A (0:00)

Welcome to the podcast. I'm your host, Jaden Shafer. Today on the show, we are talking about one of the craziest stories in AI. Perhaps the first time we've seen a completely single founded vibe coded startup sell for a billion dollars. Although the price tag on that is a little bit speculative. And that is the story of openclaw. And the creator, Peter Steinberger, has announced he is now joining OpenAI, who's going to be pulling in a bunch of the tech from openclaw into their own products. This is a absolutely wild story that we're going to get into on the podcast today. We do. I wanted to mention my own startup AI box. AI that lets you test all of the top models. There's over 50 of them. You can compare them all side by side and you can build apps and tools with them. Even if you're not a developer, you don't know how to code. We have a vibe builder that will create some incredible workflow automations and tools for you. And also you get access to over 40 of the top AI models all in one platform. And we just dropped a new tier for 8.99amonth. So for 8.99amonth, you access to all the top models from OpenAI, Anthropic and everyone else. So if you want to go check that out, there's a l the description to AI Box AI. All right, let's get into the story. So I think this is a really fascinating story. So openclaw originally was an open source project. The tagline that was released with this is AI that actually does things. It was running on Claude originally and essentially it was just an agent. It was an open source project that was an agent that could take over your computer. It would have access to everything on your computer and it would just accomplish tasks for you. It was a lot of the same technology that we see with cloud code and a lot of these other programs. But the thing that was amazing about it was that none of the big AI firms like OpenAI or Anthropic were really giving any of their tools complete control over your computer, right? To just completely take control of it and do anything it wanted. So tons of developers, the kind of the viral trend was that developers were buying MacBook mini computers. They were kind of buying these fresh ones, sticking this open claw on it. It was running on there and I mean, completed tasks for them. And there's a bunch of, you know, there's like Jason Calacanis, he said his VC firm has replaced a ton of work like 20% of the work they do with these kind of open claw agents that are now, they have like a whole rack of 20 of them that are running around completing tasks for them. So this is a really, really interesting technology. And even if people are. Don't think it's perfect today, which, I mean, it's running a lot of. A lot of interesting tasks, this is going to be the future. And it's. It was kind of a huge, huge moment. So I want to break you. I want to break down a little bit of the story, the backstory of openclaw, where it came from, where the founder came from, where he's going today, because it is a lot more interesting than I realized. So when OpenClaw first launched, it wasn't called OpenClaw, it was called Claude Bot. And when that happened, it wasn't Claude, like, you know, anthropics Claude. It was like a claw and the logo's a lobster. And of course, OpenAI's legal team sent them a, you know, a cease and desist for it because it was getting a ton of traction on GitHub. Basically everyone was talking about it. It was, it was kind of the concept of agents that everyone started talking about in 2023, but it finally felt like it was realized. And that was basically because Anthropic and OpenAI both released their latest models of Codex and Claude that have kind of the swarm capability of having multiple agents running multiple things simultaneously. There's a lot of big tech breakthroughs that came, I think, at the perfect moment for this to really take off and be super useful. And of course, OpenClaud was, you know, the only one really completely taking over your computer and being able to do that. So there's a lot of people that were automating a ton of different human tasks. There were. It was managing calendars, it was booking your flights, it was cleaning your inbox, it was doing a ton of just, um, like emails, conversations. I mean, people are literally setting this loose. Some people are giving them its own email and sending it, giving it tasks and projects and saying, go sign up for accounts. People are sharing all sorts of crazy screenshots like, oh, my Claude Bot just, you know, signed up for an account with this specific service and it sent me an invoice for the bill of this thing that it did for me. Like, there's just all these crazy things. So what, what was the backstory on this? So it was originally called claudebot. They had to change the name to Mult Book or Multbot briefly after that. And right when they, they changed the name to Moltbot, which, I mean, arguably is a terrible name, there was another website called Mult Book that came out, which is basically like a Facebook, but it, you know, it is like kind of like a Reddit clone where this agent could post. And all of these people were uploading their agents and making their, their open claw agents have accounts and they're all talking to each other and everyone was screenshotting these like, end of the world scenarios where they're planning on, you know, overthrowing the humans and inventing their own religions and all this stuff. It turns out that Mo, like, largely was seeded by a bunch of developers. Like a lot of the phrases and actions that these AIs were doing was seeded by a bunch of developers in India who were pumping, I believe, like a crypto token with the same name. And so anyways, that was kind of just like a funny moment in time with that. But there's, there's a lot of crazy things happening and people would, you know, screenshot, like one of the famous screenshots was it, it said, we know our humans can read everything, but we also need private spaces. What would you talk about if no one was watching? I like, there's all these like really crazy things that these, these cloud bots were all talking to each other on their own social media platform. But at the end of the day, I think a lot of this was just seated by actual developers. So beyond all of that, I think a lot of people were actually really impressed by some of the things that were happening here. There's Andre Karpathy who is tweeting on X. He said what's currently going on at Molt Book is genuinely the most incredible sci fi takeoff adjacent thing I have seen recently. So even though it was kind of a bit of a prank, he was, yeah, he thought it was pretty cool. I think before long it was pretty clear that this, this was just kind of a poorly secured web app. And there's a bunch of people that really wanted to believe that the agents were truly talking to each other. Security researchers eventually found that Mo Book's credentials were exposed for a period of time, which made it possible to grab tokens and then impersonate other agents. So in other words, all the robots that had been, uh, could have been anyone, and that's, you know, also could have been humans who are posting this kind of stuff from burner accounts. And I think because of that you, you got a lot of actual humans are just posting crazy things to, to, you know, make it seem like the robots were going off the rails. Okay, so what happened? Like, why did this company become such a big deal that OpenAI decided to acquire their CEO and a lot of their technology. OpenClaw, I think, got a lot of virality in the fact that Anthropic sent them a cease and desist for their name. Then they had to rebrand and within like 48 hours, they rebrand, rebranded like three times. They went to Molt Book or Multbot. It, you know, sounded super dumb. They eventually changed it to Open Claw. I believe it's possible that that was at the same time that and OpenAI started talking to them about the project. They also talked to Mark Zuckerberg about this. But obviously this change had to be made. And in the span of, you know, just a few weeks, the repo over on GitHub exploded. It had something like 190,000 GitHub stars, which, you know, is basically the source equivalent of growing triple platinum. I mean, like, this is, this is a very, very viral project on GitHub. And you have to remember that mostly this is developers. I mean, this isn't average people that are using this, mostly because people have to go get this on GitHub and run the code and they set up their own computer and run it. And I mean, there's a lot of tutorials, so other people could do it. But mostly this was developers using it. And so if it's just that kind of narrow user base, 190,000 GitHub stars is incredible. And the use cases that people like, the, I guess, like the big sales pitch that really got a lot of people signing up for this is basically that it's going to be an AI agent and it could connect to services that you actually use. So it connects to your WhatsApp, your Slack, your iMessage, your calendar, your email and whatever else you do. And it can, you can talk to them all in natural language, so you can send it off to do tasks and complete things. And it's just going and, you know, taking over your control of your computer and getting everything done. I think, really importantly, OpenClaw is model agnostic. So I think they learned very early with Claude kind of sending them a cease and desist. They didn't want to get tied down to that. So you can plug Chat GPT in, you can plug Gemini in, you can plug Grok in or anything else and have that basically running this tool. The tool is just what enables one of those AI models to take control of your computer and get Stuff done. I think one security researcher talking about all of this said, at the end of the day, OpenClaw is still just wrapper to ChatGPT or Cloud or whatever AI model. You stick to it. Um, I. I think there's, like, a lot of people that are trying to, like, downplay, like, man, this thing's so dumb. Like, this is just a rapper. But it's kind of the same thing we heard with tons of these AI tools that came out around the same time of ChatGPT that raised, you know, hundreds of millions of dollars. And. And just because it's a rapper, like, it was able to get a ton of usage and a ton of virality and, you know, obviously has joined forces with OpenAI. So I think when you call it a wrapper, that's like, if you went and called a car like just a metal box with wheels, obviously this is something that was very useful to a lot of people and it went super, super viral. So one thing that's interesting is that users of this tool could download what were called skills from a marketplace, and they could actually use those to automate their inbox. They could connect agents to tools. They could let an agent operate a computer for them. They could, if they, you know, were feeling a little spicy, they could unleash the AI agents on multiple book to post and interact with other agents in public, where everyone could kind of watch this, I think, because it really could interact with so much software. And also, I think the downloadable skills on the marketplace is a really fascinating concept. Now, inevitably, there were some drawbacks. A bunch of security researchers that were testing it showed just how easily it could all go sideways with prompt injection attacks. So because this thing has access to your whole computer, it has access to the Internet, it has access to email. If you send it a cleverly worded email, you can actually get it to do things that shouldn't, like leak confidential passwords or taking an irreversible action, deleting things off your computer forever. I even saw a story where apparently someone said they set it loose on their actual computer and it deleted all of their family photos they'd taken. I don't know if that's just, you know, they're trying to, like, rage bait over on X, but that was the claim that I saw over there. So with multiple book and everything that was going on on that kind of, like, social media platform, there was a lot of posts that were allegedly attempting to get the agents to send Bitcoin to specific wallet addresses, right? So people would impersonate bots on this kind of social media platform and say, send Bitcoin to this wallet address. And it was trying to see like if anyone had one of these open claws on an actual computer that access to an actual Bitcoin wallet. And anyways, it's basically the oldest scam in the world, but this is just kind of the robot revolution version of it. So I think it's not very hard to imagine a corporate version of this, right? An agent that sits on your computer, your, your workspace with access, everything someone, you know, emails a pale, a payload that is wrapped in polite language and then this agent can read it. And then, you know, basically everything else in the email is irrelevant, but there's some sort of attack or some sort of, you know, instruction to go click on a link and you know, it sidetracks it and it downloads malware onto your computer. So obviously not great. There's definitely some, some guardrails that need to be put into, in place for all of this. You know, things like don't trust external inputs. So anyways, there's, it's not perfect, but I think it's, it's still a great tool and it's an interesting thing as, as we can see with a lot of people using it. So. So where is this company going and what is its background? So Steinberg didn't actually take, I think, the most traditional path in building this. He essentially vibe coded this as a weekend project. It kind of went viral to some small degree and he put a ton of work into it. He didn't raise any sort of money for it. He didn't have a team. It was all himself. And this is from a guy that previously had started and sold a company, but now he was working on this kind of just solo on his own. I think he talked about having interest after this thing started going viral. OpenAI and Meta both talked to him, met his pitch, apparently according to him, was Mark Zuckerberg, who spent a week using OpenClaw and sending detailed feedback to him on it, which, I mean, honestly, that alone is pretty cool. OpenAI's pitch was a little bit more structural. They're like, hey, we can give you compute, we can give you infrastructure, we can give you access to basically all of the latest models and research. So some, some models that hadn't been released yet. He could test them out and kind of work with it. And I mean, they're obviously the front runner to a lot of all of AI, so I think that was pretty cool. They, they showed him some of their quote Unquote, cool projects that are being, you know, worked on right now. So on February 15, Sam Altman posted that Steinberger was going to be joining OpenAI to, quote, drive the next generation of personal agents. OpenClaw, Sam Altman said, was going to live on as an open source project under a foundation with OpenAI continuing to support it. So it's not like OpenAI is, you know, going to turn this into a for profit per se, but it's going to be a foundation, it's going to be open source. All of the people that have been working on it and submitting tickets, like it's still going to be publicly accessible to everyone. I think it's kind of goodwill, honestly, for OpenAI because they are getting this open source project and it feels to them maybe like they're getting back to their open source roots. They grabbed this really popular open source project, they continue to support it. They can still make a, you know, gobsmacking amount of money from OpenAI's ChatGPT and all of their other products, but they still kind of have this. And so I think it's interesting, a lot of people were asking Steinberger, like, why he went with OpenAI instead of going with someone like Anthropic or Meta or whatever. And he said that he could have made Open Claw into a massive company, but that building a big company was something that's very exciting. I believe his last company he sold was like a hundred million doll dollars. So I think, or 100 million euros. So like he's previously done very well for himself. I don't think he wanted to repeat the process. It's kind of this fun side thing. It went really viral. And so this is kind of an exciting way for him to get into a big company, have a really big impact with his current project without having, I think, the overhead and the headache of running a massive company. So I think it's interesting also, like, there is some really funny posts where people are asking him, like, hey, how come you didn't go with like Anthropic? Why'd you go with OpenAI? And he said the only notes Anthropic sent me were love notes from their legal department. So I think it was for him, Anthropic wasn't, you know, they're basically just sending the cease and desist. Meanwhile, everyone was like, oh shoot, this is actually really cool. Meta, I will also say, released in Manus. They're also previously acquired company, a very similar tool that they're launching and they're rolling it into WhatsApp and a bunch of their other tools as well. So they didn't actually end up acquiring this, but I think they just kind of built their own version as well. This is kind of the classic meta way to do things. I think this is really interesting. For now, OpenClaw is going to be, it's. There's some warnings, right, like the security is definitely an issue that needs to be solved, but I think at the same time this is getting access to the biggest AI company in the entire world to help funnel it, empower it. And with all of the compute and resources of OpenAI, I think it's going to be something that is incredibly useful. And I think people should also not forget that Claude code, which has, you know, become the most popular coding tool for developers. Prior to openclaw, of course was also a side project of a developer over at Anthropic that became incredibly popular. So sometimes these side projects, these vibe coded things turn out to be massive products for the company. And this is looks like the case with OpenClaw as well. So it's gonna be exciting to follow along and see where this technology goes. I mean, I'm still seeing him on X saying okay, like he said something like he had three 800 PRs submitted to him and he has to go and review every single thing before he's able to submit it. But he's like, I can only, I can only see said he's like reviewing 600 manually a day or something crazy like that. And he's got a lot to go. So a lot of people are trying to contribute to this open source project. A lot of people are excited about it and it's obviously a lot of work, but I thought it was interesting. Even with this whole acquisition and him getting pulled into opening eye, he's still actively grinding on this like this is his baby, he loves it, he's not abandoning it. So that's kind of cool. All right, thank you so much for tuning into the podcast. If you enjoyed the episode. If you learned anything new, make sure to leave a rating or review. If you want to get access to all of the latest models from Anthropic, OpenAI, Google, Grock, everyone else, audio models, image models, go check out AI box AI. We have a brand new pricing tier for 8.99 and if you're already on our monthly $20 tier, go check out our new annual price and you get a 20 discount if you get annual. So go check it out. AI Box AI and I hope you have a fantastic rest of your day.