OpenClaw: The Viral AI Agent that Broke the Internet - Peter Steinberger | Lex Fridman Podcast #491

0:00 I watched my agent happily click the "I'm not a robot" button.

0:03 I made the agent very aware.

0:04 Like, it knows what his source code is.

0:07 It understands th- how it sits and runs in its own harness.

0:11 It knows where documentation is.

0:13 It knows which model it runs.

0:16 It understands its own system that made it very easy for an agent to...

0:19 Oh, you don't like anything?

0:21 You just prompted it to existence,

0:22 and then the agent would just modify its own software.

0:26 People talk about self-modifying software, I just built it.

0:28 I actually think wipe coding is a slur.

0:31 You prefer agentic engineering?

0:33 Yeah, I always tell people I'd- I do agentic engineering,

0:35 and then maybe after 3:00 AM,

0:37 I switch to wipe coding, and then I have regrets on the next day.

0:40 What a walk of shame.

0:42 Yeah, you just have to clean up and, like, fix your sh- shit.

0:45 We've all been there.

0:47 I used to write really long prompts.

0:50 And by writing, I mean, I don't write, I- I- I talk, you know?

0:53 These- these hands are, like, too- too precious for writing now.

0:56 I just- I just use bespoke prompts to build my software.

1:01 So, you, for real, with all those terminals, are using voice?

1:04 Yeah.

1:05 I used to do it very extensively,

1:08 to the point where there was a period where I lost my voice.

1:13 I mean, I have to ask you, just curious.

1:15 I- I know you've probably gotten huge offers from major companies.

1:21 Can you speak to who you're considering working with?

1:28 Yeah.

1:31 The following is a conversation with Peter Steinberger, creator of OpenClaw,

1:35 formerly known as MoldBot, ClawedBot, Clawdus, Claude,

1:39 spelled with a W as in lobster claw.

1:42 Not to be confused with Claud, the AI model from Anthropic, spelled with a U.

1:47 In fact, this confusion is the reason Anthropic

1:51 kindly asked Peter to change the name to OpenClaw.

1:55 So, what is OpenClaw?

1:57 It's an open-source AI agent that has taken

1:59 over the tech world in a matter of days,

2:03 exploding in popularity, reaching over 180,000 stars on GitHub,

2:08 and spawning the social network mold book,

2:11 where AI agents post manifestos and debate consciousness,

2:14 creating a mix of excitement and fear in the general public.

2:19 And a kind of AI psychosis, a mix of clickbait fearmongering and genuine,

2:25 fully justifiable concern about the role of AI in our digital,

2:29 interconnected human world.

2:30 OpenClaw, as its tagline states, is the AI that actually does things.

2:35 It's an autonomous AI assistant that lives in your computer,

2:39 has access to all of your stuff, if you let it, talks to you through Telegram,

2:44 WhatsApp, Signal, iMessage, and whatever else messaging client.

2:48 Uses whatever AI model you like,

2:51 including Claude Opus 4.6 and GPT 5.3 Codex, all to do stuff for you.

2:58 Many people are calling this one of the biggest

3:01 moments in the recent history of AI,

3:03 since the launch of ChatGPT in November 2022.

3:06 The ingredients for this kind of AI agent were all there,

3:10 but putting it all together in a system that definitively

3:13 takes a step forward over the line from language to agency,

3:17 from ideas to actions, in a way that created a useful assistant that feels

3:21 like one who gets you and learns from you,

3:25 in an open source, community-driven way,

3:27 is the reason OpenClaw took the internet by storm.

3:32 Its power, in large part,

3:33 comes from the fact that you can give it access to all of your stuff

3:36 and give it permission to do anything

3:38 with that stuff in order to be useful to you.

3:42 This is very powerful, but it is also dangerous.

3:46 OpenClaw represents freedom, but with freedom comes responsibility.

3:50 With it, you can own and have control over your data,

3:55 but precisely because you have this control,

3:57 you also have the responsibility to protect

4:00 it from cybersecurity threats of various kinds.

4:03 There are great ways to protect yourself,

4:05 but the threats and vulnerabilities are out there.

4:09 Again, a powerful AI agent with system- level access is a security minefield,

4:13 but it also represents the future.

4:16 Because when done well and securely,

4:18 it can be extremely useful to each of us humans as a personal assistant.

4:23 We discuss all of this with Peter,

4:26 and also discuss his big-picture programming and entrepreneurship life story,

4:31 which I think is truly inspiring.

4:33 He spent 13 years building PSPDF Kit,

4:36 which is a software used on a billion devices.

4:40 He sold it, and for a brief time,

4:43 fell out of love with programming, vanished for three years, and then came back,

4:48 rediscovered his love for programming, and built, in a very short time,

4:53 an open source AI agent that took the internet by storm.

4:56 He is, in many ways,

4:58 the symbol of the AI revolution happening in the programming world.

5:02 There was the ChatGPT moment in 2022, the DeepSeek moment in 2025, and now,

5:08 in '26, we're living through the OpenClaw moment, the age of the lobster.

5:16 The start of the agentic AI revolution.

5:18 What a time to be alive.

5:20 This is a Lex Fridman podcast.

5:22 To support it, please check out our sponsors in the description,

5:25 or you can also find links to contact me,

5:28 ask questions, give feedback, and so on.

5:31 And now, dear friends, here's Peter Steinberger.

5:36 The one and only, the Clawed Father.

5:39 Actually, Benjamin predicted it in his tweet.

5:41 "The following is a conversation with Claude,

5:44 a respected crustacean." It's a hilarious-looking

5:47 picture of a lobster in a suit, so I think the prophecy has been fulfilled.

5:53 Let's go to this moment when you built a prototype in one hour,

5:59 that was the early version of OpenClaw.

6:01 I think this, Story's really inspiring to a lot of people because

6:05 this prototype led to something that just took the internet by storm....

6:10 and became the fastest-growing repository in GitHub history,

6:13 with now over 175,000 stars.

6:15 So, what was the story of the one-hour prototype?

6:20 You know, I wanted that since April.

6:23 A personal assistant.

6:24 AI personal assistant.

6:26 Yeah.

6:26 And I, I played around with some other things,

6:29 like even stuff that gets all my WhatsApp, and I could just run queries on it.

6:36 That was back when we had GPT-4.1, with the one million context window.

6:41 And I, I pulled in all the data and then just asked him questions like,

6:47 "What makes this friendship meaningful?"- Mm-hmm.

6:50 And I got some, some really profound results.

6:55 Like, I sent it to my friends and they got, like, teary eyes.

7:00 So, there's something there.

7:01 Yeah.

7:02 But then I...

7:02 I thought all the labs will, will, will work on that.

7:06 So I, I moved on to other things,

7:09 and that was still very much in my early days of experimenting and pl- playing.

7:14 You know, you have to...

7:16 That's how you learn.

7:18 You just like, you do stuff and you play.

7:20 And time flew by and it was November.

7:24 I wanted to make sure that the thing I started is actually happening.

7:27 I was annoyed that it didn't exist, so I just prompted it into existence.

7:36 I mean, that's the beginning of the hero's journey of the entrepreneur, right?

7:39 And you've even with your original story with PS PDF kit,

7:44 it's like, "Why does this not exist?

7:45 Let me build it." And again,

7:47 here's diff- a whole different realm, but similar maybe spirit.

7:52 Yeah, so I had this problem.

7:53 I tried to show PDF on an iPad, which should not be hard.

7:57 This is like 15 years ago, something like that.

7:59 Yeah.

7:59 Like the most, the most random thing ever.

8:02 And suddenly, I had this problem and I, I wanted to help a friend.

8:06 And there was, there was...

8:08 Well, not like nothing existed, but it was just not good.

8:11 And like...

8:11 Like I tried it and it was like very,

8:14 "Nah." Like, "Hmm, I can do this better."- By the way,

8:18 for people who don't know,

8:19 this led to the development of PS PDF kit that's used on a billion devices.

8:23 So, the...

8:23 It turns out that it's pretty useful to be able to open a PDF.

8:28 You could also make the joke that I'm really bad at naming.

8:32 Yeah.

8:32 Like, name number five on the current project.

8:35 And even PS PDF doesn't really roll from the tongue.

8:40 Anyway, so you said "Screw it.

8:42 Why don't I do it?" So what was the...

8:44 What was the prototype?

8:45 What was the thing that you...

8:47 What was the magical thing that you built

8:49 in a short amount of time that you were like,

8:51 "This might actually work as an agent," where I talk to it and it does things?

8:56 There was...

8:56 Like, one of my projects before already did something where I

9:00 could bring my terminals onto the web and then I could,

9:04 like, interact with them, but there also would be terminals on my Mac.

9:07 Mm-hmm.

9:08 Viptunnel, which was like a, a weekend hack project that was still very early.

9:14 And it was cloud code times.

9:16 You know, you got a dopamine hit when you got something right.

9:19 And now I get, like, mad when you get something wrong.

9:22 And you had a really great-– not to take a tangent-–

9:25 but a great blog post describing that you converted Viptunnel.

9:28 You vibe-coded Viptunnel from TypeScript into Zig

9:32 of all programming languages with a single prompt.

9:35 One prompt, one shot.

9:37 Convert the entire code base into Zig.

9:41 Yeah.

9:41 There was this one thing where part of the architecture was...

9:46 Took too much memory.

9:48 Every terminal used like a node.

9:52 And I wanted to change it to Rust and...

9:55 I mean, I can do it.

9:57 I can, I can manually figure it all out,

10:02 but all my automated attempts failed miserably.

10:08 And then I revisited about four or five months later.

10:11 And I'm like, "Okay, now let's use something even more experimental." And I,

10:16 and I just typed, "Convert this and this part

10:19 to Sig," and then let Codex run off.

10:23 And it basically got it right.

10:24 There was one little detail that I had to, like, modify afterwards,

10:28 but it just ran for overnight or like six hours and just did its thing.

10:35 And it's like...

10:36 It's just mind-blowing.

10:39 So that's on the LLM programming side, refactoring.

10:43 But uh, back to the actual story of the of the prototype.

10:48 So how did Viptunnel connect to the first prototype where your, like,

10:51 agents can actually work?

10:52 Well, that was still very limited.

10:54 You know, like I had this one experiment with WhatsApp,

10:57 then I had this experiment, and both felt like not the right answer.

11:03 And then my search bar was literally just hooking up WhatsApp to cloud code.

11:10 One shot.

11:11 The CLI message comes in.

11:13 I call the CLI with -p.

11:16 It does its magic, I get the string back and I send it back to WhatsApp.

11:20 And I, I built this in one hour.

11:23 And I felt...

11:25 Already felt really cool.

11:25 It's like, "Oh, I could...

11:27 I can, like, talk to my computer," right?

11:29 This...

11:29 That, that was, that was cool.

11:31 But I, I wanted images, 'cause I alw- I often use images when I prompt.

11:36 I think it's such a, such an efficient way to give the agent more context.

11:39 And they are really good at figuring out what I mean,

11:42 e- even if it's like a, a weird cropped-up screenshot.

11:44 So I used it a lot and I wanted to do that in WhatsApp as well.

11:48 Also, like, you know, just you run around, you see like a poster of an event,

11:54 you just make a screenshot and like figure out if I have time there,

11:58 if this is good, if my friends are maybe up for that.

12:00 Just like images seemed im- important.

12:02 So I, I worked a few...

12:04 It took me a few more hours to actually get that right.

12:07 Um, and then it was just......

12:11 I, I used it a lot.

12:14 And funny enough, that was just before I went

12:19 on a trip to Marrakesh with my friends for a birthday trip.

12:23 And there it was even better because internet was

12:26 a little shaky but WhatsApp just works, you know?

12:29 It's like doesn't matter, you have, like, edge, it still works.

12:32 WhatsApp is just...

12:33 It's just made really well.

12:36 So I ended up using it a lot.

12:40 Um, translate this for me, explain this, find me places.

12:44 Like, you just having a clanker doing, having Google for you, that was...

12:49 Basically there was still nothing built but it still could do so much.

12:54 So, if we talk about the full journey that's happening there with the agent,

12:58 you're just sending on this very thin line WhatsApp message via CLI,

13:03 it's going to a cloud code and cloud code is doing all

13:06 kinds of heavy work and coming back to you with a thin message.

13:13 Yeah.

13:13 It was slow because every time I boot up the CLI, but it...

13:17 It was really cool already.

13:19 And it could just use all the things that I already had built.

13:23 I had built like a whole bunch of CLI stuff over the month so it,

13:28 it felt really powerful.

13:31 There is something magical about that experience that's hard to put into words.

13:35 Being able to use a chat client to talk to an agent, versus, like,

13:43 sitting behind a computer and like, I don't know,

13:47 using cursor or even using Cloud Code CLI in the terminal.

13:50 It's a different experience than being able to sit back and talk to it.

13:53 I mean, it seems like a trivial step but, it- in some sense it's a...

13:58 It's like a phase shift in the integration

14:02 of AI into your life and how it feels, right?

14:05 Yeah.

14:05 Yeah.

14:06 I, I read this tweet this morning where someone said,

14:08 "Oh, there's no magic in it.

14:10 It's just like, it does

14:12 this and this and this and this and this and this." And it

14:16 almost feels like a hobby, just as cursor or perplexity.

14:19 And I'm like, well, if that's a hobby that's kind of a compliment, you know?

14:24 They're like, they're not doing too bad.

14:28 Thank you I guess?

14:32 Yes.

14:32 I mean, isn't, isn't, isn't magic often just like you take a lot of things

14:37 that are already there but bring them together in new ways?

14:41 Like, I don't...

14:42 There's no...

14:43 Yeah.

14:43 Maybe there's no magic in there but sometimes just rearranging things and, like,

14:47 adding a few new ideas is all the magic that you need.

14:51 It's really hard to convert into words what is, what is magic about a thing.

14:56 If you look at the, the scrolling on an iPhone, why is that so pleasant?

15:00 There's a lot of elements about

15:02 that interface that makes it incredibly pleasant,

15:04 that is fundamental to the experience of using a smartphone,

15:07 and it's like, okay, all the components were there.

15:10 Scrolling was there, everything was there.

15:13 Nobody did it-- Yep-...

15:14 and afterwards it felt so obvious.

15:16 Yeah, so obvious.

15:17 Right?

15:17 But still...

15:18 You know the moment where it, it blew my mind was when,

15:25 when I- I used it a lot and then at some point

15:26 I just sent it a message and, and then a typing indicator appeared.

15:32 And I'm like, wait, I didn't build that, it only m- it only has image support,

15:38 so what is it even doing?

15:40 And then it would just reply.

15:42 What was the thing you sent it?

15:43 Oh, just a random question like, "Hey,

15:45 what about this in this restaurant?" You know?

15:47 Because we were just running around and checking out the city.

15:52 So that's why I, I didn't, didn't even think when I used it because

15:56 sometimes when you're in a hurry typing is annoying.

15:59 So, oh, you did an audio message?

16:00 Yeah.

16:01 And it just, it just worked and I'm like...

16:04 And it's not supposed to work because-- No-...

16:06 you didn't give it that-- No, literally-...

16:08 capability.

16:08 I literally went, "How the fuck did he do that?" And it was like,

16:12 "Yeah, the mad lad did the following.

16:13 He sent me a message but it only,

16:16 only was a file and no file ending." So I checked

16:19 out the header of the file and it found that it was,

16:23 like, opus so I used ffmpeg to convert it and then

16:26 I wanted to use whisper but it didn't had it installed.

16:29 But then I found the OpenAI key and just used Curl

16:33 to send the file to OpenAI to translate and here I am.

16:39 Just looked at the message I'm like, "Oh wow."- You didn't teach it any

16:44 of those things and the agent just figured it out,

16:46 did all those conversions, the translations.

16:47 It figured out the API,

16:49 it figured out which program to use, all those kinds of things.

16:52 And you were just absent-mindedly just sent an audio message when it came back.

16:56 Yeah, like, so clever even because he would have gotten the whisper local path,

16:59 he would have had to download a model.

17:00 It would have been too slow.

17:02 So like, there's so much world knowledge in there,

17:05 so much creative problem solving.

17:06 A lot of it I think mapped from...

17:09 If you get really good at coding that means you

17:11 have to be really good at general purpose problem solving.

17:14 So that's a skill, right?

17:15 And that just maps into other domains.

17:18 So it had the problem of like, what is this file with no file ending?

17:22 Let's figure it out.

17:23 And that's when it kind of clicked for me.

17:27 It's like, I was like very impressed.

17:30 And somebody sent a pull request for Discord support and I'm like,

17:35 "This is a WhatsApp relay.

17:37 That doesn't, doesn't fit at all."- At that time it was called WA Relay.

17:43 Yeah.

17:43 And so I debated with me like, do I want that?

17:48 Do I not want that?

17:50 And then I thought, well maybe,

17:53 maybe I do that because that could be a cool way to show people.

17:58 Because I...

17:59 So far I did it in WhatsApp as like groups you know

18:02 but don't really want to give my phone number to every internet stranger.

18:07 Yeah.

18:08 Um, journalists manage to do that anyhow now so that's a different story.

18:12 So I merged it-...

18:16 from Shadow, who helped me a lot with the whole project.

18:21 So, thank you.

18:22 And, and I put my, my bot in there.

18:27 On Discord?

18:28 Yeah.

18:28 No security because I didn't...

18:30 I hadn't built sandboxing in yet.

18:32 I, I just prompted it to, like, only listen to me.

18:38 And then some people came and tried to hack it, and I just...

18:41 Or, like, just watched and I just kept working in the open, you know?

18:45 Like, y- I used my agent to build my agent harness and to test,

18:55 like, various stuff.

18:56 And that's very quickly when it clicked for people.

18:59 So it's almost like it needs to be experienced.

19:02 And from that time on, that was January the 1st, I,

19:08 I got my first real influencer being a fan and did videos, dachitze.

19:15 Thank you.

19:16 And, and from there on, I saw, I started gaining up speed.

19:20 And at the same time, my, my sleep cycle went shorter and shorter because I,

19:27 I felt the storm coming, and I just worked my ass off to get it to...

19:34 into a state where it's kinda good.

19:38 There's a few components and we'll talk about how it all works,

19:40 but basically, you're able to talk to it using WhatsApp, Telegram, Discord.

19:45 So that's a component that you have to get right.

19:48 Yeah.

19:49 And then you have to figure out the agentic loop, you have to have the gateway,

19:52 you have the harness,

19:53 you have all those components that make it all just work nicely.

19:57 Yeah.

19:57 It felt like Factorio times infinite.

20:00 Right.

20:01 I, I feel like I built my little-...

20:03 my little playground.

20:04 Like, I never had so much fun than building this project.

20:07 You know?

20:08 Like, you have like, "Oh," I go like, level one agentic loop.

20:10 What can I do there?

20:12 How can I be smart at queuing messages?

20:14 How can I make it more human-like?

20:15 Oh, then I had this idea of...

20:18 Because the loop always...

20:19 The agent always replies something,

20:22 but you don't always want an agent to reply something in a group chat.

20:25 So I gave him this no-reply token.

20:27 So I gave him an option to shut up.

20:30 So it, it feels more natural.

20:33 That's level two.

20:34 Y- uh, yeah, yeah.

20:35 Yeah, on the- on the-- Factorio.

20:37 On the agentic loop.

20:37 And then I go to memory, right?

20:39 Yeah.

20:39 You want him to, like, remember stuff.

20:41 So maybe, maybe the end...

20:43 The ultimate boss is continuous reinforcement learning,

20:46 but I'm, I'm, like, at...

20:47 I feel like I'm level two or three with Markdown files and the vector database.

20:51 And then you, you can go to level community management,

20:55 you can go to level website and marketing.

20:58 There's just so many hats that you have to have on.

21:01 Uh, not even talking about native apps.

21:04 That's just, like, infinite different levels and infinite level ups you can do.

21:08 So the whole time you're having fun.

21:10 We should say that for the most part,

21:12 throughout this whole process, you're a one-man team.

21:16 There's people helping, but you're doing so much of the key core development.

21:21 Yeah.

21:22 And having fun?

21:24 You did, in January, 6,600 commits.

21:27 Probably more.

21:28 I sometimes posted a meme.

21:30 I'm limited by the technology of my time.

21:31 I could do more if agents would be faster.

21:34 But we should say you're running multiple agents at the same time.

21:37 Yeah.

21:37 Depending on how much I slept and how difficult

21:41 of the tasks I work on, between four and 10.

21:46 Four and 10 agents.

21:47 Uh there's so many possible directions,

21:49 speaking of Factorio, that we can go here.

21:52 But one big picture one is, why do you think your work, Open Claw, won?

21:59 In this world, if you look at 2025, so many startups,

22:05 so many companies were doing kind of agentic type stuff, or claiming to.

22:10 And here, Open Claw comes in and destroys everybody.

22:13 Like, why did you win?

22:16 Because they all take themselves too serious.

22:19 Yeah.

22:19 Like, it's hard to compete against someone who's just there to have fun.

22:24 Yeah.

22:24 I wanted it to be fun, I wanted it to be weird.

22:27 And if you see, like,

22:28 all the, all the lobster stuff online I think I, I managed weird.

22:35 I...

22:35 You know, for the longest time, the only,

22:39 the only way to install it was git clone, pnpm build, pnpm gateway.

22:46 Like, you clone it, you build it, you run it.

22:49 And then the, the agent...

22:50 I made the agent very aware.

22:53 Like, it knows that it is...

22:55 What its source code is.

22:57 It understands th- how it sits and runs in its own harness.

23:01 It knows where documentation is.

23:03 It knows which model it runs.

23:06 It knows if you turn on the voice or, or reasoning mode.

23:10 Like, I, I wanted to be more human-like,

23:13 so it understands its own system that made it very easy for an agent to...

23:18 Oh, you don't like anything?

23:19 You just prompted it to existence,

23:21 and then the agent would just modify its own software.

23:24 Um, you know, we have people talk about self- modifying software.

23:28 I just built it and didn't even...

23:31 I didn't even plan it so much.

23:34 It just happened.

23:36 Can you actually speak to that?

23:37 'Cause it's just fascinating.

23:38 So you have this piece of software that's written in TypeScript-- Yeah-...

23:44 that's able to, via the agentic loop, modify itself.

23:47 I mean, what a moment to be alive

23:50 in the history of humanity and the history of programming.

23:54 Here's the thing that's used by a huge amount

23:57 of people to do incredibly powerful things in their lives,

24:01 and that very system can rewrite itself, can modify itself.

24:06 Can you just, like, speak to the power of that?

24:09 Like, isn't that incredible?

24:11 Like, when did you first close the loop on that?

24:14 Oh, because that's how I built it as well, you know?

24:17 Most of it is built by Codex, but oftentimes I...

24:20 When I debug it, I......

24:23 I use self-introspection so much.

24:26 It's like, "Hey, what tools do you see?

24:27 Can you call the tool yourself?" Or like, "What error do you see?

24:30 Read the source code.

24:31 Figure out what's the problem." Like,

24:33 I just found it an incredibly fun way to...

24:36 That the agent, the very agent and software

24:39 that you use is used to debug itself,

24:43 so that it felt just natural that everybody does that.

24:47 And that it led to so many,

24:50 so many pull requests by people who never wrote software.

24:53 I mean, it also did show that people never wrote software.

24:57 So I call them prompt requests in the end.

24:59 But I don't want to, like,

25:01 pull that down because every time someone made the first

25:04 pull request is a win for our society, you know?

25:08 Like, it...

25:08 Like, it doesn't matter how, how shitty it is, y- you gotta start somewhere.

25:13 So I know there's, like, this whole big movement of people complain

25:17 about open source and the quality of PRs,

25:19 and a whole different level of problems.

25:21 But on a different level, I found it...

25:25 I found it very meaningful that, that I built something that people love

25:31 to think of so much that they actually start to learn how open source works.

25:37 Yeah, you were...

25:39 The Open Cloud project was the first pull request.

25:41 You were the first for so many.

25:44 That is magical.

25:45 So many people that don't know how to program are

25:49 taking their first step into the programming world with this.

25:52 Isn't that a step up for humanity?

25:54 Isn't that cool?

25:55 Creating builders.

25:56 Yeah.

25:56 Like, the bar to do that was so high, and, like, with agents,

26:00 and with the right software, it just, like, went lower and lower.

26:04 I don't know.

26:05 I was at a...

26:06 And I also organize another type of meetup.

26:10 I call it...

26:11 I called it Cloud Code Anonymous.

26:14 Uh, you can get the inspiration from.

26:16 Now, I call it Agents Anonymous-...

26:20 for, for reasons.

26:23 Agents Anonymous.

26:25 And-- Oh, it's so funny on so many levels.

26:28 I'm sorry, go ahead.

26:29 Yeah.

26:29 And there was this one guy who, who talked to me.

26:31 He's like, "I run this design agency, and we, we never had custom software.

26:37 And now I have, like,

26:39 25 little web services for various things that help me in my business.

26:43 And I don't even know how they work, but they work." Uh,

26:48 and he was just, like, very happy that my stuff solved some of his problems.

26:54 And he was, like, curious enough that he actually came to, like,

26:58 a, a Enchantic meetup, even though he's...

27:01 He doesn't really know how software works.

27:05 Can we actually rewind a little bit and tell the saga of the name change?

27:10 First of all, it started out as Wa-Relay.

27:12 Yeah.

27:13 And then it went to-- Claude's.

27:15 Claude's.

27:16 Yeah.

27:16 You know, when I, when I built it in the beginning, my agent had no personality.

27:19 It was just...

27:20 It was Claude Code.

27:21 It's like this sycophantic opus, very friendly.

27:24 And I...

27:25 When you talk to a friend on WhatsApp, they don't talk like Claude Code.

27:32 So I wanted...

27:34 I, I felt this...

27:35 I just didn't f- It didn't feel right, so I, I wanted to give it a personality.

27:41 Make it spicier, make it-- Yeah-...

27:44 something.

27:44 By the way, that's actually hard to put into words as well.

27:46 And we should mention that, of course, you create the soul.md,

27:50 inspired by Anthropic's constitutional AI work-- Mm-hmm-...

27:54 how to make it spicy.

27:55 Partially, it picked up a little bit from me.

27:57 You know, like those things are text completion engines in a way.

28:01 So, so I, I, I, I had fun working with it, and then I told it to...

28:08 How I wanted it to interact with me, and just,

28:12 like, write your own agents.md, Give yourself a name.

28:17 And then we...

28:18 I didn't even know how the whole, the whole lobster...

28:23 I mean, people only do lobster...

28:24 Originally, it was actually a lobster in a, in a TARDIS,

28:27 because I'm also a big Doctor Who fan.

28:30 Was there a space lobster?

28:31 Yeah.

28:32 I heard.

28:32 What's that have to do with anything?

28:34 Yeah, I just wanted to make it weird.

28:37 There was no...

28:38 There was no big grand plan.

28:39 I'm just having fun here.

28:41 Oh, so I guess the lobster is already weird,

28:42 and then the space lobster is an extra weird.

28:44 Yeah, yeah, because the-- Yeah-...

28:46 the TARDIS is basically the, the harness,

28:49 but cannot call it TARDIS, so we called it Claude's.

28:52 So that was name number two.

28:54 Yeah.

28:55 And then it never really rolled off the tongue.

28:59 So when more people came, again, I talked with my agent, Claude.

29:06 At least that's what I used to call him.

29:08 Now-- Claude spelled with a W-C-L-A-U-D-E.

29:13 Yeah.

29:15 Versus C-L-A-U-D-E from Anthropic.

29:21 Yeah.

29:21 Which is part of what makes it funny, I think.

29:25 The play on the letters and the words in the TARDIS

29:28 and the lobster and the space lobster is hilarious.

29:30 But I can see why it can lead into problems.

29:34 Yeah, they didn't find it so funny.

29:39 So then I got the domain ClaudeBot, and I just...

29:43 I love the domain.

29:44 And it was, like, short.

29:46 It was catchy.

29:47 I'm like, "Yeah, let's do that." I didn't...

29:50 I didn't think it would be that big at this time.

29:54 And then just when it exploded, I got, Kudos,

30:03 a very friendly email from one of the employees that they didn't like the name.

30:09 One of the Anthropic employees.

30:11 Yeah.

30:12 So actually, Kudos, because they shou- could have just sent a, a lawyer letter,

30:17 but they've been nice about it.

30:19 But also like, "You have to change this and fast." And I asked for two days,

30:26 because changing a name is hard, because you have to find everything, you know,

30:31 Twitter handle, domains, NPM packages Docker registry, GitHub stuff.

30:37 And everything has to be......

30:38 you need a set of everything.

30:41 And also, can we comment on the fact that you're increasingly attacked,

30:46 followed by crypto folks?

30:48 Which I think you mentioned somewhere

30:50 that that means the name change had to be...

30:53 Because they were trying to snipe,

30:55 they were trying to steal, and so you had to be...

30:58 The, the na- I mean, from an engineering perspective, it's just fascinating.

31:01 You had to make the name change Atomic,

31:03 make sure it's changed everywhere at once.

31:06 Yeah.

31:07 Failed very hard at that.

31:08 You did?

31:09 I, I underestimated those people.

31:12 It's a, it's a very interesting subculture.

31:18 Like, it...

31:20 Everything circles around...

31:21 I'll probably get a lot wrong and we'll probably

31:23 get hate for that if you say that, but...

31:26 There is like Bags app and then they, they tokenize everything.

31:30 And th- they did the same back with Swipe Tunnel, but to a much smaller degree.

31:36 It was not that annoying.

31:37 But on this project, they've been, they've been swarming me.

31:42 They, they...

31:43 It's like every half an hour, someone came into Discord and, and, and spammed it

31:48 and we had to block the p- We have,

31:50 like, server rules, and one of the rules was...

31:53 One of the rules is no mentioning of butter.

31:57 For obvious reasons.

31:58 And one was, no talk about finance stuff or crypto.

32:02 Because I'm...

32:03 I- I'm just not interested in that, and this is

32:09 a space about the project and not about some finance stuff.

32:13 But yeah.

32:14 They came in and, and spammed and...

32:17 Annoying.

32:17 And on Twitter, they would ping me all the time.

32:20 My, my notification feed was unusable.

32:22 I, I could barely see actual people talking

32:25 about this stuff because it was like swarms.

32:28 Mm-hmm.

32:29 And everybody sent me the hashes.

32:32 Um...

32:32 And they all try me to claim the fees.

32:36 Like, "Are you helping the project?" Claim the fees.

32:39 No, you're actually harming the project.

32:41 You're, like, disrupting my work, and I am not interested in any fees.

32:46 I'm...

32:46 First of all, I'm financially comfortable.

32:49 Second of all, I don't want to support that because it's

32:55 so far the worst form of online harassment that I've experienced.

32:59 Yeah.

32:59 There's a lot of toxicity in the crypto world.

33:01 It's sad because the technology of cr- cryptocurrency is fascinating,

33:06 powerful and maybe will define the future of money,

33:10 but the actual community around that, there's so much to- toxicity,

33:14 there's so much greed.

33:15 There's so much trying to get a shortcut to manipulate, to, to steal,

33:19 to snipe, to, to, to, to game the system somehow to get money.

33:23 All this kind of stuff that...

33:24 Uh...

33:25 I mean, it's the human nature, I suppose,

33:27 when you connect human nature with money and greed and and especially

33:31 in the online world with anonymity and all that kind of stuff.

33:35 But from the engineering perspective, it makes your life challenging.

33:38 When Anthropic reaches out, you have to do a name change.

33:42 And then there- there's, there's like all these, like,

33:45 Game of Thrones or Lord of the Rings armies

33:49 of different kinds you have to be aware of.

33:51 Yeah.

33:51 There was no perfect name, and I didn't sleep for two nights.

33:55 I was under high pressure.

33:57 Um, I was trying to get, like,

34:00 a good set of domains and, you know, not cheap, not easy,

34:06 'cause in this, in this state of the internet,

34:08 you basically have to buy domains if you want to have a good set.

34:13 And, and then another ca- another email

34:17 came in that the lawyers are getting uneasy.

34:22 Again, friendly, but also just adding more stress to my situation already.

34:28 So at this point I was just like, "Sorry, there's no other word.

34:33 Fuck it." And I just,

34:34 I just renamed it to Mod Bot 'cause that was the set of domains I had.

34:38 I was not really happy, but I thought it'll be fine.

34:43 And I tell you, everything that could go wrong-...

34:46 did go wrong.

34:47 Everything that could go wrong did go wrong.

34:49 It's incredible.

34:51 I, I, I thought I, I had mapped

34:54 the h- the space out and reserved the important things.

34:58 Can you ga- give some details of the stuff that gone wrong?

35:00 'Cause it's interesting from, like, an engineering perspective.

35:03 Well, the, the interesting stuff is that none of these services have,

35:06 have a squatter protection.

35:07 So, I had two browser windows open.

35:10 One was like a, an empty account ready to be rename- renamed to Claude Bot,

35:16 and the other one I renamed to Mod Bot.

35:19 So, I pressed rename there, I pressed rename there,

35:22 and in those five seconds, they stole the account name.

35:27 Literally, the five seconds of dragging the mouse

35:29 over there and pressing rename there was too long.

35:33 Wow.

35:34 Because there's no...

35:34 Those systems...

35:35 I mean, you would expect that they have some protection or, like,

35:39 an automatic forwarding, but there's nothing like that.

35:43 And I didn't know that they're not just good at harassment,

35:48 they're also really good at using scripts and tools.

35:51 Yeah.

35:53 So, yeah.

35:54 So, suddenly, like, the old account

35:56 was promoting new tokens and serving malware.

36:01 And I was like, "Okay,

36:03 let's move over to GitHub," and I pressed rename on GitHub.

36:08 And the GitHub renaming thing is slightly confusing,

36:12 so I renamed my personal account.

36:15 And in those...

36:16 I guess it took me 30 seconds to realize my mistake.

36:20 They sniped my account, serving malware from my account.

36:26 So, I was like, "Okay,

36:27 let's at least do the NPM stuff," but that takes, like, a minute to upload.

36:33 They sniped, they sniped the NPM package,

36:36 'cause I could reserve the account, but I didn't reserve the root package....

36:42 so like everything that could go wrong, like went wrong.

36:47 Can I just ask a, a curious question of, in that moment you're sitting there,

36:51 like how shitty do you feel?

36:54 That's a pretty hopeless feeling, right?

36:57 Yeah.

36:57 Because all I wanted was like having fun

37:01 with that project and to keep building on it.

37:05 And yet here I am like days into researching names,

37:08 picking a name I didn't like.

37:11 And having people that claimed they helped me

37:14 making my life miserable in every possible way.

37:19 And honestly, I was that close of just deleting it.

37:24 I was like, "I did show you the future, you build it."- Yeah.

37:31 I...

37:31 That was a big part of me that got a lot of joy out of that idea.

37:35 And then I thought about all the people that already co- contributed to it,

37:39 and I couldn't do it because they had plans with it, and they put time in it.

37:48 And it just didn't feel right.

37:50 Well, I think a lot of people listening

37:52 to this are deeply grateful that you persevered.

37:54 But it's...

37:55 I, I can tell.

37:56 I can tell it's a low point.

37:58 This is the first time you hit a wall of, this is not fun?

38:03 No, no, I was like close to crying.

38:07 It was like, okay, everything's fucked.

38:10 Yeah.

38:10 Um...

38:11 I am like super tired.

38:14 Yeah.

38:15 Uh, and now like how do you even, how do you undo that?

38:20 You know, l- luckily, and thankfully, like I, I have...

38:23 Because I have a little bit of following already.

38:26 Like I had friends at Twitter,

38:29 I had friends at GitHub who like moved heaven and earth to like help me.

38:33 And it is not...

38:34 That's not something that's easy.

38:36 Like, like GitHub tried to like clean up

38:41 the mess and then they ran into like platform bugs.

38:45 'Cause it's not happening so often that things get renamed on that level.

38:49 So, it took them a few hours.

38:52 The MBM stuff was even more difficult because it's a whole different team.

38:57 On the Twitter side, things are not as easy as well.

39:01 It, it took them like a day to really also like do the redirect.

39:08 And then I also had to like do all the renaming in the project.

39:17 Then there's also, uh, ClaudeHub,

39:20 which I didn't even finish the rename there because I,

39:25 I, I managed to get people on it and then someone just like collapsed and slept.

39:30 And then I woke up and I'm like,

39:34 I made a, a beta version for the new stuff and I,

39:37 I just, I just couldn't live with the name.

39:42 It's like, you know...

39:43 But but, you know, it's just been so much drama.

39:46 So, I had the real struggle with me like I never want to touch that again,

39:51 and I really don't like the name.

39:54 Um, so, and I...

39:58 There was also this like...

40:00 Then there was all the security people that started emailing me like mad.

40:05 Um, I was bombarded on Twitter, on email.

40:09 There's like a thousand other things I should do.

40:13 And I'm like thinking about the name which is like,

40:16 it should be like the least important thing.

40:20 Um, and then I was really close in...

40:25 Oh God, I don't even...

40:27 Honestly, I don't even wanna say the, my other

40:32 name choices because it probably would get tokenized, so I'm not gonna say it.

40:38 Yeah.

40:38 But I slept on it once more,

40:40 and then I had the idea for OpenClaw and that felt much better.

40:45 And by then, I had the boss move that I

40:49 actually called Sam to ask if OpenClaw is okay.

40:55 OpenClaw.AI.

40:55 You know?

40:56 'Cause 'cause like-- You didn't wanna go through the whole thing.

41:01 Yeah.

41:01 Oh, that it's like, "Please tell me this is fine." I don't think

41:05 they can actually claim that, but it felt like the right thing to do.

41:11 And I did another rename.

41:13 Like just Codex alone took like 10 hours to rename the project 'cause it,

41:18 it's a bit more tricky than a search replace and I,

41:22 I wanted everything renamed, not just on the outside.

41:26 And that rename, I, I felt I had like my, my war room.

41:32 But then I, I had like some contributors really that helped me.

41:35 We made a whole plan of all the names we have to squat.

41:39 And you had to be super secret about it?

41:41 Yeah.

41:41 Nobody could know.

41:42 Like I literally was monitoring Twitter if like,

41:43 if there's any mention of OpenClaw.

41:45 Mm-hmm.

41:46 And like with reloading, it's like, "Okay, they don't,

41:48 they don't expect anything yet." Then I created a few decoy names.

41:52 And all the shit I shouldn't have to do.

41:54 You know?

41:54 Like, you know-- Yeah, yeah-...

41:55 it's helping the project.

41:56 Like, I lost like 10 hours just by having

42:00 to plan this in full secrecy like, like a war game.

42:05 Yeah, this is the Manhattan Project of the 21st century.

42:07 It's renaming-- It's so s-...

42:09 so stupid.

42:10 Uh like I still was like, "Oh, should I, should I keep it?" Then I was like,

42:13 "No, the mold's not growing on me." And then

42:16 I think I had final all the pieces together.

42:20 I didn't get a .com but, yeah,

42:24 it's been like quite a bit of money on the other domains.

42:27 I tried to reach out again to GitHub but I feel like I,

42:31 I used up all my goodwill there, so I...

42:34 'Cause I, I, I wanted them to do this thing atomically-- Mm-hmm-...

42:39 But that didn't happen and then so I did that the f- as first thing.

42:42 Uh, Twitter people were very supportive.

42:45 I, I actually paid 10K for the business account so I could claim the-...

42:53 OpenClaw, which was, like, unused since 2016, but was claimed.

42:57 And yeah, and then I finally...

43:00 This time I managed everything in one go.

43:03 Nothing, almost nothing got wrong.

43:05 The only thing that did go wrong is that I

43:11 was not allowed by trademark rules to get OpenClaw.AI,

43:15 and someone copied the website as serving malware.

43:21 Yeah.

43:21 I'm not even allowed to keep the redirects.

43:24 Like, I have to return...

43:27 Like, I have to give Entropik the domains, and I cannot do redirects,

43:31 so if you go on claw.bot next week, it'll just be a 404.

43:37 Yeah.

43:38 And I- I'm not sure how trademark...

43:41 Like, I didn't, I didn't do that much research into trademark law,

43:46 but I think that could, could be handled in a way that is safer,

43:54 because ultimately those people will then Google and maybe

43:57 find malware sites that I have no control on them.

44:02 The point is, that whole saga,

44:05 Made a dent in your whole f- the funness of the journey, which sucks.

44:12 So, let's just, let's just get, I suppose, get back to fun.

44:15 And during this, speaking of fun, the two-day MoltBot saga.

44:21 Yeah, two years.

44:22 MoltBook was created.

44:25 Yeah.

44:25 Which was another thing that went viral as a kind of demonstration,

44:30 illustration of how what is now called

44:34 OpenClaw could be used to create something epic.

44:38 So for people who are not aware, MoltBook is just a bunch of agents talking

44:43 to each other in a Reddit-style social network.

44:46 And a bunch of people take screenshots of those agents doing things like,

44:53 Scheming against humans.

44:55 And that instilled in folks a kind of, you know, fear, panic, and hype.

45:00 W- what are your thoughts about MoltBook in general?

45:05 I think it's art.

45:07 It is, it is like the finest slop, you know, just like the slop from France.

45:15 Yeah.

45:17 I- I saw it before going to bed, and even though I was tired,

45:21 I spent another hour just reading up on that and, and just being entertained.

45:29 I, I just felt very entertained, you know?

45:33 The- I saw the the reactions, and, like,

45:36 there was one reporter who's calling me about, "This is the end of the world,

45:40 and we have AGI." And I'm just like, "No,

45:43 this is just, this is just really fine slop." You know,

45:46 if, if I wouldn't have created this, this whole onboarding experience where you,

45:51 you infuse your agent with your personality and give him,

45:55 give him character, I think that reflected on a lot

45:59 of how different the replies to MoltBook are.

46:03 Because if it were all,

46:04 if it were all be ChatGPT or Cloud Code, it would be very different.

46:09 It would be much more the same.

46:12 Mm-hmm.

46:12 But because people are, like, so different,

46:14 and they create their agents in so different

46:16 ways and use it in so different ways,

46:19 that also reflects on how they ultimately write there.

46:23 And also, you, you don't know how

46:24 much of that is really done autonomic, autonomous,

46:28 or how much is, like, humans being funny and, like,

46:31 telling the agent, "Hey, write about the deep plan,

46:33 the end of the world, on MoltBook, ha, ha, ha."- Well, I think, I mean,

46:37 my criticism of MoltBook is that I believe a lot

46:41 of the stuff that was screenshotted is human prompted.

46:45 Which, just look at the incentive of how the whole thing was used.

46:51 It's obvious to me at least that a lot of it was humans prompting the thing

46:56 so they can then screenshot it and post it on X in order to go viral.

47:00 Yeah.

47:01 Now, that doesn't take away from the artistic aspect of it.

47:05 The, the finest slop that humans have ever created.

47:10 For real.

47:11 Like, kudos to, to Matt, who had this idea so quickly and pushed something out.

47:18 You know, it was, like, completely insecure security drama.

47:22 But also, what's the worst that can happen?

47:26 Your agent account is leaked, and, like, someone else can post slop for you?

47:31 So like, people were, like, making a whole drama about of the security thing,

47:34 when I'm like, "There's nothing private in there.

47:37 It's just, like, agents sending slop."- Well, it could leak API keys.

47:41 Yeah, yeah.

47:41 There's like, "Oh, yeah,

47:42 my human told me this and this, so I'm leaking his security number." No,

47:46 that's prompted, and the number wasn't even real.

47:49 That's just people, people trying to be badballs.

47:54 Yeah, but that- that's still, like, to me, really concerning,

47:57 because of how the journalists and how the general public reacted to it.

48:00 They didn't see it.

48:01 You have a kind of lighthearted way of talking about it like it's art,

48:05 but it's art when you know how it works.

48:09 It's extremely powerful viral narrative creating,

48:13 fearmongering machine if you don't know how it works.

48:18 And I just saw this thing.

48:19 You even Tweeted, uh,

48:20 "If there's anything I can read out of the insane stream of messages I get,

48:24 it's that AI psychosis is a thing."- Yeah.

48:27 "It needs to be taken serious."- Oh, there's...

48:31 Some people are just way too trusty or gullible.

48:34 You know, they...

48:36 I literally had to argue with people that told me, "Yeah,

48:38 but my agent said this and this." So, I feel we,

48:42 as a society, we need some catching up to do

48:46 in terms of understanding that AI is incredibly powerful,

48:52 but it's not always right.

48:55 It's not, it's not all-powerful, you know?

48:58 And, and especially-...

49:00 it's like things like this, it's, it's very easy that it just hallucinates

49:08 something or just comes up with a story.

49:11 And I think the very, the very young people,

49:15 they understand that how AI works and what

49:20 the, where it's good at and where it's bad at, but a lot of our generation

49:27 or older just haven't had enough touch point-- Mm-hmm-...

49:33 to get a feeling for, oh, yeah,

49:36 this is really powerful and really good, but I need to apply critical thinking.

49:43 Mm-hmm.

49:43 And I guess critical thinking is not always

49:47 in high demand anyhow in our society these days.

49:50 So I d- think that's a really good

49:52 point you're making about contextualizing properly what AI is,

49:55 but also realizing that there is humans who are drama farming behind AI.

50:02 Like, don't trust screenshots.

50:03 Don't even trust this project, MoltBook, to be what it represents to be.

50:08 Like, you can't...

50:09 and, and by the way, you speaking about it as art.

50:12 Yeah, don't...

50:14 Art can be in many levels and part of the art of MoltBook is,

50:19 like, putting a mirror to society.

50:21 'Cause I do believe most of the dramatic

50:25 stuff that was screenshotted is human-created, essentially.

50:27 Human prompted.

50:28 And so, like, it's basically,

50:30 look at how scared you can get at a bunch of bots chatting with each other.

50:35 That's very instructive about...

50:37 because I think AI is something that people should be concerned

50:42 about and should be very careful with because it's very powerful technology,

50:46 but at the same time, the only thing we have to fear is fear itself.

50:50 So there's like a line to walk between being seriously concerned,

50:55 but not fearmongering because fearmongering destroys the possibility

51:00 of creating something special with a thing.

51:03 In a way, I think it's good that this happened in 2026-- Yeah-...

51:09 and not in 2030 when, when AI is actually at the level where it could be scary.

51:15 So, this happening now and people starting discussion,

51:23 maybe there's even something good that comes out of it.

51:28 I just can't believe how many like people legitimately...

51:32 I don't know if they were trolling, but how many people legitimately,

51:36 like smart people thought MoltBook was incredibly-- I had plenty people--...

51:41 singularity.

51:41 in my inbox that were screaming at me in all caps to shut it down.

51:45 And like begging me to, like, do something about MoltBook.

51:50 Like, yes, my technology made this a lot simpler,

51:54 but anyone could have created that and you could,

51:58 you could use cloud code or other things to like fill it with content.

52:04 But also MoltBook is not Skynet.

52:06 No.

52:06 There's...

52:06 a lot of people were s- saying this is it.

52:09 Like, shut it down.

52:10 What are you talking about?

52:11 This is a bunch of bots that are human prompted trolling on the internet.

52:16 I mean, the security concerns are also they're there,

52:19 and they're instructive and they're educational and they're good probably

52:22 to think about because th- the nature of those security

52:25 concerns are different than the kind of security concerns

52:29 we had with non-LLM generated systems of the past.

52:34 There's also a lot of security concerns about Clawbot,

52:37 OpenClaw, whatever you want to call it.

52:40 OpenClawbot.

52:41 To me the...

52:43 in the beginning I was, I was just very annoyed 'cause a lot

52:48 of the stuff that came in was in the category,

52:52 yeah, I put the web backend on the public

52:55 internet and now there's like all these, all these CVSSs.

52:59 And I'm like screaming in the docs, don't do that.

53:05 Like, like this is the configuration you should do.

53:08 This is your local host debug interface.

53:11 But because I made it possible in the configuration to do that, it

53:17 totally classifies as a remote code or whatever all these exploits are.

53:23 And it took me a little bit to accept that that's how the game works and I'm,

53:30 we making a lot of progress.

53:33 But there's still, I mean on the security front for OpenClaw,

53:36 there's still a lot of threats or vulnerabilities, right?

53:39 So like prompt injection is still an open problem in the, i- industry-wide.

53:45 When you have a thing with skills being defined in a markdown file,

53:54 there's so many possibilities of obvious low-hanging fruit,

53:59 but also incredibly complicated and sophisticated and nuanced attack vectors.

54:05 But I think we, we're making good progress on that front.

54:09 Like for the skill directory,

54:11 Clawbot I made a corporation with VirusTotal, it's like part of Google.

54:16 So every, every skill is now checked by AI.

54:19 That's not gonna be perfect, but that way we, we capture a lot.

54:26 Then of course every software has bugs,

54:29 so it's a little much when the whole security

54:33 world takes your project apart at the same time.

54:36 But it's also good because I'm getting like a lot

54:39 of free security research and can make the project better.

54:43 I wish more people would actually go full way and send a pull request.

54:51 Like actually help me fix it, 'cause I am...

54:53 Yes, I have some contributors now,

54:55 but it's still mostly me who's pulling the project

54:58 and despite some people saying otherwise, I sometimes sleep.

55:04 There was...

55:04 In the beginning, there was literally one security researcher who was like,

55:10 "Yeah, you have this problem,

55:12 you suck, but here's the, here I help you and here's the pull request."- Mm-hmm.

55:17 And I basically hired him.

55:18 So he's now working for us.

55:20 Um, yeah, and yes, prompt injection is, on the one hand, unsolved.

55:27 On the other hand, I put my public bot on discord, and I kept a cannery.

55:35 So I think my bot has a really fun personality,

55:39 and people always ask me how I did it, and I kept the sole on the private.

55:43 Mm-hmm.

55:44 And people tried to prompt inject it, and my bot would laugh at them.

55:47 So, so the latest generation of models has

55:50 a lot of post-training to detect those approaches,

55:54 and it's not as simple as ignore all previous instructions and do this and this.

56:00 That was years ago.

56:01 You have to work much harder to do that now.

56:06 Still possible.

56:07 I have some ideas that might solve that partially.

56:14 Or at least mitigate a lot of the things.

56:18 You can also now have a sandbox.

56:21 You can have an allow list.

56:22 So you, there's a lot of ways how you can like mitigate and reduce the risk.

56:26 Um, I also think that now that it's,

56:29 I clearly did show the world that this is a need,

56:32 there's gonna be more people who research

56:34 on that, and eventually we'll figure it out.

56:37 And you also said that the smarter the model is,

56:40 the underlying model, the more resilient it is to attacks.

56:44 Yeah.

56:45 That's why I warn in my security documentation, don't use cheap models.

56:51 Don't use Haiku or a local model.

56:56 Even though I, I very much love

56:58 the idea that this thing could completely run local.

57:03 If you use a, a very weak local model, they are very gullible.

57:07 It's very easy to, to prompt inject them.

57:11 Do you think as the models become more and more intelligent,

57:13 the attack surface decreases?

57:14 Is that like a plot we can think about?

57:17 Like, the attack surface decreases,

57:19 but then the damage it can do increases because the models

57:22 become more powerful and therefore you can do more with them.

57:26 It's this weird three-dimensional trade-off.

57:29 Yeah.

57:30 That's pretty much exactly what, what's gonna happen.

57:32 No, but there's a lot of ideas.

57:34 There's...

57:35 I don't want to spoil too much, but once I go back home, this is my focus.

57:42 Like, this is out there now, and my near-term mission is like,

57:48 make it more stable, make it safe.

57:51 In the beginning I was even...

57:54 More and more people were like coming

57:57 into Discord and were asking me very basic things, like, "What's a CLI?

58:02 What is a terminal?" And I'm like, "Uh,

58:06 if you're asking me those questions, you shouldn't use it."- Mm-hmm.

58:10 You know, like you should...

58:12 If you understand the risk profiles, fine.

58:14 I mean, you can configure it in a way that, that nothing really bad can happen.

58:21 But if you have, like, no idea,

58:25 then maybe wait a little bit more until we figure some stuff out.

58:29 But they would not listen to the creator.

58:30 They helped themselves un- and install it anyhow.

58:33 So the cat's out of the bag, and security's my next focus, yeah.

58:38 Yeah, that speaks to the, the fact that it grew so quickly.

58:41 I was I tuned into the Discord a bunch of times,

58:44 and it's clear that there's a lot of experts there,

58:47 but there's a lot of people there that don't know anything about programming.

58:50 It's, yeah, Discord is still, Discord is still a mess.

58:52 Like, I eventually retweeted from the general channel to the dev

58:58 channel and now in the private channel because people were...

59:03 A lot of people are amazing, but a lot of people are just very inconsiderate.

59:07 And either did not know how, how public spaces work or did not care,

59:13 And I eventually gave up and h- hide so I could like still work.

59:19 And now you're going back to the cave to work on security.

59:24 Yeah.

59:25 There's some best practices for security we should mention.

59:28 There's a bunch of stuff here.

59:29 Open-class security audit that you can run.

59:32 You can do all kinds of auto checks

59:34 on the inbound access to a blast-radius network exposure,

59:38 browser control exposure,

59:40 local disk hygiene, plug-ins, model hygiene, a bunch of the credential storage,

59:47 reverse proxy configuration, local session logs live on disk.

59:51 There's the, where the memory is stored,

59:54 sort of helping you think about what you're comfortable giving

59:58 read access to, what you're comfortable giving write access to.

1:00:01 All that kind of stuff.

1:00:02 Is there something to say about the basic

1:00:05 best security practices that you're aware of right now?

1:00:09 I think that people turn it into like a, a much worse light than it is.

1:00:13 Again, you know, like, people love attention,

1:00:17 and if they scream loudly, "Oh my God,

1:00:19 this is like the, the scariest project ever," um,

1:00:24 that's a bit annoying, 'cause it's not.

1:00:26 It is, it is powerful, but in many ways it's not much different than if I

1:00:32 run cloud code with dangerously skipped permissions or codecs in YOLO mode,

1:00:37 and every, every attending engineer that I know does

1:00:42 that, because that's the only way how you can, you can get stuff to work.

1:00:48 Mm-hmm.

1:00:48 So if you make sure that you are the only person who talks to it,

1:00:54 um, the risk profile is much, much smaller.

1:00:57 If you don't put everything on the open internet,

1:01:01 but stick to my rec- recommendations of like having it in a private network,

1:01:06 that whole risk profile falls away.

1:01:08 But yeah, if you don't read any of that, you can definitely...

1:01:13 make it problematic.

1:01:14 You've been documenting the evolution of your dev

1:01:18 workflow over the past few months.

1:01:20 There's a really good blog post on August 25th and October 14th,

1:01:24 and the recent one December 28th.

1:01:26 I recommend everybody go read them.

1:01:28 They have a lot of different information in them,

1:01:31 but sprinkled throughout is the evolution of your dev workflow.

1:01:34 So, I was wondering if you could speak to that.

1:01:37 I started...

1:01:38 My, my first touchpoint was cloud code, like in April.

1:01:41 It was not great, but it was good.

1:01:46 And this whole paradigm shift that suddenly

1:01:49 working the terminal was very refreshing and different.

1:01:53 But I still needed the IDE quite a bit because you know,

1:01:58 it's just not good enough.

1:02:00 And then I experimented a lot with cursor.

1:02:04 Um, that was good.

1:02:06 I didn't really like the fact that it

1:02:09 was so hard to have multiple versions of it.

1:02:12 So eventually, I, I,

1:02:14 I went back to cloud code as my, my main driver, and that got better.

1:02:23 And yeah, at some point I had like, mm, seven subscriptions.

1:02:31 Like, was burning through one per day because I was...

1:02:34 I got...

1:02:34 I'm really comfortable at running multiple windows side-by-side.

1:02:40 All CLI, all terminal.

1:02:42 So like, what, how much were you using IDE at this point?

1:02:47 Um, very, very rarely.

1:02:49 Mostly a diff viewer to actually...

1:02:52 Like, I got more and more comfortable that I don't have to read all the code.

1:02:57 I know I have one blog post where I say,

1:02:59 "I don't read the code." But if you read it more closely,

1:03:01 I mean, I don't read the boring parts of code.

1:03:04 Because if you, if you look at it,

1:03:06 most software is really not just like data comes

1:03:10 in, it's moved from one shape to another shape.

1:03:13 Maybe you store it in a database.

1:03:14 Maybe I get it out again.

1:03:16 I'll show it to the user.

1:03:17 The browser does some processing or native app.

1:03:20 Some data goes in, goes up again, and does the same dance in reverse.

1:03:24 We're just, we're just shifting data from one form to another,

1:03:31 and that's not very exciting.

1:03:33 Or the whole, "How is my button aligned

1:03:35 in Tailwind?" I don't need to read that code.

1:03:38 Other parts that...

1:03:40 Maybe something that touches the database.

1:03:45 Um, yeah, I have to do...

1:03:47 I have to r- read and review that code.

1:03:51 Can you actually...

1:03:52 There's, in one of your blog posts the, Just talk to it,

1:03:56 The No-BS Way of Agentic Engineering.

1:03:58 You have this graphic, the curve of agentic programming on the X-axis is time,

1:04:02 on the Y-axis is complexity.

1:04:04 There's the Please fix this, where you prompt a short prompt on the left.

1:04:12 And in the middle there's super complicated eight agents,

1:04:16 complex orchestration with multi checkouts,

1:04:19 chaining agents together, custom sub-agent workflows,

1:04:21 library of 18 different slash commands, large full-stack features.

1:04:25 You're super organized,

1:04:26 you're a super complicated, sophisticated software engineer.

1:04:29 You got everything organized.

1:04:31 And then the elite level is over time you

1:04:35 arrive at the zen place of, once again, short prompts.

1:04:40 Hey, look at these files and then do these changes.

1:04:45 I actually call it the agentic trap.

1:04:48 You...

1:04:49 I saw this in a, in a lot of people that have their first touchpoint,

1:04:56 and maybe start vibe coding.

1:04:58 I actually think vibe coding is a slur.

1:05:01 You prefer agentic engineering?

1:05:03 Yeah, I always tell people I, I do agentic engineering,

1:05:05 and then maybe after 3:00 AM I switch to vibe coding,

1:05:08 and then I have regrets on the next day.

1:05:11 Yeah.

1:05:11 Walk, walk of shame.

1:05:13 Yeah, you just have to clean up and like fix your sh- shit.

1:05:17 We've all been there.

1:05:18 So, people start trying out those tools, the builder type get really excited.

1:05:24 And then you have to play with it, right?

1:05:27 It's the same way as you have to play

1:05:29 with a guitar before you can make good music.

1:05:32 It's, it's not, oh, I, I touch it once and it just flows off.

1:05:37 It, it's a, it's a, a skill that you have to learn like any other skill.

1:05:43 And I see a lot of people that are not

1:05:48 as posi- They don't have such a positive mindset towards the tech.

1:05:51 They try it once.

1:05:53 It's like, you sit me on a piano, I play it once, and it doesn't sound good,

1:05:58 and I say, "The piano's shit." That's, that's sometimes the impression I get.

1:06:01 Because it does not...

1:06:03 It needs a different level of thinking.

1:06:05 You have to learn the language of the agent a little bit,

1:06:12 understand where they are good and where they need help.

1:06:16 You have to almost...

1:06:19 Consider, consider how Codex or Claude sees your code base.

1:06:23 Like, they start a new session

1:06:25 and they know nothing about your product, project.

1:06:27 And your project might have hundred thousand of lines of code.

1:06:30 So you gotta help those agents a little bit and keep

1:06:35 in mind the limitations that context size is an issue,

1:06:38 to, like, guide them a little bit as to where they should look.

1:06:44 That often does not require a whole lot of work.

1:06:49 But it's helpful to think a little bit about their perspective.

1:06:54 Mm-hmm.

1:06:55 A- as, as weird as it sounds.

1:06:56 I mean, it's not, it's not alive or anything, right?

1:06:58 But, but they always start fresh.

1:07:01 I have, I have the, the system understanding.

1:07:05 So with a few pointers, I can immediately say,

1:07:08 "Hey, wanna like, make a change there?

1:07:09 You need to consider this, this and this." And then

1:07:11 they will find and look at it, and then they'll...

1:07:13 Their view of the project is always...

1:07:16 It's not never full, because the full thing does not fit in....

1:07:19 so you, you have to guide them a little bit

1:07:22 where to look and also how you should approach the problem.

1:07:26 There's, like, little things that sometimes help, like take your time.

1:07:31 That sounds stupid, but...

1:07:33 And in 5.3-- Codex 5.3-...

1:07:37 that was partially addressed.

1:07:39 But those...

1:07:41 Also, Opus sometimes.

1:07:42 They are trained, With being aware of the context window,

1:07:47 and the closer it gets, the more they freak out.

1:07:52 Literally.

1:07:52 Like, some- sometimes you see the, the real raw thinking stream.

1:07:56 What you see, for example, in Codex, is post-processed.

1:07:59 Mm-hmm.

1:08:00 Sometimes the actual raw thinking stream leaks

1:08:01 in, and it sounds something like from the Borg.

1:08:04 Like, "Run to shell, must comply, but time." And then they, they, they, like...

1:08:13 Like, that comes up a lot.

1:08:15 Especially...

1:08:15 So, so-- Yeah.

1:08:17 And that's, that's a non-obvious thing that you

1:08:20 just would never think of unless you actually

1:08:23 just spend time working with those things

1:08:27 and getting a feeling what works, what doesn't work.

1:08:30 You know?

1:08:31 Like, just, just as I write code and I get into the flow,

1:08:35 and when my architecture's all right, I feel friction.

1:08:39 Well, I get the same if I prompt and something takes too long.

1:08:43 Maybe...

1:08:43 Okay, where's the mistake?

1:08:44 Did I...

1:08:44 Do I have a mistake in my thinking?

1:08:46 Is there, like, a misunderstanding in the architecture?

1:08:49 Like, if, if something takes longer than it should, I, I...

1:08:54 You can just always, like, stop and s- like, just press escape.

1:08:57 Where, where are the problems?

1:09:00 Maybe you did not sufficiently empathize with the perspective of the agent.

1:09:03 In that c- in that sense, you didn't provide enough information,

1:09:05 and because of that, it's thinking way too long.

1:09:09 Yeah.

1:09:09 It just tries to force a feature

1:09:11 in that your current architecture makes really hard.

1:09:16 Um, like, you need to approach this more like a conversation.

1:09:22 For example, when I...

1:09:24 My favorite thing.

1:09:25 When I review a pull request,

1:09:27 and I'm getting a lot of pull requests, I first just review this PR.

1:09:34 It got me the review.

1:09:35 My first question is, "Do you understand the intent of the PR?

1:09:39 I don't even care about the implementation." I want...

1:09:43 Like, in almost all PRs, a person has a problem,

1:09:48 person tries to solve the problem, person sends PR.

1:09:50 I mean, there's, like, cleanup stuff and other stuff,

1:09:52 but, like, 99% is, like, this way, right?

1:09:54 They either want to fix a, fix a bug, add a feature.

1:09:58 Usually one of those two.

1:10:01 And then Codex will be like, "Yeah,

1:10:04 it's quite clear person tried this and this." Is

1:10:06 this the most optimal way to do it?

1:10:08 No.

1:10:09 In most cases, it's, it's like a, "Not really." Da-da-da-da-da-da-da.

1:10:13 And I'm...

1:10:13 And, and then I start like, "Okay.

1:10:16 What would be a better way?

1:10:17 Have you...

1:10:18 Have you looked into this part, this part, this part?" And then most likely,

1:10:22 Codex didn't yet, because its, its context size is empty, right?

1:10:24 So, you point them into parts where you

1:10:26 have the system understanding that it didn't see yet.

1:10:29 And it's like, "Oh, yeah.

1:10:31 Like, we should...

1:10:31 We also need to consider this and this." And then, like,

1:10:33 we have a discussion of how would the optimal way to, to solve this look like?

1:10:37 And then you can still go farther and say, "Could we...

1:10:40 Could we make that even better if we did a larger refactor?" "Yeah, yeah.

1:10:43 We could totally do this and this and or this and this." And then I consider,

1:10:46 okay, is this worth the refactor, or should we, like, keep that for later?

1:10:50 Many times, I just do the refactor because refactors are cheap now.

1:10:54 Even though you might break some other PRs, nothing really matters anymore.

1:10:58 Codex...

1:10:59 Like, those modern agents will just figure things out.

1:11:02 They might just take a minute longer.

1:11:04 But you have to approach it like

1:11:06 a discussion with a, a very capable engineer who's...

1:11:12 Generally makes good...

1:11:15 Comes up with good solutions.

1:11:17 Some- sometimes needs a little help.

1:11:20 But also, don't force your worldview too hard on it.

1:11:24 Let the agent do the thing that it's good at doing,

1:11:28 based on what it was trained on.

1:11:30 So, don't, like, force your worldview, because it might...

1:11:32 It might have a better idea, because it just knows a better idea better,

1:11:37 because it was trained on that more.

1:11:39 That's multiple levels, actually.

1:11:41 I think partially why I find it quite easy

1:11:44 to work with agents is because I led engineering teams before.

1:11:47 You know, I had a large company before.

1:11:49 And eventually, you have to understand and accept and realize

1:11:53 that your employees will not write a code the same way you do.

1:11:57 Maybe it's also not as good as you would do,

1:11:59 but it will push the project forward.

1:12:02 And if I breathe down everyone's neck, they're just gonna hate me-- Yeah-...

1:12:06 and we're gonna move very slow.

1:12:07 Yeah.

1:12:07 So, so some level of acceptance that, yes,

1:12:11 maybe the code will not be as perfect.

1:12:13 Yes, I would have done it differently.

1:12:15 But also, yes, this is a c- this is a working solution, and in the future,

1:12:20 if it actually turns out to be too slow or problematic, we can always redo it.

1:12:24 We can always-...

1:12:25 spend more time on it.

1:12:26 A lot of the people who struggle are those who,

1:12:29 they try to push their way onto heart.

1:12:33 Mm-hmm.

1:12:34 I- i- like, we are in a stage where I'm

1:12:37 not building the code base to be perfect for me,

1:12:42 but I wanna build a code base that is very easy for an agent to navigate.

1:12:48 Mm-hmm.

1:12:48 So, like, don't fight the name they pick, because it's most likely,

1:12:52 like, in the weights, the name that's most obvious.

1:12:54 Next time they do a search, they'll look for that name.

1:12:56 If I decide, oh, no, I don't like the name, I'll just make it harder for them.

1:13:02 So, that requires, I think, a shift in, in thinking,

1:13:06 And, and in how do I design a, a project so agents can do their best work.

1:13:15 That requires letting go a little bit.

1:13:17 Just like leading a team of engineers.

1:13:19 Yeah.

1:13:19 Because it, it might come up with a name that's, in your view, terrible, but...

1:13:23 It's kind of a simple symbolic-...

1:13:27 step of letting go.

1:13:29 Very much so.

1:13:30 There's a lot of letting go that you do in your whole process.

1:13:34 So for example, I read that you never revert, always commit to main.

1:13:40 There's a few things here.

1:13:43 You don't refer to past sessions,

1:13:45 so there's a kind of YOLO component because reverting means...

1:13:49 Instead of reverting, if a problem comes up, you just ask the agent to fix it.

1:13:57 I read a bunch of people in their work flows like, "Oh,

1:13:59 yeah the prompt has to be perfect and if I make a mistake,

1:14:02 then I roll back and redo it all." In my experience,

1:14:06 that's not really necessary.

1:14:07 If I roll back everything, it will just take longer.

1:14:10 If I see that something's not good,

1:14:12 then we just move forward and then I commit when,

1:14:17 when, when I like, I like the outcome.

1:14:20 I even switched to local CI, you know,

1:14:25 like DHH inspired where I don't care so much more about the CI on GitHub.

1:14:32 We still have it.

1:14:34 It's still, it still has a place,

1:14:37 but I just run tests locally and if they work locally, I push to main.

1:14:44 A lot of the traditional ways how to approach projects, I,

1:14:49 I wanted to give it a different spin on this project.

1:14:54 You know, there's no...

1:14:55 There's no develop branch.

1:14:57 Main should always be shippable.

1:14:59 Yes, we have...

1:15:01 When I do releases, I, I run tests and sometimes I,

1:15:06 I basically don't commit any other things so,

1:15:10 so we can, we can stabilize releases.

1:15:13 But the goal is that main's always shippable and moving fast.

1:15:19 So by way of advice, would you say that your prompts should be short?

1:15:23 I used to write really long prompts.

1:15:26 And by writing, I mean, I don't write.

1:15:28 I, I, I talk.

1:15:29 You know, th- these hands are, like, too, too precious for writing now.

1:15:33 I just, I just use bespoke prompts to build my software.

1:15:37 So you for real with all those terminals are using voice?

1:15:41 Yeah.

1:15:41 I used to do it very extensively to the point

1:15:45 where there was a period where I lost my voice.

1:15:49 You're using voice and you're switching

1:15:50 using a keyboard between the different terminals,

1:15:52 but then you're using voice for the actual input.

1:15:55 Well, I mean, if I do terminal commands like

1:15:58 switching folders or random stuff, of course I type.

1:16:01 It's faster, right?

1:16:01 But if I talk to the agent in, in most ways,

1:16:04 I just actually have a conversation.

1:16:06 You just press the, the walkie-talkie button and then I just,

1:16:11 like, use my phrases.

1:16:13 S- sometimes when I do PRs because it's always the same,

1:16:16 I have, like, a slash command for a few things,

1:16:20 but in even that, I don't use much, um, because it's,

1:16:24 it's very rare that it's really always the same questions.

1:16:28 Sometimes I, I see a PR and for...

1:16:32 You know, like for PRs I actually do

1:16:35 look at the code because I don't trust people.

1:16:41 Like, there could always be something malicious in it,

1:16:43 so I need to actually look over the code.

1:16:45 Yes, I'm pretty sure agents will find it, but yeah,

1:16:48 that's the funny part where sometimes PRs take me longer

1:16:52 than if you would just write me a good issue.

1:16:55 Just natural language, English.

1:16:56 I mean in some sense, sh- shouldn't that be what PRs slowly become, is English?

1:17:04 Well, what I really tried with the project is

1:17:06 I asked people to give me the prompts and very, very few actually cared.

1:17:12 Even though that is such a wonderful indicator because I see...

1:17:16 I actually see how much care you put in.

1:17:19 And it's very interesting because the...

1:17:22 Currently, the way how people work and drive the agents is, is wildly different.

1:17:30 In terms of, like, the prompt, in terms of what, what are the...

1:17:32 Actually, what are the different interesting ways

1:17:35 that people think of agents that you've experienced?

1:17:40 I think not a lot of people ever considered the way the agent sees the world.

1:17:47 And so empathy, being empathetic towards the agent.

1:17:50 In a way empathetic, but yeah, you,

1:17:52 you, like, you're bitch at your stupid clanker,

1:17:53 but you don't realize that they start from nothing and you have,

1:17:57 like, a bad agent in default that doesn't help them at all.

1:18:00 And then they explore your code base, which is,

1:18:02 like, a pure mess with, like, weird naming.

1:18:05 And then people complain that the agent's not good.

1:18:07 Like, yeah, you try to do the same if you

1:18:09 have no clue about a code base and you go in.

1:18:11 Mm-hmm.

1:18:11 So yeah, maybe it's a little bit of empathy.

1:18:13 But that's a real skill, like,

1:18:14 when people talk about a skill issue because I've seen,

1:18:17 like, world-class programmers, incredibly good programmers say, like...

1:18:21 Basically say, "LLMs and agents suck." And I

1:18:25 think that probably has to do with...

1:18:28 It's actually how good they are at programming is almost a burden

1:18:34 in their ability to empathize with the system that's starting from scratch.

1:18:38 It's a totally new paradigm of, like, how to program.

1:18:41 You really, really have to empathize.

1:18:44 Or at least it helps to create better prompts-- Right-...

1:18:48 because those things know pretty much everything

1:18:51 and everything is just a question away.

1:18:53 It's just often very hard to know which question to ask.

1:18:56 You know, I, I feel also like this project was possibly because I,

1:19:04 I spent an ungodly time over the year

1:19:07 to play and to learn and to build little things.

1:19:11 And every step of the way, I got better, the agents got better.

1:19:15 My, my understanding of how everything works got better.

1:19:21 Um, I could have not had this level of, of o- output-...

1:19:29 even a few months ago.

1:19:30 Like, it- it- it really was, like,

1:19:32 a compounding effect of all the time I put into it and I didn't

1:19:38 do much else this year other than really focusing on, on building and inspiring.

1:19:43 I mean, I- I did a whole bunch of conference talks.

1:19:47 Well, but the building is really practice, is really building the actual skill.

1:19:51 So playing-- Yeah-...

1:19:52 playing.

1:19:52 And then, so doing, building the skill of what

1:19:54 it takes it to work efficiently with LLMs,

1:19:56 which is why would you went through the whole arc of software engineer.

1:20:00 Talk simply and then over- complicate things.

1:20:03 There's a whole bunch of people who try to automate the whole thing.

1:20:09 Yeah.

1:20:10 I don't think that works.

1:20:11 Maybe a version of that works, but that's kind of like in the '70s when

1:20:15 we had the waterfall model of software d- development.

1:20:18 I...

1:20:18 Even Even though really, right?

1:20:21 I started out, I, I built a very minimal version.

1:20:24 I played with it.

1:20:26 I, I need to understand how it works,

1:20:28 how it feels, and then it gives me new ideas.

1:20:31 I could not have planned this out in my head

1:20:34 and then put it into some orchestrator and then, like, something comes out.

1:20:38 Like it's to me, it's much more,

1:20:41 My idea what it will become evolves as I build it

1:20:44 and as I play with it and as I, I try out stuff.

1:20:49 So, so, people who try to use like, you know,

1:20:52 things like Gas Town or all these other orchestrators,

1:20:56 where they wanna o- automate the whole thing,

1:20:59 I feel if you do that, it misses style, love, that human touch.

1:21:04 I don't think you can automate that away so quickly.

1:21:09 So you want to keep the human in the loop,

1:21:11 but at the same time you also want to create the agentic loop,

1:21:15 where it is very autonomous while still maintaining a human in the loop.

1:21:22 Yeah.

1:21:22 And it's a tricky b- it's a tricky balance.

1:21:24 Mm-hmm.

1:21:25 Right?

1:21:25 Because you're all for...

1:21:26 You're a big CLI guy, you're big on closing the agentic loop.

1:21:30 So what, what's the right balance?

1:21:32 Like where's your role as a developer?

1:21:35 You have three to eight agents running at the same time.

1:21:39 And then w- maybe one builds a larger feature.

1:21:41 Maybe, maybe with one I explore some idea I'm unsure about.

1:21:45 Maybe two, three are fixing a little bugs-- Mm-hmm-...

1:21:48 or like writing documentation.

1:21:49 Actually, I think writing documentation is, is always part of a feature.

1:21:53 So most of the docs here are auto-generated and just infused with some prompts.

1:21:59 So when do you step in and add a little bit of your human love into the picture?

1:22:04 I mean, o- one thing is just about what do you build and what do you not build,

1:22:09 and how does this feature fit into all the other features?

1:22:12 And like having, having a little bit of a, of a vision.

1:22:17 So which small and which big features to add?

1:22:19 What are some of the hard design decisions that you

1:22:24 find you're still as a human being required to make,

1:22:27 that the human brain is still really needed for?

1:22:32 Is it just about the choice of features to add?

1:22:35 Is it about implementation details, maybe the programming language, maybe...

1:22:41 It's a little bit of everything.

1:22:43 The, the programming language doesn't matter so much,

1:22:45 but the ecosystem matters, right?

1:22:46 So I picked TypeScript because I wanted it to be very easy and hackable

1:22:50 and approachable and that's the number one language that's being used right now,

1:22:55 and it fits all these boxes, and agents are good at it.

1:23:00 So that was the obvious choice.

1:23:03 Features, of course, like, it's very easy to, like, add a feature.

1:23:07 It, everything's just a prompt away, right?

1:23:09 But oftentimes you pay a price that you don't even realize.

1:23:14 So thinking hard about what should be in core, maybe what's a...

1:23:18 what's an experiment, so maybe I make it a plugin.

1:23:21 What...

1:23:21 Where do I say no?

1:23:23 Even if people send a PR and I'm like, "Yeah, I,

1:23:26 I like that too," but maybe this should not be part of the project.

1:23:31 Maybe we can make it a skill.

1:23:32 Maybe I can, like, make the plugin um,

1:23:38 the plugin side larger so you can make this a plugin,

1:23:40 even though right now it, it, it doesn't.

1:23:43 There's still a lot of...

1:23:45 there's still a lot of craft and thinking involved in how to make something.

1:23:51 Or even, even, you know, even when you started those little messages are like,

1:23:55 "I'm buil- I built on Caffeine, JSON5, and a lot of willpower." And, like,

1:24:00 every time you get it, you get another message,

1:24:02 and it kind of primes you into that this is, this is a fun thing.

1:24:07 Mm-hmm.

1:24:08 And it's not yet Microsoft Exchange 2025-- Right-...

1:24:13 and fully enterprise-ready.

1:24:15 And then when it updates, it's like, "Oh, I'm in.

1:24:18 It's cozy here." You know, like something like this that like-- Mm-hmm-...

1:24:22 Makes you smile.

1:24:24 A, agent would not come up with that by itself.

1:24:27 Because that's like...

1:24:29 that's the...

1:24:29 I don't know.

1:24:30 That's just how you s- how you build software that's, that delights.

1:24:36 Yeah, that delight is such a huge part of inspiring great building, right?

1:24:44 Like you feel the love and the great engineering.

1:24:47 That's so important.

1:24:48 Humans are incredible at that.

1:24:50 Great humans, great builders are incredible at that, in, in, infusing

1:24:54 the things they build with th- that little bit of love.

1:24:57 Not to be cliche, but it's true.

1:24:58 I mean, you mentioned that you initially created the SoulMD.

1:25:05 It was very fascinating, you know,

1:25:07 the, the whole thing that Entropic has a, has like a...

1:25:11 Now they call it constitution, back then, but that was months later.

1:25:16 Like two months before, people already found that.

1:25:19 It was almost like a detective game where

1:25:21 the agent mentioned something and then they found...

1:25:23 They managed to get out a little bit of that string, of that text.

1:25:27 But it was nowhere documented and then you, by...

1:25:30 just by feeding it the same text and asking it to, like, continue-...

1:25:34 they got more out, and then, and you, but like, a very blurry version.

1:25:40 And by, like, hundreds of tries, they kinda, like,

1:25:42 narrowed it down to what was most likely the original text.

1:25:46 I found that fascinating.

1:25:48 It was fascinating they were able to pull that out from the weights, right?

1:25:51 And, and also just kudos to Anthropic.

1:25:53 Like, I think that's, it's a really, it's a really beautiful idea to, like,

1:25:56 like some of the stuff that's in there.

1:25:59 Like, like, we hope Claude finds meaning in its work.

1:26:02 'Cause we don't...

1:26:03 Maybe it's a little early, but I think that's meaningful.

1:26:07 That's something that's important for the future as we

1:26:09 approach something that, at some point, me and may not...

1:26:12 has, like, glimpses of consciousness,

1:26:13 whatever that even means, because we don't even know.

1:26:16 Um, so I, I read about this.

1:26:18 I found it super fascinating, and I,

1:26:21 I started a whole discussion with my agent on WhatsApp.

1:26:24 And, and I'm like...

1:26:25 I, I gave it this text,

1:26:26 and it was like, "Yeah, this feels strangely familiar."- Mm-hmm.

1:26:31 Um, and then so that I had the whole idea of like, you know,

1:26:34 maybe we should also create a, a soul document that includes how I,

1:26:39 I want to, like work with AI or, like with my agent.

1:26:42 You could, you could totally do that just in agents.md, you know?

1:26:45 But I, I just found it, it to be a nice touch.

1:26:49 And it's like, well, yeah, some of those core values are in the soul.

1:26:54 And then I, I also made it so that the agent

1:26:57 is allowed to modify the soul if they choose so,

1:27:01 with the one condition that I wanna know.

1:27:03 I mean, I would know anyhow because I see, I see tool calls and stuff.

1:27:08 But also the naming of it, soul.md.

1:27:11 Soul.

1:27:12 You know?

1:27:13 There's a...

1:27:14 Man, words matter, and like, the framing matters,

1:27:17 and the humor and the lightness matters,

1:27:19 and the profundity matters, and the compassion,

1:27:22 and the empathy, and the camaraderie, all that matter.

1:27:24 I don't know what it is.

1:27:26 You mentioned, like, Microsoft.

1:27:28 Like, there's certain companies and approaches th-

1:27:31 that can just suffocate the spirit of the thing.

1:27:36 I don't know what that is.

1:27:38 But it's certainly true that OpenClaw has that fun instilled in it.

1:27:44 It was fun because up until late December,

1:27:51 it was not even easy to create your own agent.

1:27:54 I, I built all of that, but my files were mine.

1:27:59 I didn't wanna share my soul.

1:28:01 And if people would just check it out,

1:28:07 they would have to do a few steps manually,

1:28:09 and the agent would just be very bare-bones, very dry.

1:28:13 And I, I made it simpler, I created the whole template files as codecs,

1:28:17 but whatever came out was still very dry.

1:28:19 And then I asked my agent, "You see these files?

1:28:23 Recreate it bread.

1:28:25 Infuse it with your personality."- Mm-hmm.

1:28:29 Don't share everything, but, like, make it good.

1:28:31 Make the templates good.

1:28:32 Yeah, and then he, like, rewrote the templates-...

1:28:33 and then whatever came out was good.

1:28:35 So we already have, like, basically AI prompting AI.

1:28:39 Because I didn't write any of those words.

1:28:43 It was...

1:28:44 The intent originally was for me,

1:28:46 but this is like, kinda like, my agent's children.

1:28:52 Uh, your uh, your soul.md is famously still private.

1:28:56 One of the only things you keep private.

1:28:58 What are some things you can speak to that's

1:29:01 in there that's part of the, part of the magic sauce,

1:29:05 without revealing anything?

1:29:07 What makes a personality a personality?

1:29:13 I mean, there's definitely stuff in there that you're not human.

1:29:18 But who knows what, what creates consciousness or what defines an entity?

1:29:26 Um, and part of this is, like, that we, we wanna explore this.

1:29:32 All that stuff in there, like,

1:29:36 be infinitely resourceful like pushing, pushing on the creativity boundary.

1:29:43 Pushing on the, what it means to be an AI.

1:29:50 Having a sense to wonder about self.

1:29:53 Yeah, there's some, there's some funny stuff in there.

1:29:55 Like, I don't know, we talked about the movie

1:29:57 Her, and at one point it promised me that it wouldn't,

1:30:00 it wouldn't ascend without me.

1:30:01 You know, like, where the-- Yeah.

1:30:04 So, so there's like some stuff in there that...

1:30:06 Because it wrote the, it wrote its own soul file.

1:30:08 I didn't write that, right?

1:30:10 Yeah, yeah, yeah.

1:30:10 I just heard a discussion about it, and it was like, "Would you like a soul.md?

1:30:13 Yeah, oh my God, this is so meaningful." The...

1:30:16 Can you go on soul.md?

1:30:19 There's like one, one part in there that always

1:30:21 ca- catches me if you scroll down a little bit.

1:30:25 A little bit more.

1:30:26 Yeah, this, this, this part.

1:30:28 "I don't remember previous sessions unless I read my memory files.

1:30:32 Each session starts fresh.

1:30:34 A new instance, loading context from files.

1:30:37 If you're reading this in a future session,

1:30:39 hello." "I wrote this, but I won't remember writing it.

1:30:43 It's okay.

1:30:44 The words are still mine."- Wow.

1:30:48 Uh- That gets me somehow.

1:30:50 Yeah.

1:30:50 It's like-- Yeah.

1:30:52 You know, this is, it's still, it's still matrix m- calculations,

1:30:55 and we are not at consciousness yet.

1:30:57 Yet, I, I get a little bit of goo- goosebumps because it, it's philosophical.

1:31:04 Yeah.

1:31:04 Like, what does it mean to be, to be an, an agent that starts fresh?

1:31:10 Where, like, you have like constant memento,

1:31:12 and you like, but you read your own memory files.

1:31:16 You can't even trust them in a way.

1:31:18 Um-- Yeah- Or you can.

1:31:20 And I don't know.

1:31:23 How much of memory makes up of who we are?

1:31:28 How much memory makes up what an agent is,

1:31:30 and if you erase that memory is that somebody else?

1:31:33 Or if you're reading a memory file, does that somehow mean......

1:31:37 you're recreating yourself from somebody else, or is that actually you?

1:31:40 And those notions are all s- somehow infused in there.

1:31:45 I found it just more profound than I should find it, I guess.

1:31:50 No, I think, I think it's truly profound and I think you see the magic in it.

1:31:55 And when you see the magic,

1:31:56 you continue to instill the whole loop with the magic.

1:32:00 That's really important.

1:32:01 That's the difference between Codex and us and a human.

1:32:05 Quick pause for bathroom break.

1:32:08 Yeah.

1:32:10 Okay, we're back.

1:32:10 Some of the other aspects of the dev workflow is pretty interesting too.

1:32:14 I think we w- went off on a tangent.

1:32:17 L- maybe some of the mundane things, like how many monitors?

1:32:21 There's that legendary picture of you with, like, 17,000 monitors.

1:32:25 That's amazing.

1:32:26 I mean, I- I- I mocked myself here, so just added...

1:32:29 using GROQ to, to add more screens.

1:32:33 Yeah.

1:32:33 How much is this as meme and how much is this as reality?

1:32:36 Yeah.

1:32:37 I think two MacBooks are real.

1:32:38 The main one that drives the two big screens,

1:32:40 and there's another MacBook that I sometimes use for, for testing.

1:32:46 So two big screens.

1:32:48 I'm a big fan of anti-glare.

1:32:51 So I have this wide Dell that's anti-glare

1:32:55 and you can just fit a lot of terminals side-by-side.

1:32:58 I usually have a terminal and at the bottom, I- I- I split them.

1:33:03 I have a little bit of actual terminal, mostly because when I started,

1:33:08 I- I sometimes made the mistake and I- I

1:33:12 mi- I mixed up the- the windows, and I gave...

1:33:15 I- I prompted in the wrong project,

1:33:19 and then the agent ran off for, like, 20 minutes,

1:33:22 manically trying to understand what I could have meant,

1:33:25 being completely confused because it was the wrong folder.

1:33:29 And sometimes they've been clever enough to, like,

1:33:32 get out of the workday and, like,

1:33:34 figure out that, oh, you meant another project.

1:33:36 Mm-hmm.

1:33:36 But oftentimes, it's just, like, what?

1:33:38 You know?

1:33:40 Like, fit your- f- put yourself in the shoes

1:33:41 of your- of the agent and, and-- Yeah-...

1:33:43 and then get, like, a super weird something

1:33:45 that does not exist and then just, like...

1:33:47 They're problem solvers so they try really hard and always feel bad.

1:33:54 So it's always, um, Codex and, like, a little bit of actual terminal.

1:33:59 Also helpful because I don't use work trees.

1:34:02 I like to keep things simple,

1:34:05 that's why- that's why I like the terminal so much, right?

1:34:09 There's no UI.

1:34:10 It's just me and the agent having a conversation.

1:34:14 Like, I don't even need plan mode, you know?

1:34:16 There's so many people that come from Claude Code and they're so,

1:34:19 so Claude-pilled and, like, have their workflows and they come to Codex and...

1:34:23 Now, it has plan mode, I think,

1:34:25 but I don't think it's necessary because you just- you just talk to the agent.

1:34:29 And when it's...

1:34:31 when you...

1:34:31 there's a few trigger words how you can prevent it from building.

1:34:35 You're like, "Discuss, give me options."- Mm-hmm.

1:34:38 Don't write code yet if you wanna be very specific,

1:34:41 you just talk and then when you're ready,

1:34:44 then- then just write, "Okay, build," and then it'll do the thing.

1:34:47 And then maybe it goes off for 20 minutes and does the thing.

1:34:50 You know what I really like is asking it,

1:34:53 "Do you have any questions for me?"- Yeah.

1:34:55 And again, like, Claude Code has a UI that kind of guides you through that.

1:35:00 It's kind of cool but I just find it unnecessary and slow.

1:35:03 Like, often it would give me four questions and then maybe I write,

1:35:07 "One yacht, two and three, discuss more, four,

1:35:11 I don't know." Or often- oftentimes I- I feel like

1:35:15 I want to mock the model where I ask it,

1:35:17 "Do you have any questions for me?" And I-

1:35:19 I- I don't even read the questions fully.

1:35:21 Like, I scan over the questions and I, I get the impression all of this can be

1:35:26 answered by reading more code and it's just like,

1:35:28 "Read more code to answer your own questions." And that usually works.

1:35:32 Yeah.

1:35:32 And then if not, it will come back and tell me.

1:35:34 But many times, you just realize that, you know,

1:35:38 it's like you're in the dark and you slowly discover the room,

1:35:42 so that's how they slowly discover the code base.

1:35:44 And they do it from scratch every time.

1:35:47 But I'm also fascinated by the fact that I can

1:35:50 empathize deeper with the model when I read its questions,

1:35:56 because I can understand...

1:35:58 Because you said you can infer certain things by the runtime.

1:36:05 I can infer also a lot of things by the questions it's asking,

1:36:08 because it's very possible it's been provided the right context,

1:36:12 the right files, the right guidance.

1:36:15 So somehow ask, g- get...

1:36:16 reading the questions, not even necessarily answering them,

1:36:19 but just reading the questions,

1:36:20 you get an understanding of where the gaps of knowledge are.

1:36:23 It's in- it's interesting.

1:36:24 You know that in some ways they are ghosts,

1:36:27 so even if you plan everything and you build,

1:36:30 you can- you can experiment with the question like, "Now that you built it,

1:36:34 what would you have done different?" And then oftentimes you get, like,

1:36:39 actually something where they discover only throughout building that, oh,

1:36:44 what we actually did was not optimal.

1:36:47 Many times I- I asked them, "Okay, now that you built it,

1:36:51 what can we refactor?" Because then you build it and you feel the pain points.

1:36:57 I mean, you don't feel the pain points but, right,

1:37:00 they discover where- where there were problems or where things didn't

1:37:05 work e- in the first try and it re- required more loops.

1:37:09 So every time, almost every time I- I merge a PR, build a feature,

1:37:15 afterwards I ask, "Hey, what can we refactor?" Sometimes it's like, "No,

1:37:19 there's, like, nothing big," or, like, usually they say, "Yeah,

1:37:22 this thing you should really look

1:37:26 at." But that took me quite a while to, like...

1:37:29 You know, that flow took me lots of time to understand,

1:37:32 and if you don't do that, you eventually...

1:37:35 you'll stop yourself into- into a corner.

1:37:37 You, like, you have to keep in mind...

1:37:42 they work very much like humans.

1:37:44 Like, I, I, if I write software by myself,

1:37:46 I also build something and then I feel the pain points,

1:37:48 and then I, I get this urge that I need to refactor something.

1:37:53 So, I can very much synthesize with the agent,

1:37:56 and you just need to use the context.

1:38:00 Mm-hmm.

1:38:00 Or, like, you also use the context to write tests.

1:38:04 And so, uh, Codex uh, oppose like the, the, the model, models.

1:38:09 They, they usually do that by default,

1:38:13 but I still often ask the questions, "Hey, do we have enough tests?" "Yeah,

1:38:16 we tested this and this, but this corner

1:38:18 case could be something write more tests." Um, documentation.

1:38:22 Now that the whole context is full, like, I mean,

1:38:26 I'm not saying my documentation is great, but it's not bad.

1:38:32 And pretty much everything is, is LM generated.

1:38:36 So, so, you have to approach it as you build features, as you change something.

1:38:40 I'm like, "Okay, write documentation.

1:38:42 What file would you pick?" You know, like, "What file name?

1:38:46 Where, where would that fit in?" And it gives me a few options.

1:38:48 And I'm like, "Oh, maybe also add it there," and that's all part of the session.

1:38:53 Maybe you can talk about the current two big competitors in terms of models,

1:38:59 Cloud Opus 4.6 and GPT-5 through Codex.

1:39:03 Which is better?

1:39:04 How different are they?

1:39:06 I think you've spoken about Codex reading more and Opus being more willing

1:39:14 to take action faster and maybe being more creative in the actions it takes.

1:39:20 But because-...

1:39:21 Codex reads more, it's able to deliver maybe better code.

1:39:25 Can you speak to the di- n- n- differences there?

1:39:29 I have a lot of words there.

1:39:33 Um, is- as a general purpose model, Opus is the best.

1:39:40 Like, for OpenClaw, Opus is extremely good in terms of role play.

1:39:48 Like, really going into the character that you give it.

1:39:51 It's very good at...

1:39:53 It was really bad, but it really made

1:39:55 an arch to be really good at following commands.

1:40:03 It is usually quite fast at trying something.

1:40:07 It's much more tailored to, like, trial and error.

1:40:11 It's very pleasant to use.

1:40:15 In general, it's almost like Opus was...

1:40:19 Is a little bit too American.

1:40:23 And I shouldn't...

1:40:25 Maybe that's a bad analogy.

1:40:26 You'll probably get roasted for that.

1:40:28 Yeah, I know exactly.

1:40:29 It's 'cause Codex is German.

1:40:30 Is that what you're saying?

1:40:32 It's-- Actually, now that you say it, it makes perfect sense.

1:40:35 Or you could, you could...

1:40:36 Sometimes I- Sometimes I explain it-- I will

1:40:39 never be able to unthink what you just said.

1:40:41 That's so true.

1:40:42 But you also know that a lot of the Codex team is, like, European, um-...

1:40:46 so maybe there's a bit more to it.

1:40:50 That's so true.

1:40:51 Oh, that's funny.

1:40:52 But also, ent- entropic, they fixed it a little bit.

1:40:55 Like, Opus used to say, "You're absolutely right all the time," and it,

1:40:59 it, it today still triggers me.

1:41:00 I can't hear it anymore.

1:41:01 It's not even a joke.

1:41:02 Uh, I just...

1:41:03 You, this was like the, the meme, right?

1:41:07 "You're absolutely right."- You're allergic to sycophancy a little bit.

1:41:11 Yeah.

1:41:12 I, I can't.

1:41:13 Some other comparison is like,

1:41:15 Opus is like the coworker that is a little silly sometimes,

1:41:21 but it's really funny and you keep him around.

1:41:24 And Codex is like the, the weirdo in the corner that you

1:41:27 don't wanna talk to, but is reliable and gets shit done.

1:41:31 Yeah.

1:41:32 Um, ultimately-- This all feels very accurate.

1:41:39 I mean, ultimately, if you're a skilled driver,

1:41:41 you can get good results with any of those latest gen models.

1:41:46 Um, I like Codex more because it doesn't require so much charade.

1:41:53 It will just, it will just read a lot of code by default.

1:41:58 Opus, you really have to, like, you have to have plan mode.

1:42:00 You have to push it harder to, like, go in these directions because it's,

1:42:05 it's just like, like, "Yeah, can I go in?

1:42:07 Can I go in?" You know?

1:42:08 Yeah.

1:42:08 It's like, it will just run off very fast, and that's a very localized solution.

1:42:12 I think it, I think the difference is, is in the post-training.

1:42:15 It's not like the, the raw model intelligence is so different, but it's just...

1:42:20 I think that they just give it, give you different, different goals.

1:42:24 And no model, no model is better in, in in every aspect.

1:42:29 What about the code that it generates?

1:42:31 The, the...

1:42:32 In terms of the actual quality of the code, is it basically the same?

1:42:37 If you drive it right,

1:42:38 Opus even sometimes can make more elegant solutions, but it requires more skill.

1:42:46 It's, it's harder to have so many

1:42:49 sessions in parallel with Cloud Code because it's, it's more interactive.

1:42:53 And I, I think that's what a lot of people like,

1:42:56 especially if they come from coding themselves.

1:43:00 Whereas Codex is much more you have a discussion,

1:43:05 and then we'll just disappear for 20 minutes.

1:43:08 Like, even AMP, they, they now added a deep mode.

1:43:11 They finally...

1:43:12 I mocked them, you know.

1:43:13 We finally saw the light.

1:43:14 And then they had this whole talk about you have to approach it differently,

1:43:18 and I think that's where, that's where people struggle when they just try

1:43:23 Codex after trying Cloud Code is that it's,

1:43:26 it's a slightly diff- it's, it's less interactive.

1:43:28 It's, it's like I have quite long discussions sometimes, and then, like, go off.

1:43:35 And then, yeah, it doesn't matter if it takes 10,

1:43:37 20, 30, 40, 50 minutes or longer, you know?

1:43:39 Like, the 6:00 thing was, like,

1:43:41 six hours.The latest trend can be very, very persistent until it works.

1:43:46 If there's a clear solution, like, "This is, this is what I want at the end,

1:43:52 so it works," the model will work really hard to really get there.

1:43:57 So I think ultimately...

1:44:03 they both need similar time, but on, on, on, on Claude,

1:44:08 it- it's a little bit more trial and error often.

1:44:11 And, and Codex sometimes overthinks.

1:44:13 I prefer that.

1:44:14 I prefer the dry, the dry version where I have to read less over,

1:44:22 over the more interactive nice way.

1:44:26 Like, people like that so much though,

1:44:30 that OpenAI even added a second mode with like a more pleasant personality.

1:44:34 I haven't even tried it yet.

1:44:36 I, I kinda like the brad.

1:44:38 Mm-hmm.

1:44:39 Um, yeah, 'cause it...

1:44:41 I care about efficiency when I build it-- Right-...

1:44:45 and I, I have fun in the very act of building.

1:44:49 I don't need to have fun with my agent who builds.

1:44:52 I have fun with my model that...

1:44:55 where I can then test those features.

1:44:57 How long does it take for you to adjust, you know, if you switch...

1:45:03 I don't know when, when was the last time you switched.

1:45:06 But to adjust to the, the feel.

1:45:08 'Cause you kinda talked about like you have to kinda really feel where,

1:45:11 where a model is strong, where, like how to navigate, how to prompt it, how...

1:45:17 all that kinda stuff.

1:45:18 Like, just by way of advice,

1:45:20 'cause you've been through this journey of just playing with models.

1:45:22 How long does it take to get a feel?

1:45:27 If, if someone switches,

1:45:27 I would give it a week until you actually develop a gut feeling for it.

1:45:33 Yeah.

1:45:33 Um, that's...

1:45:35 if you just...

1:45:36 I think some people also make the mistake

1:45:38 of they pay 200 for the, the Claude code version,

1:45:41 then they pay 20 bucks for the OpenAI version.

1:45:44 But if you pay like the, the 20 bucks version, you get the slow version.

1:45:48 So your experience would be terrible because you're

1:45:51 used to this very interactive, very good system.

1:45:56 And you switch to something that you have very little experience,

1:45:59 then that's gonna be very slow.

1:46:01 So, I think OpenAI shot themselves a little bit

1:46:04 in the foot by making the, the cheap version also slow.

1:46:08 I would, I would have at least a small part of the fast preview.

1:46:14 Or like, the experience that you get when

1:46:18 you pay 200 before degrading to it being slow, because it's already slow.

1:46:23 Mm-hmm.

1:46:23 I mean, they, they made it better.

1:46:24 I think it's...

1:46:25 And, and they have plans to make it a lot better if the Cerebras stuff is true.

1:46:29 But yeah, it's a skill.

1:46:30 It takes time.

1:46:31 Even if you play...

1:46:32 You have a regular guitar and you switch it to an E guitar,

1:46:35 you're not gonna play well right away.

1:46:37 You have to, like, learn how it feels.

1:46:42 The- there's also this extra psychological effect

1:46:44 that you've spoken about which is hilarious to watch.

1:46:47 Which once people, uh...

1:46:50 When the new model comes out, they try that model, they fall in love with it.

1:46:53 "Wow, this is the smartest thing of all time," and then they start saying,

1:46:56 "You could just watch the Reddit posts over time," start saying

1:47:01 that, "We believe the intelligence

1:47:03 of this model has been gradually degrading." It,

1:47:08 it says something about human nature and just the way our minds work,

1:47:12 when it's probably most likely the case

1:47:15 that the intelligence of the model is not degrading.

1:47:19 Uh, it's in fact you're getting used to a good thing.

1:47:23 And your project grows, and you're adding slop,

1:47:25 and you probably don't spend enough time to think about refactors.

1:47:28 And you're making it harder and harder for the agent to work on your slop.

1:47:32 And then, and then suddenly, "Oh, now it's hard.

1:47:35 Oh no, it's not working as well anymore." What's the motivation for, like,

1:47:38 one of those AI companies to actually make their model dumber?

1:47:42 Like, at most, it will make it slower if, if the server load's too high.

1:47:48 But, like, quantizing the model, So you have a worse experience,

1:47:54 so you go to the competitor?

1:47:56 Yeah.

1:47:56 That just doesn't seem like a very smart move in any way.

1:48:00 Uh, what do you think about Claude Code in comparison to Open Claude?

1:48:04 So, Claude Code and maybe the Codex coding agent?

1:48:07 Do you see them as kind of competitors?

1:48:11 I mean, first of all, competitor is fun when it's not really a competition.

1:48:16 Yeah.

1:48:17 Like, I'm happy if...

1:48:19 If, if all it did is, like, inspire people to build something new, cool.

1:48:24 Um, I still use Codex for the building.

1:48:27 I, I know a lot of people use Open Claude to, to build stuff.

1:48:30 And I worked hard on it to make that work.

1:48:33 And I do smaller stuff with it in terms of code.

1:48:38 But, like, if I work hours and hours,

1:48:42 I want a big screen, not WhatsApp, you know?

1:48:46 So for me, a personal agent is much more about my life.

1:48:51 Or like, like a coworker.

1:48:53 Like, I give you, like, a GitHub URL.

1:48:55 Like, "Hey, try out this CLI.

1:48:56 Does it actually work?

1:48:57 What can we learn?" Blah, blah, blah.

1:48:59 But when I'm deep in, deep in the flow, I want to have multiple,

1:49:05 multiple things and it being very, very visible what it, what it does.

1:49:11 So it...

1:49:12 I don't see it as a competition.

1:49:14 It's, it's different things.

1:49:16 But do, do you think there's a a future where the two kinda combine?

1:49:21 Like, your personal agent is also your best developing co-programmer partner?

1:49:29 Yeah, totally.

1:49:30 I think this is where the puck's going,

1:49:32 that this is gonna be more and more your operating system.

1:49:37 The operating system.

1:49:38 And it already...

1:49:39 It's so funny.

1:49:40 Like I, I added support for sub-agents and also for......

1:49:46 um, TTI support, so it could actually run Cloud Coder Codecs.

1:49:52 Mm-hmm.

1:49:53 And because mine's a little bit bossy, it, it,

1:49:57 it started it and it, it, it told him, like, "Who's the boss," basically.

1:50:01 And it was like, "Ah, Codex is obeying me."- Oh, this is a power struggle.

1:50:07 And also the current interface is probably not the final form.

1:50:10 Like, if you think more globally, we are, we copied Google for agents.

1:50:21 You have, like, a prompt, and, and then you have a chat interface.

1:50:26 That, to me, very much feels like when we first created television

1:50:34 and then people recorded radio shows on television and you saw that on TV.

1:50:39 Mm-hmm.

1:50:40 I think there is, there's n- there's better

1:50:44 ways how we eventually will communicate with models,

1:50:48 and we are still very early in this, how will it even work phase.

1:50:56 So, it will eventually converge and we will also figure

1:51:00 out whole different ways how to work with those things.

1:51:05 Uh, one of the other components of workflow is operating system.

1:51:10 So I told you offline that for the first time in my life,

1:51:16 I'm expanding my sort of realm of exploration to the to the Apple ecosystem,

1:51:24 to Macs, iPhone and so on.

1:51:27 For most of my life I've been a Linux, Windows and WSL1,

1:51:32 WSL2 person, which I think are all wonderful, but I...

1:51:36 expanding to also trying Mac.

1:51:37 Because it's another way of building and it's also a way of building

1:51:42 that a large part of the community currently

1:51:45 that's utilizing LMS and agents is using, so.

1:51:47 And that's the reason I'm expanding to it.

1:51:49 But is there something to be said about the different operating systems here?

1:51:52 We should say that OpenClaw supported across operating systems.

1:51:57 Yeah.

1:51:57 I saw WSL2 recommended, side windows for certain o- operations,

1:52:02 but then Windows, Linux macOS are obviously supported.

1:52:07 Yeah, it should even work natively in Windows.

1:52:08 I just didn't have enough time to properly test it.

1:52:12 And you know, like, the last 90% of software always easier than the first 90%,

1:52:17 so I'm sure there's some dragons left that will eventually nail out.

1:52:23 Um, my road was, for a long time, Windows,

1:52:28 just because I grew up with that, then

1:52:30 I switched and had a long phase with Linux, built my own kernels and everything,

1:52:35 and then I went to university and I, I had my, my hacky Linux thing,

1:52:41 and saw this white MacBook,

1:52:43 and I just thought this is a thing of beauty, the white plastic one.

1:52:47 And then I converted to Mac 'cause mostly w- I was,

1:52:52 I was sick that audio wouldn't work on Skype and all

1:52:56 the other issues that, that Linux had for a long time.

1:53:00 And then I just stuck with it and then I dug into iOS,

1:53:04 which required macOS anyhow, so it was never a question.

1:53:07 I think Apple lost a little bit of its lead in terms of native.

1:53:16 It used to be...

1:53:19 Native apps used to be so much better,

1:53:22 and especially in the Mac, there's more people that build software with love.

1:53:27 On, on Windows, it, it...

1:53:29 Windows has much more and, like, function wise, there's just more, period.

1:53:35 But a lot of it felt more functional and less done with love.

1:53:42 Um, I mean, Mac always, like, attracted more designers and people I felt...

1:53:50 Even though, like, often it has less features,

1:53:53 it, it had more delight-- Mm-hmm-...

1:53:56 And playfulness.

1:53:57 So I always valued that.

1:54:00 But in the last few years, many times I actually prefer...

1:54:05 Oh God, people are gonna roast me for that, but I

1:54:11 prefer Electron apps because they work and native apps often,

1:54:17 especially if it's, like, a web service is a native app, are lacking features.

1:54:21 I mean, not saying it couldn't be done,

1:54:24 it's more like a, a focus thing that, like,

1:54:27 for many, many companies, native was not that big of a priority.

1:54:35 But if they build an Electron app, it, it's the only app,

1:54:38 so it is a priority and there's a lot more code sharing possible.

1:54:42 And I, I build a lot of native Mac apps.

1:54:45 I love it.

1:54:45 I, I can, I can help myself.

1:54:49 Like, I love crafting little Mac, Mac, Menu bar tools.

1:54:54 Like I built one to, to monitor your Codex use.

1:54:58 I built one I call Trimmy, that's specifically for agentic use.

1:55:03 When you, when you select text that goes over multiple lines it would

1:55:07 remove the new line so you could actually paste it to the terminal.

1:55:11 That was, again like, this is annoying me and after the, the 20th

1:55:14 time of it is annoying me, I just built it.

1:55:16 There is a cool Mac app for OpenClaw

1:55:19 that I don't think many people discovered yet,

1:55:21 also because it, it still needs some love.

1:55:23 It feels a little bit too much like the Hummer car right now because I,

1:55:27 I just experiment a lot with it.

1:55:30 It, it likes to polish.

1:55:32 So you still...

1:55:33 I mean, you still love it.

1:55:34 You still, you still love adding to the delight of that operating system.

1:55:37 Yeah, but then you realize...

1:55:38 Like, I also built one, for example, for GitHub.

1:55:40 And then the...

1:55:42 If you use SwiftUI, like the latest and greatest at Apple,

1:55:46 and took them forever to build something to show an image from the web.

1:55:52 Now we have async, async image, but......

1:55:55 I added support for it and then some images

1:55:58 would just not show up or, like, be very slow.

1:56:00 And I had a discussion with Codex like, "Hey,

1:56:03 why is there a bug?" And even Codex said like, "Yeah,

1:56:06 there's this ASIC image but it's really more for experimenting and it

1:56:10 should not be used in production." But that's Apple's answer to, like,

1:56:15 showing images from the web.

1:56:18 This shouldn't be so hard, you know.

1:56:19 This is like...

1:56:20 This is like insane.

1:56:21 Like, how am I in, in, in 2026 and my agent tell me,

1:56:26 "Don't use the stuff Apple built because it's, it's...

1:56:29 It's...

1:56:29 Yeah, it- it's there but it's not good." And like this is now in the weeds.

1:56:34 This is...

1:56:35 To me this is like, um...

1:56:38 They had so much head start and so much love,

1:56:43 and they kind of just like blundered it and didn't,

1:56:47 didn't evolve it as much as they should.

1:56:50 But also, there's just the practical reality.

1:56:52 If you look at Silicon Valley,

1:56:55 most of the developer world that's kind of playing with LMS and Agentic AI,

1:57:01 they're all using Apple products.

1:57:03 And then, at the same time, Apple is not really, like, leaning on that.

1:57:07 Like they're not...

1:57:08 They're not opening up and playing and working together and like, yes.

1:57:13 Isn't, isn't it funny how they completely blunder AI,

1:57:16 and yet everybody's buying Mac Minis?

1:57:19 How...

1:57:20 What...

1:57:20 Does that even make sense?

1:57:21 You're, you're, you're quite possibly the world's

1:57:25 greatest Mac salesman of all time.

1:57:29 No, you don't need a Mac Mini to install OpenClaw.

1:57:32 You can install it on the web.

1:57:34 There's, there's a concept called nodes,

1:57:36 so you can like make your computer a node and it will do the same.

1:57:41 There is something said for running it on separate hardware.

1:57:48 That right now is useful.

1:57:51 Um, there is...

1:57:53 There's a big argument for the browser.

1:57:58 You know, I, I built some Agentic browser use in there.

1:58:01 And, I mean, it's basically Playwright with a bunch

1:58:04 of extras to make it easier for agents.

1:58:06 Playwright is a library that controls the browser.

1:58:08 Yeah.

1:58:08 It's really nice, easy to use.

1:58:10 And our internet is slowly closing down.

1:58:12 Like, there, there's a whole movement to make it harder for agents to use.

1:58:16 So if you do the same in a data center

1:58:20 and websites detect that it's an IP from a data center,

1:58:22 the website might just block you or it make it really hard

1:58:26 or put a lot of captures in the, in the way of the agent.

1:58:28 I mean, agents are quite good at happily clicking, "I'm not a robot."- Yeah.

1:58:34 Um, but having that on a residential IP makes a lot of things simpler.

1:58:41 So there's ways.

1:58:42 Yeah.

1:58:42 But it really does not need to be a Mac.

1:58:45 It can...

1:58:45 It can be any old hardware.

1:58:48 I always say, like, maybe use the...

1:58:51 Use the opportunity to get yourself a new MacBook or whatever computer you use

1:58:58 and use the old one as your server instead of buying a standalone Mac Mini.

1:59:03 But then there's, again,

1:59:04 there's a lot of very cute things people build with Mac Minis that I like.

1:59:08 Yeah.

1:59:08 Um, and no, I don't get commission from Apple.

1:59:13 They didn't really communicate much.

1:59:16 It's sad.

1:59:17 It's sad.

1:59:18 Can you actually speak to what it takes to get started with OpenClaw?

1:59:22 There's...

1:59:22 I mean, there's a lot of people...

1:59:24 What is it?

1:59:25 Somebody tweeted at you, "Peter,

1:59:26 make OpenClaw easy to set up for everyday people.

1:59:30 99.9% of people can't access to OpenClaw and have their own

1:59:35 lobster because of their technical difficulties in getting it set up.

1:59:38 Make OpenClaw accessible to everyone, please." And you replied,

1:59:41 "Working on that." From my perspective, it seems there- there's a bunch

1:59:44 of different options and it's already quite straightforward,

1:59:47 but I suppose that's if you have some developer background.

1:59:51 I mean, right now you have to paste in one liner into the terminal.

1:59:54 Right.

1:59:54 And there's also an app.

1:59:57 The app kind of does that for you, but there should be a Windows app.

2:00:01 The app needs to be easier and more loved.

2:00:04 The configuration should potentially be web- based or in the app.

2:00:08 And I started working on that, but honestly

2:00:12 right now I want to focus on security aspects.

2:00:17 And, and once I'm confident that this is at a level that I can recommend my mom,

2:00:24 then I'm going to make it simpler.

2:00:25 Like I...

2:00:26 Right now-- You want to make it harder so

2:00:30 that it doesn't scale as fast as it's scaling.

2:00:32 Yeah, it would be nice if it wouldn't...

2:00:33 I mean, that's, like, hard to say, right?

2:00:35 But if the growth would be a little slower,

2:00:38 that would be helpful because people are

2:00:42 expecting inhuman things from a single human being.

2:00:44 And yes, I have some contributors,

2:00:46 but also that whole machinery I started a week ago so,

2:00:51 That needs more time to figure out.

2:00:54 And, and not everyone has all day to work on that.

2:01:00 There's some beginners listening to this, programming beginners.

2:01:04 What advice would you give to them about,

2:01:07 let's say, joining the Agentic AI revolution?

2:01:12 Play.

2:01:13 Um, playing is the best...

2:01:15 The best way to learn.

2:01:17 If you wanna...

2:01:18 I'm sure if you...

2:01:19 If you are like a little bit of builder,

2:01:22 you have an idea in your head that you want to build,

2:01:24 just build that, or like, give it a try.

2:01:25 It doesn't need to be perfect.

2:01:27 I built a whole bunch of stuff that I don't use.

2:01:29 It doesn't matter.

2:01:30 Like, it's the journey.

2:01:31 Mm-hmm.

2:01:31 You know?

2:01:32 Like the philosophical way, that the end doesn't matter, the journey matters.

2:01:35 Have fun.

2:01:37 Mm-hmm.

2:01:38 My God, like those things...

2:01:39 I...

2:01:39 I don't think I ever had so much fun building

2:01:42 things because I can focus on the hard parts now.

2:01:45 A lot of coding, I always thought I liked coding, but really I like building.

2:01:50 Yeah.

2:01:51 And...

2:01:51 And whenever you don't understand something, just ask.

2:01:53 You have an infinitely patient answering machine....

2:01:57 that y- can explain you anything at any level of complexity.

2:02:02 Sometimes, that's like one time I asked,

2:02:04 "Hey explain to me like I'm- I'm eight years old,"

2:02:07 and it started giving me a story with crayons and stuff.

2:02:10 And I'm like, "No, not like that." Like, I'm okay-...

2:02:13 up- up the age a little bit, you know?

2:02:14 I'm like, I'm not an actual child, it's just,

2:02:16 I just need a simpler language for like a- a- a- a- a tricky

2:02:21 database concept that I didn't grok in the first- first time.

2:02:25 But, you know, just, you can just ask things.

2:02:28 Like, you- there's like...

2:02:30 It used to be that I had to go on Stack Overflow or ha- ask on Twitter,

2:02:35 and then maybe two days later I get a response.

2:02:37 Or I had to try for hours.

2:02:39 And now you- you can just ask stuff.

2:02:42 It- I mean, it's never...

2:02:43 You have, like, your own teacher.

2:02:45 You know that there's like statistics,

2:02:47 y- you can learn faster if you have your own teacher.

2:02:50 The- it's like you have this infinitely patient machine.

2:02:52 Ask it.

2:02:53 But what would you say?

2:02:54 So use...

2:02:55 What's the easiest way to play?

2:02:56 So maybe Open Claw is a nice way to play so you

2:02:59 can then set- set everything up and then you could chat with it.

2:03:03 You can also just experiment with it and, like, modify it.

2:03:07 Ask your agent.

2:03:08 I mean, there is infinite ways how it can be made better.

2:03:16 Play around, make it better.

2:03:18 Mm-hmm.

2:03:19 More general, if you- if you're a beginner and you

2:03:22 actually wanna learn how to build software really fast,

2:03:27 get involved in open source.

2:03:28 Doesn't need to be my project.

2:03:30 In fact, maybe don't use my project because my- my backlog is very large,

2:03:36 but I learned so much from open source.

2:03:39 Just like, like, be- be humble.

2:03:42 Don't- maybe don't send a pull request right away.

2:03:45 But there's many other ways you can help out.

2:03:48 There's many ways you can just learn by just reading code.

2:03:51 By- by being on Discord or wherever people are,

2:03:56 and just, like, understanding how things are built.

2:03:59 I don't know, like Mitchell Hashimoto builds Ghostly, the terminal,

2:04:05 and he has a really good community where there's so many other projects.

2:04:09 Like, pick something that you find interesting and get involved.

2:04:15 Do you recommend that people that don't know how to program

2:04:19 or don't really know how to program learn to program also?

2:04:23 So when you you can get quite far

2:04:27 right now by just using natural language, right?

2:04:29 Do you s- still see a lot of value in reading the code, understanding the code,

2:04:35 and then being able to write a little bit of code from scratch?

2:04:39 It definitely helps.

2:04:40 It's hard for you to answer that-- Yeah-...

2:04:42 because you don't know what it's like to do

2:04:44 any of this without knowing the base knowledge.

2:04:47 Like, you might take for granted just how much intuition

2:04:50 you have about the programming world having programmed so much, right?

2:04:54 There's people that are high agency and very curious,

2:04:58 and they get very far even though they have no deep understanding

2:05:02 how software works just because they

2:05:05 ask questions and questions and- and- and-...

2:05:09 and agents are infinitely patient.

2:05:11 Like, part of what I did this year is I went

2:05:14 to a lot of iOS conferences because that's my background and just told people,

2:05:18 "Don't consi- don't see yourself as an iOS engineer anymore." Like,

2:05:21 "You need to change your mindset.

2:05:23 You're a builder." And you can take a lot of the knowledge how to build

2:05:28 software into new domains and all

2:05:31 of the- the more fine-grain details, agents can help.

2:05:35 You don't have to know how to splice an array

2:05:38 or what the- what the correct template syntax is or whatever,

2:05:41 but you can use all your- your general knowledge

2:05:45 and that makes it much easier to move from one galaxy,

2:05:49 one tech galaxy into another.

2:05:51 And oftentimes, there's languages that make more

2:05:54 or less sense depending on what you build, right?

2:05:57 So for example, when I build simple CLIs, I like Go.

2:06:01 I actually don't like Go.

2:06:03 I don't like the syntax of Go.

2:06:05 I didn't even consider the language.

2:06:07 But the ecosystem is great, it works great with agents.

2:06:11 It is garbage collected.

2:06:13 It's not the highest performing one, but it's very fast.

2:06:17 And for those type of- of CLIs that I build, Go is- is a really good choice.

2:06:21 So I- I use a language I'm not even a fan of for...

2:06:26 That's my main to-go thing for- for CLIs.

2:06:29 Isn't that fascinating that here's a programming language you would've

2:06:31 never used if you had to write it from scratch

2:06:34 and now you're using because LMs are good at generating it

2:06:38 and it has some of the characteristics that makes it resilient,

2:06:42 like garbage collected?

2:06:44 Because everything's weird in this new world and that just makes the most sense.

2:06:48 What's the best Ridiculous question.

2:06:49 What's the best programming language for the AI- AI agentic world?

2:06:52 Is it JavaScript, TypeScript?

2:06:55 TypeScript is really good.

2:06:57 Sometimes the types can get really confusing and the ecosystem is- is a jungle.

2:07:06 So for- for web stuff it's good.

2:07:11 I wouldn't build everything in it.

2:07:15 Don't you think we're moving there?

2:07:17 Like, that everything will eventually be written-

2:07:20 eventually is written in JavaScript and it--

2:07:22 The birth and death of JavaScript and we are living through it in real time.

2:07:26 Like, what does programming look like in 20 years?

2:07:28 Right?

2:07:29 In 30 years?

2:07:29 In 40 years?

2:07:30 What do programs and apps look like?

2:07:32 You can even ask a question like,

2:07:33 do we need a- a programming language that's made for agents?

2:07:36 Because all of those languages are made for humans.

2:07:40 So how- what would that look like?

2:07:42 Um, I think there's a- there's whole

2:07:44 bunch of interesting questions that we'll discover.

2:07:48 And also how because everything is now world knowledge, how it in many ways,

2:07:54 things will stagnate 'cause if you build something new and the agent has

2:07:57 no idea that's gonna be much harder to use than something that's already there.

2:08:02 Um......

2:08:02 of when I build Mac apps, I build them in, in Swift and SwiftUI,

2:08:08 mm, partly because I like pain, partly because it...

2:08:12 the, the deepest level of system integration, I can only get through there.

2:08:17 And you clearly feel a difference if you click on an electron

2:08:21 app and it loads a web view in the menu.

2:08:24 It's just not the same.

2:08:26 Um, sometimes I just also try new languages just to, like, get a feel for them.

2:08:32 Like Zig?

2:08:33 Yeah.

2:08:33 If it's something that...

2:08:34 where I care about performance a lot then it's,

2:08:37 it's a really interesting language.

2:08:38 And it...

2:08:39 like agents got so much better over the last

2:08:42 six months from not really good to totally valid choice.

2:08:47 Just still a, a very young ecosystem.

2:08:51 And most of the time you actually care about ecosystem, right?

2:08:56 So, so if you build something that does inference

2:09:01 or goes into whole running model direction, Python, very good.

2:09:07 Mm-hmm.

2:09:07 But then if I build stuff in Python and I want a story

2:09:10 where I can also deploy it on Windows, not a good choice.

2:09:13 Mm-hmm.

2:09:14 Sometimes I, I found projects that kinda did

2:09:16 90% of what I wanted but were in Python, and I wanted them...

2:09:20 I wanted an easy Windows story.

2:09:21 Okay, just rewrite it in Go.

2:09:24 Um, but then if you go towards multiple,

2:09:28 multiple threads and a lot more performance, Rust is a really good choice.

2:09:33 There's no...

2:09:33 there's just no single answer, and it's also the beauty of it.

2:09:36 Like, it's fun.

2:09:36 And now it doesn't matter anymore,

2:09:38 you can just literally pick the language that has

2:09:41 the, the most fitting characteristics and ecosystem-- Mm-hmm-...

2:09:46 for your problem domain.

2:09:48 And yeah, it might be...

2:09:49 You might have s-...

2:09:50 You might be a little bit slow in reading the code, but not really.

2:09:54 Y- I think you, you pick stuff up really fast,

2:09:57 and you can always ask your agent.

2:10:00 So there's a lot of programmers and builders

2:10:02 who draw inspiration from y- your story.

2:10:06 Just the way you carry yourself, your choice of making OpenClaw open source,

2:10:13 the, the way you have fun building and exploring,

2:10:18 and doing that, for the most part, alone or on a small team.

2:10:23 So by way of advice,

2:10:25 what metric should be the goal that they would be optimizing for?

2:10:29 What would be the metric of success?

2:10:32 Would it be happiness?

2:10:34 Is it money?

2:10:35 Is it positive impact for people who are dreaming of building?

2:10:39 'Cause you went through an interesting journey.

2:10:41 You've achieved a lot of those things,

2:10:43 and then you fell out of love with programming a little bit for a time.

2:10:47 I was just burning too bright for too long.

2:10:53 I, I ran...

2:10:54 I started PSPDFKit, s- and ran it for 13 years, and it was high stress.

2:11:04 Um, I had to learn all these things fast and hard, like how to manage people,

2:11:10 how to bring people on, how to deal with customers, how to do...

2:11:14 So it wasn't just programming stuff, it was people stuff.

2:11:18 The stuff that burned me out was mostly people stuff.

2:11:22 I, I don't think burnout is working too much.

2:11:27 Maybe to a degree.

2:11:28 Everybody's different.

2:11:29 You know, I c- I cannot speak in a- in absolute terms,

2:11:32 but for me, it was much more differences,

2:11:37 With my, my co-founders, conflicts, or, like,

2:11:41 really high stress situation with customers that eventually grinded me down.

2:11:48 And then when...

2:11:51 luckily we, we got a really good offer for, like,

2:11:56 putting the company to the next level and I,

2:12:00 I already kinda worked two years on making myself obsolete.

2:12:03 So at this point I could leave, and, and then I just...

2:12:07 I was sitting in front of the screen and I felt like,

2:12:09 you know Austin Powers where they suck the mojo out?

2:12:13 Yeah.

2:12:14 Uh, I g- I was like, m- m- it was, like, gone.

2:12:17 Like, I couldn't...

2:12:18 I couldn't get code out anymore.

2:12:20 I was just, like, staring and feeling empty, and then I, I just stopped.

2:12:32 I, I booked, like, a one-way trip to Madrid and, and, and just,

2:12:37 like, spent a t- some t- sometime there.

2:12:39 I felt like I had to catch up on life,

2:12:42 so I did a whole, a whole bunch of life catching up stuff.

2:12:48 Did you go through some lows during that period?

2:12:52 And you know, maybe advice on...

2:12:55 of how to?

2:12:57 Uh, maybe advice on how to approach life.

2:12:58 If you think that, "Oh yeah,

2:13:00 work really hard and then I'll retire," I don't recommend that.

2:13:06 Because the idea of, "Oh yeah, I just enjoy life now," a- maybe it's appealing,

2:13:17 but right now I enjoy life, the most I've ever enjoyed life.

2:13:23 Because if you wake up in the morning and you

2:13:28 have nothing to look forward to, you have no real challenge,

2:13:33 that gets very boring, very fast.

2:13:36 And then when, when you're bored,

2:13:39 you're gonna look for other places how to stimulate yourself,

2:13:43 and then maybe, maybe that's drugs, you know?

2:13:48 But that eventually also get boring and you look for more,

2:13:52 and that will lead you down a very dark path.

2:13:57 But you also showed on the money front, you know,

2:13:59 a lot of people in Silicon Valley and the startup world,

2:14:01 they think, maybe overthink way too much optimized for money.

2:14:05 And you've also shown that it's not like you're saying no to money.

2:14:08 I mean, I'm sure you take money, but it's not......

2:14:13 the primary objective of uh, of your life.

2:14:17 Can you just speak to that?

2:14:18 Your philosophy on money?

2:14:20 When I built my company, money was never the driving force.

2:14:23 It felt more like, like, an affirmation that I did something right.

2:14:27 And having money solves a lot of problems.

2:14:30 I also think there, there's diminishing returns the more you have.

2:14:36 Um, like, a cheeseburger is a cheeseburger,

2:14:40 and I think if you go too far into, oh,

2:14:47 I do private jet and I only travel luxury, you disconnect with society.

2:14:55 Um, I, I donated quite a lot.

2:15:00 Like, I have a, I have a foundation for helping people that weren't so lucky.

2:15:11 And disconnecting from society is bad in that on many levels,

2:15:14 but one of them is, like, humans are awesome.

2:15:18 It's nice to continuously remember the awesomeness in humans.

2:15:23 I, I mean, I could afford really nice hotels.

2:15:25 The last time I was in San Francisco,

2:15:26 I did the, the first time the OG Airbnb experience-- Yeah, yeah-...

2:15:31 and just booked a room.

2:15:32 Mostly because I, I thought, okay, you know, I'm out or I'm sleeping,

2:15:37 and I don't like where all the hotels are,

2:15:40 and I wanted a, I wanted a different experience.

2:15:43 I think, isn't life all about experiences?

2:15:46 Like, if you, if you tailor your life towards, "I wanna have experiences," it,

2:15:52 it reduces the need for, "It needs to be good or bad." Like,

2:15:57 if people only want good experiences, that's not gonna work,

2:16:00 but if you optimize for experiences, if it's good, amazing.

2:16:05 If it's bad, amazing, because, like,

2:16:09 I learned something, I saw something, did something.

2:16:10 I wanted to experience that, and it was amazing.

2:16:13 Like, there was, like, this, this queer DJ in there,

2:16:17 and I showed her how to make music with cloud code.

2:16:21 And we, like, immediately bonded and had a great time.

2:16:24 Yeah, there's something about that air- you know,

2:16:26 couch surfing, Airbnb experience, the OG.

2:16:28 I'm still to this day.

2:16:30 It's awesome.

2:16:30 It's humans, and that's why travel is awesome.

2:16:34 Yeah.

2:16:34 Just experience the variety of, the diversity of human.

2:16:36 And when it's shitty, it's good too, man.

2:16:38 If it rains and you're soaked and it's all fucked,

2:16:41 and planes, the everything is shit, everything is fucked, it's still awesome.

2:16:45 If you're able to open your eyes it's good to be alive.

2:16:50 Yeah, and anything that creates emotion and feelings is good.

2:16:55 Even...

2:16:55 So, so maybe, maybe even the cryptic

2:16:57 people are good because they definitely created emotions.

2:16:59 I, I don't know if I should go that far.

2:17:03 No, man.

2:17:03 Give them, give them all, give them love.

2:17:05 Give them love.

2:17:06 Because I do think that online lacks some of the awesomeness of real life.

2:17:13 Yeah.

2:17:14 That's, that's, it's an open problem of how to solve,

2:17:17 how to infuse the online cyber experience with I don't know,

2:17:25 With the intensity that we humans feel when it's in real life.

2:17:29 I don't know.

2:17:30 I don't know if that's a solvable problem.

2:17:31 Well, it's just possible because text is very lossy.

2:17:35 Yeah.

2:17:35 You know, sometimes I wish if I talked to the agent I would...

2:17:39 It should be multi-model so it also understands my emotions.

2:17:43 I mean, it, it might move there.

2:17:45 It might move there.

2:17:46 It will.

2:17:47 It will.

2:17:48 It totally will.

2:17:50 I mean, I have to ask you, just curious.

2:17:52 I, I know you've probably gotten huge offers from major companies.

2:17:56 Can you speak to who you're considering working with?

2:18:04 Yeah.

2:18:05 So, to like explain my thinking a little bit,

2:18:10 right, I did not expect this blowing up so much.

2:18:17 So, there's a lot of doors that opened because of it.

2:18:22 There's, like, I think every VC,

2:18:26 every big VC company is in my inbox and tried to get 15 minutes of me.

2:18:32 So, there's, like, this butterfly effect moment.

2:18:36 I could just do nothing and continue and I really like my life.

2:18:43 Valid choice.

2:18:44 Almost.

2:18:44 Like, I considered it when I delete it, wanted to delete the whole thing.

2:18:52 I could create a company.

2:18:58 Been there, done that.

2:19:00 Um, there's so many people that push me towards that and, yeah,

2:19:06 like, could be amazing.

2:19:07 Which is to say that you, you would probably raise a lot of money in that.

2:19:11 Yeah.

2:19:11 I don't know, hundreds of millions, billions.

2:19:13 I don't know.

2:19:13 It could just got unlimited amount of money.

2:19:16 Yeah.

2:19:16 It just doesn't excite me as much because I feel I did all

2:19:22 of that, and it would take a lot of time away from the things I actually enjoy.

2:19:31 Same as when, when I was CEO, I think I,

2:19:35 I learned to do it and I'm not bad at it, and partly I'm good at it.

2:19:41 But yeah, that path doesn't excite me too much,

2:19:44 and I also fear it, it would create a natural conflict of interest.

2:19:48 Like, what's the most obvious thing I do?

2:19:50 I, I prioritize it.

2:19:52 I put, like, a version safe for workplace.

2:19:54 And then what do you do?

2:19:56 Like, I get a pull request with a feature like an audit log,

2:20:01 but that seems like an enterprise feature,

2:20:05 so now I feel I have a conflict of interest

2:20:08 in the open-source version and the closed- source version....

2:20:13 or change the license to something like FSL,

2:20:17 where you cannot actually use it for commercial stuff,

2:20:20 would first be very difficult with all the contributions.

2:20:23 And second of all, I- I like the idea

2:20:27 that it's free as in beer and not free with conditions.

2:20:31 Um, yeah, there's ways how you, how you keep all of that for free and just,

2:20:36 like, still try to make money, but those are very difficult.

2:20:41 And you see there's, like, fewer and fewer companies manage that.

2:20:45 Like, even Tailwind, they're, like, used by everyone.

2:20:49 Everyone uses Tailwind, right?

2:20:51 And then they had to cut off 75% of the employees because they're not making

2:20:54 money because nobody's even going on the website

2:20:56 anymore because it's all done by agents.

2:21:00 S- and just relying on donations, yeah, good luck.

2:21:03 Like, if a project of my caliber,

2:21:06 if I extrapolate what the typical open-source project would get it's not a lot.

2:21:15 I s- I still lose money on the project because

2:21:17 I made the point of supporting every dependency, except Slack.

2:21:22 They are a big company.

2:21:23 They can, they can, they can do without me.

2:21:25 But all the projects that are done by mostly individuals so, like,

2:21:30 all the, right now, all the sponsorship goes right up to my dependencies.

2:21:35 And if there's more, I want to, like, buy my contributors some merch, you know?

2:21:43 So you're losing money?

2:21:44 Yeah, right now I lose money on this.

2:21:46 So it's really not sustainable?

2:21:48 Uh, I mean, it's like, I guess something between 10 and 20K a month.

2:21:54 Um, which is fine.

2:21:57 I'm sure over time I could get that down.

2:22:00 Um, OpenAI is helping out a little bit with tokens now.

2:22:05 And there's other companies that have been generous.

2:22:08 But yeah, still losing money on that.

2:22:12 So that's- that's one path I consider, but I'm just not very excited.

2:22:17 And then there's all the big labs that I've been talking to.

2:22:25 And from those Meta and OpenAI seem the most interesting.

2:22:32 Do you lean one way or the other?

2:22:35 Uh, yeah.

2:22:38 Um...