Mark Zuckerberg needed ten minutes. He was finishing code.
Peter Steinberger had called him on WhatsApp without scheduling anything. "I don't like calendar entries," he told Lex Fridman last week. "Let's just call now." Zuckerberg asked for a brief pause, then picked up. The first ten minutes devolved into an argument about whether Claude Code or Codex was the better programming tool. The CEO of a trillion-dollar company squabbling with a solo developer from Vienna over IDE preferences.
That was two weeks ago. Zuckerberg ran OpenClaw on his own machine afterward. Gave feedback that was blunt and specific, calling features "great" or "shit" in real time. Used it until it broke, then sent notes on what to fix. Steinberger called it "the biggest compliment" because "it shows they actually care about it."
On Sunday, Steinberger announced he's joining OpenAI. Not Meta. Sam Altman posted within hours that Steinberger would "drive the next generation of personal agents." Greg Brockman and Fidji Simo both posted within the hour. Three executives in a coordinated burst of enthusiasm that OpenAI hasn't shown for a hire since the $6.4 billion Jony Ive acquisition last year.
Every outlet is covering this as a talent acquisition win. The framing is wrong. OpenAI already employs thousands of good engineers. What it couldn't build internally is proof that ordinary people will hand an AI agent full access to their digital lives without hesitation.
What $10,000 a month buys at the negotiating table
The Breakdown
• OpenAI hired OpenClaw creator Peter Steinberger; both Meta and OpenAI reportedly made offers valued in the billions
• OpenClaw's 300,000+ line codebase was never the prize; 196,000 GitHub stars and 2 million weekly visitors were
• Steinberger publicly rejects AGI but chose OpenAI's distribution over Meta's personal attention from Zuckerberg
• Anthropic's models powered OpenClaw, but the framework's creator now works for the competition
Steinberger has been losing money on OpenClaw since November. Ten to twenty thousand dollars a month, by his own count. He routes sponsorship revenue to the developers who maintain his dependencies rather than keeping it. The project hit 196,000 GitHub stars and pulled 2 million visitors in a single week while its creator subsidized everything from savings.
He didn't need their money. He spent 13 years building PSPDFKit into a PDF tools company worth over $100 million before selling to Insight Partners. Three years of what he described as soul-searching followed. Therapy. Ayahuasca. Forty-three failed projects. Then OpenClaw caught fire.
His negotiating edge came from something no check could replicate. Both Meta and OpenAI made concrete offers, reportedly valued in the billions. VCs lined up. Steinberger told Fridman he doesn't care. "I don't give a fuck" were his exact words. When you've already sold a company and your next project goes viral by accident, the dynamics flip completely. He wasn't selling. They were auditioning.
The code nobody wanted
If you think these companies wanted OpenClaw's codebase, look at the inventory. Somewhere north of 300,000 lines of code, nobody is sure exactly how many. Unaudited. Developer Gavriel Cohen evaluated it for NanoClaw, found it too bloated for any security team to review properly, and rebuilt the core logic in roughly 500 lines of TypeScript. His team audited the entire replacement system in an afternoon.
The security record is just as ugly. Moltbook leaked 1.5 million API keys to the open internet. ClawHub, OpenClaw's skills marketplace, hosted 335 packages distributing Atomic Stealer malware to Mac users. Three days after Moltbook launched, RentAHuman.ai went live, a marketplace where OpenClaw agents hire real humans for physical tasks. Forty thousand people signed up to take orders from bots. Payment in stablecoins.
No serious engineer looks at that inventory and says ship it. What Meta and OpenAI wanted is the adoption graph. Six hundred contributors. Ten thousand commits. All in under three months. The WhatsApp integration dropped an AI agent into a messaging app 3 billion people carry in their pockets. OpenClaw became the first consumer brand in AI agents without spending a dollar on marketing.
OpenClaw's momentum even bent hardware markets. Tom's Hardware reported that Mac delivery times for high-memory configurations stretched to six weeks, driven partly by users buying machines to run local AI agents. Apple CEO Tim Cook acknowledged the company was chasing memory supply. One solo developer's weekend project created enough demand to disrupt Apple's supply chain.
You can rewrite 400,000 lines of messy code in a quarter. You cannot fabricate that kind of pull. That distinction, according to the bidding behavior of two of the most powerful companies in technology, is worth billions.
Specialized intelligence inside the AGI machine
The deal gets uncomfortable when you look at what Steinberger actually believes.
He told a Y Combinator podcast this month that AGI is the wrong goal. "What can one human being actually achieve? Do you think one human being could make an iPhone or one human being could go to space?" he said. "As a group we specialize, as a larger society we specialize even more." His vision runs on specialized agents collaborating. Not one god model that handles everything.
OpenAI's entire corporate identity rests on achieving artificial general intelligence. The name says it. So does the $500 billion valuation. Steinberger just joined a company whose stated mission he publicly rejects.
Ignore the ideology for a second. The logistics explain everything. Steinberger does not want to run a company. Thirteen years of it was enough. He wants to build agents everyone can use. That means compute, APIs, and 300 million people already opening ChatGPT every week. This is pragmatism dressed as alignment.
Meta offered something different. More personal, in fact. Zuckerberg's hands-on engagement impressed Steinberger. Steinberger recounted it on Fridman's show, acting out the reactions. "Mark basically, 'Oh, this is great. Oh, this is shit. Oh, it needs to change this.'" He noted the contrast with OpenAI. "I didn't get the same on the OpenAI side."
Stay ahead of the curve
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
He chose OpenAI anyway. Zuckerberg codes, gives real feedback, and clearly cares. None of that ships a product to hundreds of millions of users. For a builder who wants reach without management overhead, that settled it.
Chrome won. Chromium exists.
Steinberger drew the comparison himself. OpenClaw would follow the Chrome and Chromium model. A foundation to hold the open-source project. A corporate partner to build the commercial version. "I think this is too important to just give to a company and make it theirs," he told Fridman.
He's describing a pattern with a known ending.
Google open-sourced Chromium. Chrome is what everyone downloads. Android AOSP is open. Google's Android with Play Services runs the planet. MySQL went to Oracle. The community forked it into MariaDB. MariaDB survives. MySQL still owns the market.
Every foundation arrangement in tech follows the same gravitational pull. The corporate version gets full-time engineers, marketing, distribution, and the daily attention of the person who created the project. The open-source twin gets volunteers and good intentions. Gravity always wins.
Altman committed in public. "OpenClaw will live in a foundation as an open source project that OpenAI will continue to support." Those words are sincere until they collide with product timelines. When OpenAI ships a consumer agent built on Steinberger's ideas, using OpenAI's models, and that product competes with what the foundation maintains, the foundation will not get the best features first. OpenAI has shareholders. Shareholders beat charters every time.
Steinberger is too experienced to miss this. Thirteen years running a company taught him what public commitments look like right before they erode. The foundation language gives both sides something comfortable to say in public. It hands the community a story to hold onto. And it gives OpenAI a grace period before anyone asks which version is getting Steinberger's best hours.
Anthropic built the runway. OpenAI caught the flight.
Nobody is talking about who this hurts most.
OpenClaw launched on Claude. Steinberger called himself "the biggest unpaid Codex advertisement show," but the framework's recommended setup pointed users to "Anthropic Pro/Max plus Opus 4.5 for long-context strength and better prompt-injection resistance." Thousands of OpenClaw users routed heavy agent workloads through Anthropic's $20 consumer subscriptions.
Then Anthropic pushed back. Steinberger said in a recent interview that Anthropic "doesn't like it anymore." He recommended API keys instead. The economics made sense from Anthropic's side. A subscription designed for individual chat sessions was never priced for autonomous agents burning through tokens at 3 AM while their owner slept in the next room.
But the result is a category disaster. Anthropic's models powered the most visible agent framework in history. Its subscriptions funded the consumer proof of concept. And now the creator of that framework works for a competitor. He announced the move on a weekend while Anthropic's $380 billion valuation round closed the same week.
OpenClaw users will keep running Claude. The models are good. But the face of the agent movement now works at a rival lab, and every keynote, podcast, and product launch will carry OpenAI's logo behind him.
Anthropic's discomfort won't be loud. It will be the quiet kind. The kind where you helped prove that agent interfaces matter more than models, then watched someone gift-wrap that proof for the competition. At $380 billion, Anthropic can absorb the sting. What it cannot do is un-train a community that learned to think of AI agents through OpenClaw's interface, an interface now housed inside OpenAI's strategy.
The gravity starts now
Zuckerberg is still coding at Meta. He'll build agents regardless. Google has the largest mobile distribution surface on earth through Android and has done nothing visible with consumer agents. Apple has iMessage, Siri, and 1.5 billion active devices. Silence.
The foundation will publish its charter. Volunteers will submit pull requests. Steinberger will show up at community events and champion the open-source spirit he clearly values. He means it. That sincerity is real.
And slowly, because this is how gravity works in tech, the best ideas and the best engineering hours will flow toward OpenAI's agent products. Steinberger spends his days there now. So will his attention. So will the features that matter.
Chrome won. Chromium exists. The claw is the law, until corporate gravity says otherwise.
Frequently Asked Questions
Q: Why did Steinberger choose OpenAI over Meta?
A: Despite Zuckerberg's hands-on engagement, Steinberger chose OpenAI for its distribution infrastructure. OpenAI has 300 million weekly ChatGPT users and can ship agent products at scale. Steinberger wanted reach without management overhead.
Q: What happens to OpenClaw as an open-source project?
A: Steinberger and Altman committed to a foundation model where OpenClaw remains open-source while OpenAI builds a commercial version. The pattern mirrors Chrome and Chromium, where the corporate version historically absorbs the best features and talent.
Q: How much was the Steinberger acquisition reportedly worth?
A: Both Meta and OpenAI made offers reportedly valued in the billions. VCs also lined up with proposals. Steinberger, who previously sold PSPDFKit for over $100 million, said he did not care about the money.
Q: What security issues has OpenClaw faced?
A: Moltbook leaked 1.5 million API keys. ClawHub hosted 335 packages distributing Atomic Stealer malware to Mac users. RentAHuman.ai launched days later, letting AI agents hire humans for physical tasks with stablecoin payment.
Q: How does the Steinberger hire affect Anthropic?
A: OpenClaw launched on Claude and recommended Anthropic subscriptions for agent workloads. Thousands of users routed heavy tasks through Anthropic's $20 plans. The framework's creator now works at a rival lab during Anthropic's $380 billion valuation round.
