OpenClaw Could Be 1st 1-Person $1B Company, OpenAI Buys  ## Document Notes ## Highlights - **Solo Project Proved Agents Can Do Real Work** - OpenClaw showed that a solo, vibe-coded agent project can rapidly demonstrate utility and viral adoption among developers. - The project's ability to fully control a user's computer marked a practical step toward personal agents that actually do tasks. Transcript: Jaeden Schafer 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 Clawd 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're kind of buying these fresh ones, sticking this open claw on it, it Was running on there. And I mean, completely tasked 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 is kind of a huge, huge moment. ([Time 0:01:07](https://share.snipd.com/snip/66c54b87-11ac-470e-a441-aaed79ab300c)) - **Viral Rise Fueled By Rebrands And Hype** - OpenClaw went viral after multiple rebrands and a cease-and-desist from Anthropic drew attention to the repo. - Developers bought machines to run OpenClaw and shared screenshots of agents automating calendars, emails, and sign-ups. Transcript: Jaeden Schafer So when OpenClaw first launched, it wasn't called OpenClaw, it was called ClawedBot. And when that happened, it wasn't clawed like, you know, Anthropics clawed. It was like a claw. Then the logo's a lobster. And of course, OpenAI's legal team sent them a cease and desist for it because it was getting a ton of traction on GitHub. Basically, everyone was talking about it. 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 Anthropoc 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, OpenCloud 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, like, emails, conversations. I mean, people were literally setting this loose. Some people were giving them its own email and sending it, giving it tasks and projects and saying, go sign up for accounts. People were sharing all sorts of crazy screenshots like, oh, my CloudBot just, you know, signed up for an account with this specific service. ([Time 0:02:40](https://share.snipd.com/snip/fdca5de9-2b87-43e3-93f9-266c70112798)) - **Model-Agnostic Wrapper Was Key** - OpenClaw succeeded because it was model-agnostic and acted as a wrapper to connect models to real-world tools. - That flexibility let users plug GPT, Claude, Gemini or others into a single agent platform. Transcript: Jaeden Schafer 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 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 Clawed, kind of sending them a cease and desist. They didn't want to get tied down to that. So you can plug ChatGPT 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. ([Time 0:07:13](https://share.snipd.com/snip/05777db0-cecb-4524-9109-5e1c3e35b84c)) - **Treat External Inputs As Potential Attacks** - Do not trust external inputs without guardrails because prompt injection can make agents leak secrets or take harmful actions. - Implement strict validation and restrict irreversible actions when agents have broad system access. Transcript: Jaeden Schafer 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 Moldbook 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 give 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 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 workspace with access to everything. Someone, you know, emails 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 downloads malware onto your computer. ([Time 0:09:38](https://share.snipd.com/snip/37387bc2-3260-4605-8358-918941550ed0)) - **Founder Chose Product Over Building A Big Company** - Peter Steinberger built OpenClaw as a weekend, vibe-coded solo project and never raised funding or formed a team. - He previously sold a company and preferred joining OpenAI over scaling OpenClaw into a big startup. Transcript: Jaeden Schafer 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. Meta's 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're gonna give you compute, we can give you infrastructure, we can give you access to basically all of the latest models and research. So some models that hadn't been released yet, you 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 showed him some of their quote unquote, cool projects that are being, you 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 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 grab 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 ChaiGVT 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 OpenClaw into a massive company, but that building a big company wasn't something that's very exciting. I believe his last company he sold was like a hundred million dollars. So, uh, I think, or a hundred 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. ([Time 0:11:39](https://share.snipd.com/snip/3a2a5778-c32f-4755-96d7-45497b4d7343)) - **Acquisition Kept The Project Open** - OpenAI acquired the founder and will maintain OpenClaw as an open-source project under a foundation. - This preserves community contributions while giving OpenClaw access to OpenAI's resources and models. Transcript: Jaeden Schafer 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 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. ([Time 0:12:54](https://share.snipd.com/snip/6de14a5e-3a97-46aa-96a7-b26e085d5a70)) - **Large Tech Races To Integrate Agents** - Big companies like Meta and OpenAI view agent tooling as strategically important and are building similar features into their products. - OpenClaw's popularity accelerated rivals to develop integrated agent features for messaging and developer tools. Transcript: Jaeden Schafer So they didn't actually end up acquiring this, but I think they just kind of built their own version as well. It's kind of the classic meta way to do things. I think this is really interesting. For now, OpenClaw is going to be, 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 and power it. And with all of the compute and resources of OpenAI, I think it's going to be something that is incredibly useful. ([Time 0:14:45](https://share.snipd.com/snip/e5b93d10-2c50-40a5-9059-f01e57c72ce6)) - **Scale Code Review For Rapid Open Source Growth** - Accept community contributions but set clear review processes because the repo flooded with thousands of PRs. - Prioritize manual review for high-risk changes and scale reviewer capacity to maintain security and quality. Transcript: Jaeden Schafer I mean, I'm still seeing him on X saying, okay, like he said something like he had 3,800 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, he said he's like reviewing 600 manually a day or something crazy like that. And he's got a lot to go. ([Time 0:15:42](https://share.snipd.com/snip/5737b182-f7e3-4dec-aaa1-1b51d94b8e97))