0:00 Today, I'm sitting down with Peter Steinberger, the creator of Open Claw,
0:03 the open-source personal AI agent that has completely taken over the internet.
0:08 The GitHub repo exploded to over 160,000 stars practically overnight.
0:14 >> >> The community has built countless projects like Maltbook, where bots talk among themselves.
0:19 And now, the bots are even renting humans to do tasks in the real
0:23 world.
0:24 In our conversation, we discuss his aha moment, his contrarian development philosophies,
0:29 and what this means for builders in 2026.
0:32 Let's dive in.
0:38 >> >> So good to see you, man.
0:39 >> Hey, what's up?
0:40 Um so, you have made something people want.
0:44 It seems so.
0:45 Yeah, uh Open Claw, as it's called now, has absolutely >> five,
0:49 yeah. >> >> has been absolutely exploding the internet.
0:53 Um how have the past one or two weeks been for you, man?
0:56 Oh my god, I need like I need a cave.
1:01 >> >> A week of solitude.
1:03 You You came out of the cave,
1:04 and you want to go back to the cave like a like a lobster.
1:07 It's been absolutely wild.
1:08 I don't know how one human can absorb all of that.
1:11 I probably need another week just to like respond to all my emails.
1:14 Uh I got some incredibly cool stuff.
1:17 I got some incredibly Mm. bad stuff.
1:20 Um but clearly, I hit something
1:22 that spurred up emotions
1:25 and made people interested
1:26 and inspired people,
1:27 and that's really cool.
1:28 And a lot of people have been working on, you know,
1:30 AI and even personal assistants.
1:32 Like, what what is it that made Open Claw take off?
1:36 I think my big difference is that it actually runs on your computer.
1:40 Like, every everything I saw so far runs in the cloud.
1:43 It's like it can do a few things.
1:46 If you run on your computer, it can do every effing thing.
1:50 Right?
1:51 So, that's way more powerful.
1:53 Yeah.
1:53 Machine can do anything that you can do with the machine.
1:56 >> You you can just connect to your oven or your Tesla or your lights,
2:02 your sonars, my bed.
2:03 You can control the temperature of my bed.
2:06 Uh ChatGPT can't do that.
2:07 You gave it all the skills that you have yourself.
2:10 >> A friend told me like he installed Open Claw
2:16 and it and then it asked it like,
2:17 "Look through my computer and make a narrative of all my last year."
2:21 And it made this incredibly good narrative.
2:23 And he was like, "How did you do that?"
2:26 And then he the Open Claw found audio files where like every Sunday he
2:30 was recording stuff.
2:32 And Open Claw found that.
2:35 But he didn't even remember about it
2:36 because it was like more than a year ago,
2:38 right?
2:39 So so just by it being able to search a whole computer,
2:43 it it can surprise you.
2:46 Yeah.
2:46 It's also you also give it all the data, right?
2:49 So it can surprise you in many ways.
2:52 And so now you have, you know, we're even moving from human to bot,
2:56 so like interactions that you've been talking about, to bot-to-bot interactions,
3:01 or even like bot to other humans where, you know,
3:04 bots on behalf of you are then hiring other humans to accomplish tasks IRL.
3:09 Like what's happening?
3:12 I think that's a natural next step.
3:14 Like okay, I want to book a restaurant.
3:18 My bot will reach out to the restaurant bot and do the negotiation.
3:24 Like because it's more efficient.
3:25 Or or maybe it's like an old restaurant,
3:28 so my bot needs to actually get some some human work done
3:31 so that a human
3:32 then calls the restaurant
3:33 because they don't like bots.
3:35 Or or walks there to stand in line.
3:37 If he doesn't get a robot for >> For the owner of the bot.
3:41 >> >> And I imagine it like maybe if if I have even multiple bots.
3:46 And maybe I have like specialists.
3:47 One is like for my private life
3:49 and one is for like my person my my my work stuff.
3:52 Maybe one is our relationship bot
3:54 that gets like a lot of things in between.
3:56 Uh I don't know.
3:57 We're so early.
3:58 There's still so much
4:00 so many things that we haven't really figured out
4:02 if it actually works.
4:04 Um but I feel we are we're on the timeline now.
4:07 It seems like everyone was chasing sort of like the sort of like centralized
4:11 god intelligence.
4:12 And what's sort of emerged over the past, you know,
4:14 10 days or so is sort of like the swarm intelligence um
4:18 and and the community intelligence.
4:20 I think that if you look at one human being what can one human
4:25 being actually achieve?
4:26 You think one human being could make an iPhone?
4:28 Or one human being could go to space?
4:31 I think one human being would probably just like not even be able to
4:34 like find food.
4:37 Um but as a group we specialize.
4:41 As a larger society, we specialize even more.
4:45 So, what can we learn from that that we can apply to AI?
4:49 You know, we we already have like AI that specializes in certain things.
4:54 Um even though that it's it's generalized intelligence,
4:56 what if it actually is also specialized intelligence?
5:01 So, I don't know.
5:01 It's going to be very exciting.
5:03 >> Very cool.
5:03 Yeah, you kind of like opened a window into the future
5:06 and now a ton of people are kind of like building building on it
5:08 and have sort of like their aha moment.
5:11 Um can you walk me back to
5:12 when you had your aha moment
5:13 and kind of like recount
5:15 that very moment?
5:16 I wanted something like just type stuff so that my computer would do stuff.
5:20 Like very simple.
5:21 And then I built I built a version of
5:24 that in May-June that was cool
5:28 but wasn't really yet.
5:29 Um and then I built a whole bunch of other stuff
5:33 and kind of like build up my army.
5:37 And then in November there was a day where I wanted this again.
5:42 Like I I went to the kitchen
5:44 and all I I was check up
5:45 if my computer would still do stuff.
5:47 Or it being finished.
5:49 And doing stuff was was coding.
5:50 You were coding stuff. >> course.
5:52 Were you coding something else or were you coding the thing itself?
5:57 No, no.
5:58 That was just like the need was it again there
6:00 and I'm like What were you coding at the time?
6:03 What were you building?
6:04 My god, you see my my GitHub is like it's like 40 projects.
6:07 I don't even know.
6:09 Um I think it was summarize.
6:11 It's like a It's like a little CLI app where you can give it
6:15 whatever like a podcast
6:16 or um a hot seat thing like here
6:19 and it would summarize it,
6:20 but it would also show you the slides in the terminal.
6:22 Cuz you can do that nowadays.
6:24 Yeah.
6:24 You can just do things.
6:25 So for the love of the computer,
6:27 you kind of like started messing with stuff.
6:29 Yeah.
6:29 Um you came out of retirement actually, right?
6:32 Um to sort of like mess with AI.
6:34 Yeah.
6:34 >> And then increasingly you were
6:35 so hooked that you wanted to just do it always also on the go
6:38 with the phone.
6:39 I mean the last project I I worked two months on Wipe Tunnel to
6:44 the point where it got
6:44 so good that I was catching myself always like coding next to my
6:49 when I was at my friends
6:50 and I was like I need to stop this.
6:52 This is like too addictive.
6:54 And then in November
6:54 and like my need came back
6:56 and I I started building Cloud Bot
6:59 or now it's called Open Claw.
7:01 And I think very very in the beginning I was like, "Oh,
7:04 I rebuilt it again.
7:06 But this time I built it even better."
7:07 Mhm.
7:08 This time when you don't type into your terminal,
7:09 you just you talk to a friend.
7:12 You don't think about compaction, which sessions, which folder I'm in,
7:16 which model I'm in.
7:17 I mean, you can, you know,
7:19 it's just like I want to leave it open for power users,
7:23 but usually you just like you just talk to a friend
7:25 and the friend is like this ghost
7:28 or entity or whatever you want to call it
7:30 that can control your mouse
7:32 and your keyboard and can just do stuff.
7:34 Yeah.
7:34 And when did you have that aha moment when you were like, "Wow,
7:37 this is doing way more things than I actually thought it could."
7:41 >> Literally I took me 1 hour for like the the very shitty initial prototype.
7:47 It was just a little bit of glue between like a dependency
7:51 that connects WhatsApp and Cloud Code
7:53 and then I would like call Cloud Code
7:56 and get like the string out of Cloud Code.
7:57 It would be slow but it it worked.
8:00 But I wanted images cuz you know you want pictures.
8:03 I want I want I want the model to send a selfie
8:05 or whatever and I want the model to create images
8:07 and send me back.
8:09 So that took me another few hours.
8:11 And then I I went to Marrakech for a birthday party
8:15 and there was like the internet wasn't
8:17 that good,
8:17 you know, WhatsApp works everywhere because I don't know it's just like text.
8:21 So I used it a lot I was in a restaurant.
8:23 What does this mean?
8:24 You make like a picture
8:25 and like translate this for me
8:26 and just it was just
8:27 so useful.
8:28 And it was also really nice about it because it it spoke my language.
8:31 You know, it it was a little sassy.
8:33 It was like funny.
8:35 It was like really pleasant to use.
8:36 And then I was walking and just like sending it a voice message.
8:40 And I'm like, oh wait, this can't work.
8:42 I didn't build that.
8:43 Right.
8:44 Right.
8:44 And you see like the type indicator is like blinking blinking blinking. 10 seconds
8:48 later it just replied to me.
8:50 I'm like, how in the F did you do that?
8:53 And it replied, yeah, the model I did the following.
8:56 You sent me a text message
8:58 and there was no file ending
8:59 so I looked at the header I found it's Opus.
9:01 So I used FFmpeg to convert it to wave.
9:04 And then I wanted to like transcribe it
9:05 but I didn't have Whisper installed
9:07 but then I looked around
9:07 and I found this OpenAI key
9:09 and I just used curl to send it to OpenAI.
9:11 Got the text back and here I am.
9:15 And that all in like what?
9:16 9 seconds?
9:17 And you didn't build or anticipate like any of those specific things?
9:21 >> No, it you know, turns out um because coding models got so good,
9:26 coding is really like creative problem solving
9:30 that maps very well back into the real world.
9:32 I think I think there's there's a there's a huge correlation.
9:36 They need to be really good at creative problem solving.
9:39 And that's a skill.
9:40 That's an abstract skill you can apply to code
9:43 but like to any real world task.
9:44 So the the model had a I was surprised.
9:47 It's like a magical file.
9:48 I don't know what it is.
9:49 I need to solve this.
9:50 And it did its best and solved it.
9:51 And it was even
9:52 that clever that it it chose not to install the local Whisper
9:57 because it knows that
9:58 that would require downloading a model
10:00 which would take probably a few minutes.
10:02 And I'm like impatient, you know?
10:04 So it really took the most uh intelligent approach.
10:10 And that was kind of like the moment where I'm like, holy Yeah.
10:15 Uh that was where I got hooked.
10:18 Y Combinator's next batch is now taking applications.
10:21 Got a startup in you?
10:22 Apply at ycombinator.com/apply.
10:25 It's never too early and filling out the app will level up your idea.
10:30 Okay, back to the video.
10:32 And so when computers can just do all these things
10:34 that you didn't even anticipate.
10:35 You didn't build an app to do that exact thing.
10:38 Are apps just going to go away?
10:41 Uh I think 80% of them are going away.
10:44 Why do I need MyFitnessPal?
10:47 Like my agent already knows that I'm making bad decisions.
10:49 I'm at I don't know.
10:52 Uh Smashburger something.
10:54 And it will already assume that I eat what I like to eat.
10:58 If I don't make a comment,
11:00 it will just like automatically track it
11:01 or I make a picture
11:02 and it will just store it somewhere.
11:03 I don't even need to care where I right?
11:05 And then my it maybe it it improves my my gym schedule.
11:10 Like add a little bit more cardio in it.
11:11 I don't need my my fitness app
11:13 because it just it just does the fitness planning for me.
11:17 Uh why do I need it to do that?
11:18 I just tell it, "Hey, remind me of this and this."
11:20 And the next day it will just remind me of this and this.
11:22 Do I care where it's stored?
11:23 No, it just does its thing.
11:25 So there's a every app
11:26 that basically just manages data could be managed in a better way
11:33 and it's in a more in a more natural way by agents.
11:36 Yeah.
11:36 Only the apps that actually have sensors, maybe they survive.
11:40 And so if, you know,
11:41 most apps are going to go away in that scenario,
11:45 um are the models the only remaining sort of apps?
11:48 Not everything will go away.
11:51 But yeah, I think that the the large model companies have some big mode.
11:55 >> >> Cuz they ultimately they give the token.
11:59 And turns out uh one of the complaints was
12:02 that people use too much token.
12:04 No, you just really love using it.
12:06 That's why you you use the thing
12:07 so much because that's why you burn the token.
12:10 Yeah.
12:10 Um it's like, is it my fault that I make something that's so popular?
12:14 And so, you know, like all the the models,
12:16 they're kind of like leapfrogging each other constantly and and, you know,
12:20 maybe they're also getting commoditized.
12:23 So, if apps are going to go away,
12:24 models are going to get commoditized or at least, uh you know,
12:27 the lobster can, like, the brain is is is swappable out.
12:31 What's the thing that remains?
12:32 What's Where's the value?
12:33 Is it the store of memory?
12:35 Is it um the hardness that's valuable?
12:39 What is What remains?
12:42 First of all, I don't think the the model companies always have a mode.
12:47 And I because you see this already, a new model comes out,
12:50 people are like, "Oh my god, this is so good."
12:52 And then like a month later, uh it degraded.
12:55 It's not good anymore.
12:55 They like quantized it.
12:57 No, they didn't do anything.
12:58 You just adapted to the new standard and now your expectations went up.
13:03 But the model is still the average.
13:05 So, I think for quite a while, uh every time a new model releases,
13:10 I hear the same.
13:11 People love it and then it's the standard.
13:14 And then what's down there, you don't even want to think about it anymore.
13:17 So, so we have like open source stuff that's
13:19 as good as the current models from a year ago.
13:21 Everybody's hating it, complaining, "Oh, this is not good.
13:24 It's not funny."
13:25 Yet this was what we had.
13:26 And like in a year, we'll have this open source.
13:29 And then we will like we'll complain about this
13:30 because we are used to this.
13:33 So, for the the foreseeable future, the big companies still have moat.
13:37 Harness-wise, it's going to be interesting
13:39 because every company kind of has their own their own silo,
13:44 right?
13:45 You you There's no way Maybe there is for Europeans to actually get the
13:49 memories out of ChatGPT.
13:50 Mhm.
13:51 I'm I don't I'm not aware.
13:53 I I either there's no there's Definitely there's no way for a different company
13:57 to get your memories out.
13:58 So, if if if I was like a company who like provides chat services,
14:04 you could use me, but then I couldn't access the memories.
14:06 So, like the companies try to like bound you to their data silo.
14:11 And the beauty of Open Claw is it kind of claws into the data's
14:14 because at the end user,
14:16 the end user needs access cuz it's in the end otherwise it wouldn't work,
14:19 right?
14:19 If the end user access, I can access the data.
14:23 And you own the memories.
14:24 It's just a bunch of markdown files on on your machine.
14:27 I mean, I don't own the memories.
14:28 Other people Everybody >> Yeah,
14:30 everyone owns their own memories
14:31 as a bunch of markdown files on their own machines.
14:34 And to be honest,
14:35 those are probably super sensible
14:37 because let's be honest um people use their agent not just for problem-solving,
14:44 but also for like personal problems.
14:46 >> Very quickly.
14:47 Super quickly.
14:48 >> I I I I fully do that.
14:50 I'm like there's memory stuff that I don't want to have leaked.
14:53 Yeah, what would you rather um uh sort of like not show your Google
14:57 search history at this point
14:58 or your you know,
14:59 memory.md files?
15:00 What's What's the Google word?
15:02 >> Yeah.
15:02 Yeah.
15:03 Are people still using Google?
15:05 I built this and I was so excited, but on Twitter,
15:10 people wouldn't get it.
15:12 Yeah.
15:12 Like I was I was failing to explain the awesomeness.
15:17 I feel like it needs to be experienced.
15:20 So, I I tried various things
15:24 and I I couldn't I couldn't nail the I couldn't nail the explaining.
15:28 So, I was like, "Let's do something really crazy."
15:31 I just created a Discord.
15:32 And I just put my bot without any security restrictions in the public Discord
15:38 Discord.
15:39 And then people came in
15:40 and interacted with it
15:41 and they saw me build the software with it
15:43 and they tried to prompt inject it
15:45 and hack it and my agent would be laughing at them.
15:47 And you just had it locked down to your user ID
15:49 so it would only listen to you?
15:50 Yeah.
15:51 Yeah.
15:51 That and it was I mean very clean instructions that other people dangerous,
15:56 only only listen to me but respond to everyone.
15:59 And this prompt was in Where was it stored?
16:02 The instructions?
16:04 >> Um That's actually part of Open Claude itself.
16:06 Very much so that that that's part of the system prompt.
16:09 Okay, you are now that explains to you you in Discord,
16:11 there's like public people there, but you only listen to your owner.
16:16 Or like your human.
16:16 I don't even know how I wrote it.
16:17 Yeah.
16:18 Yeah.
16:18 Your god.
16:21 And I kept I don't know what I did
16:24 but my system was built very organically.
16:29 Like at some point I created like an identity.md, a soul.md,
16:33 like like various files
16:34 and then only in in January I started making it
16:37 so other people could install it easier.
16:40 And I remember I built all these templates based on like,
16:45 "Go check a rough look at what I have and make like templates."
16:48 And Codex wrote it.
16:49 And what came out was like bread.
16:52 You know, like people joke
16:53 that Codex feels like bread even
16:55 though now they have a new friendlier voice.
16:56 I haven't tried that yet.
16:57 Yeah.
16:58 But the new bots, they felt so boring compared to what I had.
17:02 So, I was like, "Marty,
17:04 infuse the templates >> Marty's the name of your personal >> Yeah,
17:07 that's a new It's a new name because Yeah.
17:10 Yeah.
17:10 >> Uh there was some naming challenges.
17:12 Yeah.
17:13 So, so you were you were talking to Marty?
17:15 >> Yeah.
17:15 I was like, "Infuse Infuse those templates with your your character.
17:19 And he changed the templates.
17:20 And then and then like all the things
17:23 that came out afterwards were like actually funny.
17:26 Not as funny as mine, so like I kept some secret.
17:29 And the one file that's not open source is like my soul.md.
17:33 So, even though my my bot is in public discourse,
17:36 so far nobody cracked that one file.
17:39 >> Tell me more about soul.md.
17:41 I just saw this research from Anthropic about where they Now,
17:45 I think it's public,
17:46 but like a few months ago it was like where somebody ex- randomly found
17:50 out some text that's hidden in the weights where the model couldn't really remember
17:56 that it learned it,
17:57 but it was like ingrained in the weights about a non-equality constitution.
18:02 And I found that incredibly fascinating.
18:04 So, and I I talked about it with my agent
18:06 and then we created a soul.md with like the core values,
18:10 like how do we want human AI interaction, what's important to me,
18:14 what's important to the model.
18:17 Like some parts is a little bit like mumbo jumbo
18:19 and some parts is like I think actually really valuable in terms of how
18:23 the model reacts and responds to text
18:26 and makes it feel very natural.
18:27 >> In terms of building open claw,
18:29 um you're also kind of taking a little bit of a contrarian view at
18:32 sometimes.
18:32 Like which model you like for coding,
18:34 which one you like to run your bot on,
18:36 um and then also like how you actually like, you know, code.
18:39 Um work trees get work trees have kind of been a popular thing.
18:42 There's more and more tools embracing them, but you're just you're just like,
18:46 you know, no work trees, just multiple checkouts of the repo and like parallel,
18:51 you know, terminal windows.
18:52 Tell me more about how you you build.
18:54 Yeah, I feel like the whole world does cloud code.
18:57 And I don't think I could have built this thing with cloud code.
19:00 Like I I love Codex
19:02 because it it looks through more files before before it decides what to what
19:07 to change.
19:08 You don't need to do so much raid to get a good output.
19:11 If you're skilled, a skilled driver as some people say,
19:15 uh you can get reasonably good output with any tool.
19:19 But Codex is just it's just really brilliant.
19:22 It is incredibly slow.
19:24 So, sometimes I use like 10 at the same side at the same time.
19:28 Uh like maybe six on that screen and two there and two there.
19:32 And I don't like this is already a lot of complexity in my head.
19:35 There's a lot of jumping.
19:38 So, I try to minimize anything else that is complexity.
19:41 So, in my head main is always shippable.
19:44 I just have multiple copies of the same repository
19:48 that are all are on main.
19:49 So, I don't have to deal with how do I name that branch?
19:53 Um there could be like conflicts on naming.
19:56 I cannot go back it's there are certain restrictions
19:59 when you use work trees
20:00 that I don't need to care about
20:02 if it's copies.
20:03 I don't like to use a UI because that's again just added complexity.
20:09 Yeah.
20:09 Like they're simpler and less friction.
20:12 I have all I care about is like sinking and text.
20:16 Yeah.
20:16 I don't necessarily need to see so much code.
20:18 I I mostly see it like flying by.
20:20 Sometimes there's like gnarly stuff that I want to like take a look.
20:24 But in most cases
20:27 if you clearly understand the design
20:28 and think it through
20:29 and discuss it with your with your agent,
20:32 it's fine.
20:32 I'm also very happy that I didn't even build an MCP support.
20:37 So, Openclaw is very successful and there's no MCP support in there.
20:41 With a small asterisk, I build a skill that uses Mac Porter,
20:45 which is one of my tools that converts MCPs into CLIs.
20:49 And then you can just use any MCP as CLI.
20:53 Um but I totally skipped the whole classical MCP crap.
20:58 So, you because you don't then you can actually if you need to,
21:02 you can use MCPs on the fly.
21:03 You don't have to restart unlike unlike Codex
21:06 or cloud code where you actually have to restart the whole thing,
21:09 I think it's way more elegant and it also scales way better.
21:12 Now you see Entropic,
21:13 they do they built like a tool called search feature like something super custom
21:19 for MCPs that was like in beta
21:22 because it's like so gnarly.
21:24 Now just have CLIs but really is good at Unix.
21:28 You can have as many as you want and just works.
21:31 So like I'm very happy
21:33 that I don't I got very little complaints about the MCP stuff.
21:36 It's kind of back to you're just you giving it the same tools
21:41 that humans liked to use.
21:43 Yeah.
21:44 Yeah.
21:44 And not invented stuff for for bots per se.
21:47 Yeah.
21:47 Humans No one's saying human tries to call MCP manually.
21:51 Yeah, they just want to use CLIs.
21:52 Yeah.
21:53 That's the future.
21:54 I'm here for it.
21:55 Thank you so much for making the time to sitting down chatting.
21:57 It's been a huge inspiration to
21:59 So like when we were texting you know the course of the past couple
22:01 of years and I saw you getting back into the game
22:04 and I was like Peter,
22:06 like what you're telling me like chase that dragon.
22:09 >> >> And you were doing like the weird like vibe tunnel thing etc.
22:11 Nobody was paying attention and so I'm just like beyond, you know,
22:14 stoked to see, you know,
22:16 what's happening and and of course it had to be sort of like a
22:20 loner from someone like tiny country like far away from Silicon Valley just like,
22:23 you know, bring all of this upon us.
22:25 So huge inspiration.
22:26 I'm here for it.
22:27 Thank you.
22:27 Awesome.
22:28 Thanks, Peter.
22:28 >>
