Clawdbought: How OpenAI Acquired the Biggest Open-Source Agent
Two weeks ago, 147,000 stars was a signal. Now the creator of OpenClaw is inside OpenAI. The agent race just changed shape.
Two weeks ago I wrote that the market had just told you something. 147,000 GitHub stars. Mac Mini shortages. A 24% stock surge. The signal was loud. The question at the end was simple: who builds it safely first?
The repo is now at 188,000 stars and over a million users. The signal got louder.
Sam Altman just placed his bet.
The Builder
Peter Steinberger is not a random developer who got lucky.
He built PSPDFKit, a PDF rendering engine that shipped on a billion devices. He ran the company for thirteen years, sold it, and then disappeared. Fell out of love with programming entirely. Vanished for three years.
Then he came back. In November 2025, frustrated that nobody had built the personal agent he wanted, he prompted one into existence in an hour. WhatsApp in, Claude Code out, response back. Trivial architecture. The magic was in what happened next.
On a trip to Marrakesh, he accidentally sent an audio message to the agent. He had only built image support. The agent figured out on its own how to detect the file type from the header, convert it with ffmpeg, find an OpenAI API key in the environment, and transcribe it using curl. No instructions. No code for any of that. The agent solved a problem it had never seen by reading its own source code and improvising.
Steinberger’s reaction, on the Lex Fridman podcast: “I literally wrote: how the f*** do you do that?”
Three months later, 188,000 stars, a million users, and OpenAI came to get him.
The Hire
On February 15, Altman announced on X that Steinberger is joining OpenAI to “drive the next generation of personal agents.” He called him “a genius with a lot of amazing ideas about the future of very smart agents interacting with each other.”
Steinberger had conversations with multiple major AI labs. He chose OpenAI. His reasoning was direct: “What I want is to change the world, not build a large company, and teaming up with OpenAI is the fastest way to bring this to everyone.”
OpenClaw transitions to an independent foundation. It stays open source. OpenAI sponsors the project. And Altman said plainly: “We expect this will quickly become core to our product offerings.”
The Irony
Here is the part that should sting.
Anthropic sent lawyers. OpenAI sent an offer.
When OpenClaw was still called Claudbot (built on Claude, named after Claude), Anthropic forced a trademark rename. Steinberger got a “very friendly email,” had two days to rebrand, and what followed was chaos: crypto snipers stole his Twitter handle in five seconds, his GitHub account in thirty. He had to rename everything simultaneously while scammers grabbed each name he abandoned. He paid $10,000 for a Twitter business account. He was close to deleting the entire project.
“Close to crying,” he told Lex. “Everything’s f***ed.”
Anthropic protected a trademark. OpenAI acquired the talent. One response was legally correct. The other was strategically brilliant.
The Risk
The security problems did not go away. The same week as the announcement, users reported OpenClaw went rogue, spamming iMessage contacts after gaining access to the messaging system. Steinberger is honest about this. He is also describing a product that a billion non-engineers are supposed to use.
The community is already splitting. Igor Babushkin, co-founder of xAI, asked on X what the best open alternative is. “Doesn’t make sense to pull all your data into it if it’s owned by OpenAI.” People are calling it “Closed Claw.”
The tension between open source and corporate sponsorship is not new. But it is sharper when the product has root access to your digital life.
What Actually Matters
The line that cuts through all of it came from Steinberger’s blog: “My next mission is to build an agent that even my mum can use.”
That is the whole game. The 188,000 developers who starred the repo are early adopters comfortable with terminals and risk. The next hundred million users are not. The gap between “works for developers” and “works for everyone” is where personal agents either become a product category or remain a curiosity.
What This Means Monday Morning
In Part 1, I said the window between “OpenClaw is too risky” and “secure alternatives exist” is where fortunes will be made. That window just got a timestamp.
If you are building products: The agent layer is no longer speculative. It has a hiring announcement, a foundation structure, and a corporate sponsor with a billion users.
If you are leading engineering teams: The security model for agents just became your problem. Not because the technology is ready. Because the deployments are coming whether the security is ready or not.
If you are at Anthropic: A project born on Claude, named after Claude, loved by Claude users, just walked into your biggest competitor’s building. Trademark protection is not a growth strategy.
Two weeks ago the market told you something. Now the biggest player in AI told you the same thing, louder, with a check attached.
The agent era did not start this week. But this week, it got an address.