen happening in the community a lot already is that whenever you need something of a word or you need something of a Photoshop, you can on the fly basically just make this happen.

13:12 Now, we're getting this weird early stage look of what AI is, and it runs at about 50 tokens a second. And if you turn on ultra fast, it runs at 12. That's not enough for things to feel alive. Like, uh, Fireworks just two weeks ago at their big conference demonstrated 1,500 tokens a second. They were doing an example with a game. In four seconds, they went from the prompt, "Hey, can you make me a Tetris game?"

13:39 to the game running. Then they made another game. Let make me a sidescroller, a shooter. Um, the game is up on the screen. The game is running. He just prompts at it. Hey, add lasers. Oh, great. Add bombs to it. No bigger monsters. Everything just mutates live. That was such a preview of what the future of software is going to be. Everything is just going to mutate live.

14:02 And that's 1500 tokens a second. We can do that today. What's going to happen next year? What's going to happen two years from now when these models are going to run at a 100,000 tokens a second, a million tokens a second. So directionally, we're going towards a place where all these historic barriers for anyone to launch and succeed with a new ecosystem is just going to disappear because not only can the ecosystems be created far more easily, I mean, the current version that we have was made over the course of about

14:31 two and a half months part-time. me and Ryan Hughes, my main partner on this project. That was the version that turned into this thing that's now taking off like crazy. What we're competing against thousands of programmers over at Microsoft, tens of thousands of programmers around the world working at these huge companies and now suddenly a slap thing here of a handful of people can make something that's not just competitive but superior.

14:58 That's completely unheard of. That was never going to happen before. Now you look at venture capital, right? Like there's literally a what two decade long history where all the money just plowed into developers because you needed developers to build the thing and you also needed a runway of several years to have these developers build things. Now suddenly all of that is going getting compressed down to days, weeks, at worst months.

15:24 And that's what the models we have right now. Just at the time I've been developing Omachi, we went from um whatever opus 48 was what I started this new phase of it and it was it was good and then we got soul and now we got opus 55 like the distance between opus 48 and 55 >> enormous. Yeah. >> And are we going to get another drop before Christmas? Probably.

15:47 And if not then we're going to get it early next year. And these hops are accelerating. So anyone who still thinks that there is moat in source code even millions of lines of it or that that is going to protect their ecosystem are delusional. >> You know it's it's interesting for uh for for me at least to hear you speak this way because I've always viewed you as such a craftsman.

16:13 Like your your design aesthetic is so good. Your ergonomics are so good. I mean you're you're super legendary and it's kind of funny. I play Starcraft. I still play Starcraft. I'm 50 years old. Um, and they just came out with a patch recently where they updated the stats of the ghost. Like the ghost is like, you know, it's been around for 20 years and it's like these minor tweaks.

16:31 And so, like, at some level, like I I I get this like we will just vibe, not vibe, but we will just like have a an LLM code something very quickly. On the other hand, >> I look at what you've done and I look at games like Starcraft. I realized that like software is these minor tweaks that are kind of divorced from the lines of code and and in at some level over time that's what it is.

16:56 It's not the lines of code and like that feels like a moat to me. And I actually think you know if you're if you're playing or dealing with these things like this longtail design choices is what it is. And so is it is it also the case that that maybe the change isn't that big and you still need people that love software and will kind of imbue it in the same way?

17:13 It's just you've lost the writing the lines of code. >> That's exactly it. I feel like I can get more of my taste, more of my vision, more of my perfectionism into the software because the barriers between me spotting something that's not quite right, a few pixels that are off and wanting to fix it has just collapsed. >> I love it. I now see any problem, any problem at all, any piece of software that is not perfect as an opportunity to go pro until it's perfect.

17:45 This is why I'm working 16our days right now. Now, historically, I wrote a whole book called It doesn't have to be crazy at work. It's totally part of your manifesto is is not >> it's totally [ __ ] crazy at work right now. And the reason it is so crazy is that if you are of that type of personality that I am, that complete perfectionism, that craftsmanship of let's just keep whittling things away, you've just been granted superpowers and that is intoxicating if not downright um sort of dangerous to your to your health

18:17 and sleep patterns at least at least for a while, right? But >> I think that that is the moment that we're going through. You're completely right like anyone could do this. Anyone could have done a lot of things over long periods of time and they didn't. It still takes some sense of vision towards this. It still takes some sense of care and it takes wanting to put all the pieces together and having some determination to see it through.

18:39 Now, what we're seeing with a lot of vibe coding projects is they start out and they're kind of promising and then someone loses interest and they just get to the most gratifying for some maybe even for most people is the first 80%. >> Yeah. let me just get something done in the first 80% and then the last 20s where things are a little jangy they're not quite right they have some bugs whatever and they lose interest now for whatever reason I am blessed with the entire dopamine payoff in the last 2% I don't actually

19:09 give a [ __ ] about the first 98% I want the last two >> the last two is where I get my satisfaction of building software right so if you are of that type that you just keep adding and keep adding keep at it until it's literally perfect. This is a extremely addictive moment in history because there at least for the time being is still a place for someone to drive these agents.

19:33 I've not yet found the correct loop or setup where you can just oneshot everything and the design comes out perfect and everything comes out perfect. You need the eye for a moment still and to be able to supply that and then feel like you have an entire army below you with no coordination cost. This is one of the things I've already said. I was never interested in trading my company at 37 Signals.

19:58 We're about 60 people for a company of 6,000. Yeah, sure. Or 60,000. You could do more, but then I'd have to deal with [ __ ] 60,000 people. That sounds like misery. absolute pain and torture. >> Yes. >> But dealing with the productive potential of 60,000 people through agents. Now, now you have me interested, right? No [ __ ] argument, no HR, no committee meetings, just me, high bandwidth with a fleet of agents executing on a distilled vision for what the perfect computer could be like.

20:33 Holy crap. Um Toby Ludka calls this when he does it over the weekend and then has to go back into the office of a literal company of 7,000 people. Like coming out of hyperdrive >> when it's just you and the agents, you're in hyperdrive. Everything's going amazing and then screech, you go into the office, you got to deal with [ __ ] HR and teams and updates and reports and all the other nonsense that involves running a company.

21:00 And um you don't have to do that if you do it this other way. and you can get similar, if not better, results. That's what's so mindblowing about this current moment. Even more so is this um fact of just how cheap it is. I mean, I made the quip on X, I don't know, a month ago. I have unlimited tokens, right? Because, okay, I've been working for 20 years.

21:25 I'm rich at this point. I could buy as many tokens as I want. Do you know how many tokens I buy? I got three claw subscriptions. That's it. And I can remember two weeks out of the last three months where I actually maxed or got close to maxing those out. Most weeks I'm on one $200 20x sub and I could do everything. So don't tell me that this moment is just reserved for people who literally have unlimited money.

21:52 It totally is not. And this is Frontier stuff. We're not even talking about openweight models or any of the cheaper models or running things with Luna or optimizing any of those things. I'm just [ __ ] Opus 55 fix everything and there we go, right? Like I am saturated in terms of what I can move forward and we can move forward at this lightning pace.

22:13 So yeah, I think there's still room for humans, but I'm also not so vain to believe that that's necessarily going to last forever. I think it's entirely possible that by next year we will have true super intelligence that's just better at humans at everything and they will be better even than me at polishing the perfect computer. And you could just say me, make me a great computer.

22:31 Whoopity doop doop. It does a Linux kernel bespoke for you and everything else. And uh and there we go. So I'm savoring this moment where the human for maybe a blink of an eye is still extra, is still additive to the process while having this amazing capacity. And then I don't know, maybe we're all dead at the end of next year. So enjoy it. I have to ask.

22:55 So, do you have in your head your like David's super intelligence test that you like test every So, I'll give you an example. So, here's my super intelligence test is um every time there's a new model, I I ask it to like go buy all of the stuff it needs to embody itself. Like, like my house constant my my house constantly has like robotics parts showing me.

23:19 I'm not kidding. So, I gave it I give it like a $20,000 budget and I'm like, just go build your robot, you know, and then so things show up and like, you know, OLED displays show up and, you know, whatever. And like we're just not at the point where it's even close to being able to do that. But that's like my ASI test. As soon as I have like a walkable autonomous robot that it built itself by ordering parts.

23:42 So, I'm just wondering if you have like like a similar like now you know it's arrived. I don't. I probably should, but what I've have is this sensation that for software de development, we already have super intelligence. >> Oh, for sure. Yeah. >> The the AI is better than any programmer I've ever known when you consider the breadth of its ability. There are individual parts where individual super programmers I know are still able to one up the agents in very narrow fields.

24:11 But you take the whole thing together and you take the pace by which they do it, we already have super intelligence, right? So I can build anything literally anything that I want just by prompting and and sort of steering a little bit. Right. But what's interesting about your example with the robots was I was just in China whatever two weeks ago and we went to see unit and we went to see the other one talking to the guys at Unity Tree.

24:35 They have a very different perspective of this from what I hear from Silicon Valley which is I was just watching this Doress interview who's talking to someone who's saying like by next year maybe we're going to have the super intelligence going to have rapid take off unit guys are like ah I don't know four years five years maybe 10. >> Yeah. So, there's a weird mix here where the folks working on the software side of things are like, "Oh my god, this is just about to happen."

24:58 And the people working on actuators and mechanics and whatever are like, "Ah, I don't know, maybe it's actually quite hard to peel an onion or whatever." >> So, for what it's worth right now, I mean, Grockbot thinks that the right way to like have an AI in my house is basically home automation. So, it keeps ordering home automation stuff for my house.

25:17 So, I I think we're going to get like I think at least I'm going to get the Star Trek like the display on the wall with the face that talks to you that like controls the house before the actual. But I I think it's very pragmatic actually on its part. >> By the way, this was its decision, not mine. So, >> and this is how you end up a paper clip, by the way, if you're not careful yourself.

25:38 >> You're like, "Here's a vague goal. Optimize human happiness." And bam, you're a vague. >> We need Martine to pace his frontier. Now please please Martin can you by the way can you talk a bit about you know you mentioned open source I've always viewed you so iconoclastic you know and having strong opinions on kind of consolidation of power you've you've already talked about it now like how do you think about the frontier labs versus open source >> I think it's absolutely essential that we have open source and we have

26:04 open weights to keep the whole game honest we need competition I mean I have spent >> two phases of my career being very dismayed by duopolies splitting everything up. And we only got this brief refuge with the internet from about uh 2000ish to 2009. This glorious age where the most important platform was completely open, completely free. Anyone could build anything and you didn't have to ask for permission.

26:36 And then we got the mobile platforms. >> Yeah. >> And the game was up. >> Yeah. Suddenly you went back into this draconian uh economics scenario where you had to beg for permission for your application to be out there and therefore for your business to exist. >> Yeah. >> And I just find that personally completely intolerable. So, as much as I appreciate Opus 55 and what Open AI is doing, I think it would be a nightmare scenario if we end up with a duopoly position, especially one that's enforced by the government with

27:09 big guns. That would be very bad for us. Now, this industry went through something kind of similar with the encryption wars in the '90s where we got pretty close to banning public key encryption because NSA and others said like, well, how are we going to ease drop on everything? Like, imagine the trajectory of the internet if we did not have encryption.

27:33 >> How you going to do commerce? How you going to do communication? It would just have been an enormous injury to the world economy, to the development of the world, and to the web itself. And I think we are at that precipice with AI right now. I keep hearing plenty of smart people say like, "Ah, maybe the government should just take it over." Jesus.

27:53 >> Yeah. >> And even if they don't, even if we just end up in a commercial situation where these two labs seemingly divide the market between them, that would be catastrophic. I mean, we need to avoid that. Now the good thing is that there is more competition within closed models than any other area or phase of computer industry that I've ever observed.

28:16 The fact that we don't just have these two labs that are ahead, but we have SpaceX and AI quite close to it. We have uh Google that seemingly one moment is in the game and then the next moment somewhere off smoking something funny on a island somewhere. But there's at least some potential. And then you have a handful of other labs that are nibbling at it as well.

28:37 Now I think that's hugely important. But what's even more important is that we have these openweight models that are getting really good too. I mean I don't know how to quantify exactly how far behind but GLM53 or Deepseek Flash >> I don't know six seven months behindish on some tasks not even. There's a lot of benchmaxing going on and I don't really buy that.

29:01 They put out the one benchmark. Oh, it's just as good at uh as whatever Opus 48 and then I try I'm like ah I don't know on your benchmark maybe not in the real world but so what they're in the game they are in the game these models are useful you can run them on your own local hardware this is the thing that's mind-blowing to me I wrote a piece I don't know six months ago or something basically saying like this is stupid why people buying fast personal computers the models we had at the time that could run locally were

29:28 so lame like as in they couldn't do anything. They couldn't move anything that mattered. And then thankfully I was totally wrong. And here we are not even at the end of the same year. And a DGX Spark can run a useful model you can actually use for something. Uh Quen 38 uh Flash Next is a totally usable model. There is intelligence in a box that you can fit on your damn desk that you can buy for singledigit thousands of dollars.

29:57 That's incredible. and we desperately need that. But I want all of it, right? There are other people in the open source community that are very hardline on this. It is either open- source or bust. I don't subscribe to that at all. I cheer for the fact that the US have these two frontier labs and that they do have the lead because we get to then build with that stuff.

30:19 And I think it's amazing as long as they're kept honest. as long as they don't just have an infinite money glitch where there's no competition and they can just turn the price of tokens up to $500 a million or something like that because we would pay. We would absolutely pay. And we've seen that story play out. When social media took off and the web at large sort of took off, it got divided between two companies, Google and Meta.

30:42 And those two companies sucked out 99% of the value chain in online advertisement because everyone was willing to just get one extra incremental cent back on every dollar they spent. And to me that's perverse. As much as I appreciate the contributions to some extent in some areas that these two companies have made, that is dystopian when you capture essentially 99% of the value being created.

31:10 who labs would totally do that if left to their own devices because that's just how capitalism works and in fact I think it's kind of a feature not a bug most of the time because that's what inspires all this investment and everyone wants to end up with one of these companies everyone wants to end up with shares in the next Google and meta or now anthropic or or open AI but all these ideas in your head at the same time we need the open weights we can cheer on the advances at the frontier if we get all of it everything

31:40 just gets better for everyone and we get to live in this ploom scenario where there's just infinite excess and prosperity for all. So David, one of the things I noticed is like we're all sort of climbing the ambition ladder as these models get more and more capable. This is actually one of the reasons I think even for the people who are building software where they get the dopamine hit in the first 10%.

32:01 It turns out that you need to ratchet up the level of ambition to keep getting the hits, right? Because you build something that would have been magical 5 years ago and you know now you're like this kind of sucks. >> You want to get a like on X. >> You're like 3D thing. It's like no I built this like selfmodifying game that does all these incredible things.

32:17 You get like 40 likes. What am I doing here? Right. >> So I think it's actually cool and like here we are talking about this new desktop Linux revolution that you're leading in this distribution. So, what does it look like for your ambition to be taken to the next like what is something that you never envisioned you'd be able to make that you might be building in the next 12 or 24 months?

32:38 >> Well, I'd already acknowledged the ambition inflation that's been happening on this project and I only just started it last year and I started by hand if you could believe it. I was writing out little bash scripts with my 10 fat fingers um swearing at bash a fair amount and being very limited in what was possible. I just thought, hey, do you know what?

32:58 If I could make a nicer computer for myself, a place I can live where Apple doesn't control my destiny. I'll be happy. I'll do that. I'll do it for me and whatever my five friends and a few people on the internet. And we did that and it was it was going well. We had tens of thousands of people using like nice here. got a nice little uh open source project going and I have a better computer.

33:18 Wonderful. Then we got the agents and I went from like, "Oh, it'd be nice to build a slightly better computer to oh my god, we can literally build the perfect computer. Everything I've ever wanted out of a computer over the course of 40 years where I've been using these machines, I can now make happen more or less by myself. Now, it's more fun to do it with friends.

33:42 So, I've invited a few hundred thousand people to come along, but by myself, I can make the perfect computer. And knowing that this is not just possible, but seeing it happen, seeing people experience Linux as not just an alternative to a Mac OS, not just an alternative to Windows, but vastly superior is so damn gratifying that of course you're going to get delusions of grandeur.

34:09 And my current delusion of grandeur is that within, let's call it 3 years, we're going to be bigger than Mac OS. There is only 180 million Mac OS users in the world. The Mac is not that big. I know it looks like it in America and especially if you walk around Silicon Valley where everyone uses a Mac, but outside of that, tiny. 180 million. Like that's not even a [ __ ] Ctier internet service, right?

34:32 Like biggest of these services have billions of users. Of course, we can get Linux above that. Of course, we can. Now that sounds delusional because the Linux folks have literally been talking about that prophecy for 25 years making Linux happening on the desktop and now suddenly we're here and we're accelerating and it's happening right so that's how my ambition ladder has been escalating just in the last I call it like let's say four or five months or maybe not even that seeing the traction we've gained with this

35:03 latest version of um the quattro release where we went all in on the agents we built it with agents as we put agents in the hands of everyone and seeing hundreds of thousands of people pick it up. I went like, well, okay, maybe 3 years. That's the delusional thing. Hey, how about let's check in in 12 months. Could we unseat the Mac in 12 months? Something that they literally spent whatever two decades building up this position with uh with Mac OS.

35:29 And before that, I mean, this is a company from the goddamn 70s that had a lineage that led to that point. I think it's entirely possible. Now, I wouldn't put majority odds on it for the next 12 months, but I also wouldn't rule it out. I mean, if there are smart people at the labs thinking we have about 12 months left before recursive self-improvement is going to give us the super god of AI, well, surely we can kick off a whatever second ranked operating system in the world.

35:57 Now, and this is just me being ambitious within my own little circle here, within my own area of focus. Um, I'm actually not that good at being ambitious on the grand scale in the same way. I mean, I I leave that to the Elons of the world. Like, let's populate all the planets. That sounds awesome. Just give me a seat on the Starship going to Mars. I'd love to do it.

36:18 That's not me. Um, I try to be ferociously ambitious within this narrow scope and put all the energy into that. And then also just not try to think about what 12 months from now looked like because I find that that is the shest way to develop a severe case of AI psychosis. >> Yeah. >> I like to have AI euphoria. I think that's a much milder version of it where you're just really excited and optimistic about the future because you're focused on what's possible right now.

36:47 What you're building right now with the tools as they look. You try to imagine two, three, four model hops ahead. You're just going to go crazy, right? Because then why is there even a role for anyone? What do humans even need to damn well exist? Are we all going to end up as paper clips? I think that is a bad path to go down. And I think this is why so many of these AI safety people end up [ __ ] crazy cuz that's what they do all day long.

37:10 They just sit and think about like what if all the worst things happened over the next 12 months? Wouldn't it be bad? Yeah, probably [ __ ] would. If you were sitting down next to Oenheimimer in 44 and going, "Oh my god, by the time this thing is out, can you imagine what's going to happen in 5 years? It's going to be a desolate planet. Nuclear waste everywhere."

37:33 And what happened? The bomb was the best thing that ever happened to world peace. We literally got 50 years of a beautiful cold war that did not include total nuclear annihilation. So very hard to predict the future. it is. I try to stay out of that for as much as possible. And the only prediction of the future I enjoy is the William Gibson quote. The future's already here.

37:56 It's just not evenly distributed. So the future I'm seeing as I'm working with the system, I feel very comfortable doing some predictions on if everyone was able to do what I'm doing right now and could see what was possible. Of course, GDP acceleration of 3, four, 5, 10% a year is within the realm of the possible. So that seems easier to predict, but where the models are going to be in a year.

38:22 Yeah. Shake the eightball. Ask again. You know, David, uh, actually Anish and I were talking before, you know, right before this. One very unexpected, wonderfully unexpected phenomenon with this whole AI wave is how many of like the great that we've grown up with kind of showed up and helped guide the way. And then listen, there's a lot of confusion.

38:46 There's a lot of fear and like it was you and it's Eric Raymond like he's back and it's amazing and it's John Carmarmac and like these just absolute stalwarts from the last 30 years. And so first off, thank you so much for everything you've done. I think your talk by the way where you're like you know software is over now we're builders like I was like yes you know by myself you know I really think that you've kind of you're leading a a generation through this kind of very confusing time so thank you and then B like >>

39:15 who who do you look towards as as you do these things is kind of also kind of a guiding light who you think is like right about this who you think we should kind of follow more as we kind of march forward because I just think we're at ultimate confusion right now and Not everybody lives through the internet like we did and not everybody lives through the great recession like we did and not everybody lives through kind of like the mobile confusion share economy and so I I do think that a I think what you're saying has

39:41 been very helpful to people and you know the people that maybe you look up to could also provide a similar level of guidance. Yeah, I think it is very interesting that everyone that I deeply respect for the programming chops, 100% of them are now fully pilled, >> fully into it because of course they are because anyone who have built something the hard way and achieved something with it, undeniable that these tools are able to accelerate that work, whatever that work is.

40:16 And I guess for perhaps reasons of success in many cases they're not threatened by it because I think we have this uncomfortable middle which is I don't know if it's unfortunate but just is most programmers most places are not comfortably independently wealthy. They are somewhere in their career where they feel like they've earned a place because they put in a lot of hard labor to get there.

40:43 They learn a bunch of skills and suddenly they're feeling the anxiety that what if it all goes away tomorrow? What if we all stop being programmers now and then we're all just makers and like am I a good maker? I don't know. So I do think we have to have some sympathy with that as you should have had sympathy with >> autoworkers in the 70s in Detroit as you should have had sympathy with the Ludits, the original weavers in England.

41:09 as you should have had sympathy with plenty of people who felt intimately on their own profession the disruption that's coming now it is possible to have that have the empathy and also go yeah I'm not going back to pre-industrial age I'm not [ __ ] substance farming here with my own hoe and an ox that is not a world I want to return to I don't even want to return to the assembly line work world where an auto worker in Detroit could make a decent living by putting things muttonally together by hand.

41:40 I don't think we should return to that. We got to lean forward into it. We got to find a way to to bring everyone along into that. And some of that is vibes. Some of it is vibes. And right now there is a battle of the vibes. And in that battle you have a bunch of people who for whatever reason maybe their own personal anxieties which can be very real and sometimes also just sort of finds this audience where like I'm going to be the anti-AI person here and there's not just one there's like a billion of them.

42:09 Whole communities actually are getting addicted to this stuff. If you draw the Ben diagram with the most insufferable woke people with blue hairs and the anti-A people, it's a [ __ ] circle, right? So in some ways we are all fidget we are replaying the last round we're replaying every battle ever because that's how the law lines are now drawn up and I think unfortunately the way for them that is the way this uh cookie crumble was that the people they didn't like leaned into this first and that left them only one

42:45 position to take which was the opposite >> and therefore they had to become anti-AI people right So, I mean, I do think there's a beautiful irony in there because there's a bunch of those insufferable people. I don't mind at all having a bit of anxiety about the future here, right? Um, a bunch at the rest of it had quite some anxiety about the future about 5 years ago when um they seemingly had all the cultural power and could cancel anyone, anytime, any place.

43:10 Now, the shoe's a little different, isn't it? But not all of it. And we shouldn't kind of fall into that trap that it's all just going to be tribal and there's this side or the other side or that everyone who is uncomfortable with AI is a blue-haired idiot. It's just that every blue-haired idiot is also anti- AI. But um getting to to sort of bring them along and to realize or or help them realize that you know what you can feel about this whichever way but there is only one way forward and that's through.

43:44 You're we're not going back. No one is going to willingly give up this technology ever unless we end up as paper clips on a stack somewhere in AI warehouse. We're not giving this up. This is here now. The world is just different as it was so many other times before. After the Wright brothers flew, the world was different. We're not going to ship everything with with boat and so forth.

44:07 So, we have this obligation to push in. And who I look to in that is first all my fellow travelers in this. I mean, as you mentioned, Carmarmac or Tim Sweeney or Mitchell Masimoto or there's a million of these people I deeply respect who've worked on software products that I've been using my entire career. 100% of them are leaned in. They're pil um then I also look to this really curious phenomenon that's been happening in both the Linux world but also in the AI world.

44:38 The return of the pseudo anonymous hacker. Now, the security world has always had this, but now it's leaking into the mainstream. On the Umachi uh core team and the Omicron Foundation staff list, we have two pseudo anonymous people. one outf foxed the creator of quick shell the malible shell that we're using for all the desktop interactions and handcore one of the artists in there >> which to me is a total throwback to the '90s and like freaker culture elite message boards and all this stuff and I love that I love that

45:14 we have this >> energy around computers again like it was in the '9s the 80s where hackers and experimenters are suddenly at the foreront front of everything because maybe that was true for a split second with the internet and then the internet got commercialized and then it got taken over and we got platforms and we got social media and then it was dead.

45:34 There was none of that anymore, right? It didn't last that long as sort of this free spirit. Local AI for example has a ton of this stuff too. Um 0x0 is someone we uh sponsor through the Yamcom Foundation. Mia's lab um also both pseudo anonymous 000 is I guess his identity is known but goes by a handle. I love this energy that now individuals thank to the power of AI can actually play a pivotal role in the forward progression of the industry because as I said once the internet got sort of bottled up and commercialized

46:09 a lot of the sort of frontier work to push things forward just got gobbled up by the big companies. Oh, you had Google doing Chrome and then that was the only browser in town and then to create another browser was basically impossible after the Linux kernel, the browser engine is the most complex piece of software humanity has ever produced. So we just we closed off all these domains of the consequential bits of the computer and only the biggest companies in the world could work on them.

46:37 And now we're redemocratizing it. We're returning it to the hacker culture. returning it to the indiv individual who can actually move some of this stuff forward. So, I look to those people basically anyone doing cool [ __ ] with a pseudo anonymous handle online or like, hey, hey, I got to I got to follow this person. Something's going on over here.

46:56 Um, I love that. It's like it's anti-LinkedIn culture. Anti-LinkedIn culture is what it is, right? It's just the complete opposite. And a big part of it is infused by X. I >> X has played such an enormous connecting tissue role for this whole revolution for the artist, for the pseudo anonymous hackers, for the local AI movement is all getting connected through X in a way that if um those blue-haired bandits had somehow managed to hold on to that [ __ ] back before Elon brought it, we might not have had any of it at all.

47:30 It might all have ended up being linked in human slob. Yeah. I just wanted I just very very quick throwback also another thing that I love. Do you remember demo culture in the 90s? >> Oh my god. Yes, >> dude. It is so back in the creative aspect. It's amazing. Like some of the most sophisticated demos are by like these random, you know, >> artists that have no interest actually in like >> um you know, building businesses is purely for the love of art.

48:00 And I don't remember the last time we had such a movement and it's awesome. I'm trying to pull all of that into Amachi. So, I don't know if you've tried Amachi yet, but one of the signature visual effects is the screen saver. And the screen saver is literally asy art running through 40 demo effects, >> just looking totally out of like a 1995 demo party intro, right?

48:23 We're pulling all of those effects into it, keeping that terminal aesthetic, >> which is by the way also completely different aesthetic from Mac OS or from Windows because I think Linux historically a lot of the distributions thought if we would just make a Teemo version of Windows, everyone would love us. No, no one wants a [ __ ] cheap copy of some [ __ ] that wasn't that great to begin with.

48:45 They want something bespoke. They want something unique. And the aesthetic of Amachi is totally 90s demo culture. Totally that. That's exactly the kind of vibe that I'm going for. In fact, I just before we hop into this call, I'm working on some how to set up local AI on your own machine. And one of the things we have is booting a local AI model can take a while, like uh I don't know, a few seconds, maybe in some cases even a few minutes.

49:07 So, while it's loading, I jammed in another asy effect just like [ __ ] cool characters floating around on the screen. Does it have any functionality? No. Absolutely [ __ ] not. just there to make you feel like I'm using a cool ass computer because that's what the '9s and demo culture felt to me. Like >> I was into pirated games at the time, too. And all these pirated games would have this intro cracked by Razer 1911 or whatever and the text would swirl around and I just felt a little cooler, a little cooler that day

49:41 that I got to witness that. I mean BBS was like that too like that >> if you >> even if your goddamn job is pushing an Excel spreadsheet around at least you feel alive for that hot moment while a machining >> yeah like BBS culture was like that too you know I mean all the protocols you remember X modem Y modem Z modem Kermit why is it Kermit no one knows but it seems badass >> I ran an elite bulletin board system out of my bedroom.

50:11 Three lines in. I was 14 years old. I can tell you when the guy from the Danish >> Telecom Monopoly showed up to install the third line and he saw this 14-year-old kid was like, "What the hell is going on here? This got to be illegal." And he was totally right. It was totally illegal. It was totally awesome. And I can still hear the US robotics >> modem sound, the 33.6 connecting.

50:40 It's just such an iconic sound. And we got to have more of that. Computers should be a little mysterious. They should be a lot of cool and they should not be so polished. As much as I love Apple, I think Jonathan IV essentially aluminum polished all character out of the computer in that regard. And maybe that was good for a time, but that time is over.

50:59 We're [ __ ] getting the hackers back. >> David, you're such a legend, man. We really appreciate everything that you're doing now and everything that you've done. Um, so we really appreciate you coming on and enjoy the conversation. >> Thanks so much. >> Well, thank you. It is always a pleasure to talk about the 90s demo culture, hackers, and cool [ __ ] >> Let's do more together, man.

51:18 What a pleasure. >> Yeah. >> Where are you basically? >> Absolutely. In person next time. >> By the way, is the office real? Is the office real? Like the the the picture of like your beautiful office view? >> It's very real. I'm not in that office right now. Otherwise, I would pan the camera around. But, uh, yes, that office in is in Malibu. And, um, >> gorgeous.

51:36 >> It never gets old. I mean, this is why I I keep posting that picture just for my own satisfaction. Like, I cannot believe I work here. And uh yeah, one of the spoils of capitalism. >> This office was brought to you by Commercial Software. Thankfully, >> really appreciate this. Really appreciate the time. >> Thank you. >> Thanks so much. >> Thank you for having me on. Let's do it again in person. All right. Sounds good. Take care, man. Byebye. Happy.