Anthropic's Biggest Miss of 2026 Just Happened.

Peter Steinberger built OpenClaw — the hottest AI agent on the planet right now.

180K+ GitHub stars. 3.4M views on the announcement tweet. Baidu integrating it. Used in 100+ countries. The fastest-growing GitHub repository ever by star count.

And he built it alone. One person. One lobster mascot.

Yesterday, Sam Altman announced Steinberger is joining OpenAI. The open-source project will live in a foundation. The agentic AI race just tilted.

But this story isn't about OpenAI's win. It's about Anthropic's loss.

What Just Happened

On February 15, 2026, Sam Altman posted what might be the most consequential hiring announcement of the year:

Let that sink in. The creator of the most viral AI agent in history — a project that was built on Claude, named after Claude, and recommended Claude Opus as its default model — just walked into OpenAI's front door.

And Anthropic basically held it open for him.

The Man Who Didn't Need the Money

Peter Steinberger isn't some 22-year-old founder chasing a payday. He's a serial builder from Austria who already had his exit.

He founded PSPDFKit in 2011, grew it to 70+ employees, and sold it to Insight Partners for over $100 million. He took a three-year break. Then he started building again — this time with AI at the center.

His philosophy is stark. He told The Pragmatic Engineer: "I ship code I don't read." He runs 5-10 AI agents simultaneously on different features. He views pull requests as "prompt requests." He's the purest example of what AI-native development looks like when the developer already has a decade of systems architecture experience.

OpenClaw started as a weekend hobby project. A personal AI assistant that could clear your inbox, make reservations, handle flight check-ins, and integrate with WhatsApp and Slack. Nothing fancy on paper.

It hit 9,000 stars in 24 hours. 60K in two weeks. 100K in 48 hours after going viral. It became the fastest-growing GitHub repository in history by star count — more Google searches than Claude Code or Codex combined at its peak.

And the whole time, Claude was the engine under the hood. anthropic/claude-opus-4-6 was literally the default model in the config file.

Peter Steinberger was Anthropic's biggest unpaid evangelist. Every one of those 180K+ stars was an implicit endorsement of Claude's capabilities.

How Anthropic Fumbled Their Biggest Evangelist

On January 27, 2026, while OpenClaw was in the middle of its explosive growth phase, Anthropic's legal team sent a trademark notice. The original name — "Clawdbot" — was too phonetically similar to "Claude."

From a pure trademark perspective, they weren't wrong. Legally defensible. Strategically catastrophic.

Three name changes in four days. The fastest triple rebrand in open-source history. Security researchers used the chaos to publish vulnerability reports. Hundreds of misconfigured instances were exposed. The whole Claude ecosystem took collateral damage.

Here's the part that should make Anthropic's leadership wince. Let's compare responses:

The shift to "OpenClaw" was deliberate. The new name positioned the project as model-agnostic infrastructure — not "Claude's assistant" but a platform that works with any AI model. The rename Anthropic forced literally removed Claude's branding advantage from the most popular AI agent on Earth.

The Bidding War: Zuckerberg vs. Altman

After the rename dust settled, Steinberger spent a week in San Francisco meeting with every major AI lab. Both Mark Zuckerberg and Sam Altman personally courted him. Satya Nadella caught up with him at an event in Vienna.

The Zuckerberg interaction was pure Silicon Valley theater. Zuckerberg reached out via WhatsApp. Steinberger insisted on calling immediately. Zuckerberg said he needed 10 minutes to finish coding.

Then they argued for 10 minutes about whether Claude Code or Codex was better.

Steinberger was impressed. He noted that Zuckerberg was actively tinkering with OpenClaw rather than just doing corporate due diligence. Zuckerberg called Steinberger "eccentric, but brilliant."

Altman's approach was different. Steinberger described their conversation as "really, really cool" and called Altman "very thoughtful, brilliant." But what sealed the deal wasn't charm — it was alignment on mission.

Steinberger's stated goal: build an agent "even my mum can use." That required frontier model access, cutting-edge research, and the infrastructure to ship to billions. OpenAI had the distribution.

His words: "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 will live in a foundation. Open source. Model agnostic. With OpenAI sponsoring development and giving Steinberger dedicated time to keep building it.

Steinberger's condition was non-negotiable: OpenClaw stays open source. He compared the intended model to Chrome and Chromium — an open core with commercial layers on top.

The One-Person Billion-Dollar Company That Chose Not To Be

This is the part that hits different for the "one-person billion-dollar company" crowd.

Sam Altman has publicly predicted that AI will power the first one-person billion-dollar company. Andrej Karpathy said the same. Dario Amodei talked about it. There's reportedly a betting pool among tech CEOs for the first year it happens.

Peter Steinberger was the closest anyone has come.

180K stars. Global adoption across 100+ countries. Zero employees. Zero VC. Acquisition offers from both Meta and OpenAI implying billion-dollar valuations. And the man was losing $10,000-$20,000 per month running the thing.

And he walked away from the company-building path entirely.

Not because he couldn't do it. Because he didn't want to. His words: "It's not really exciting for me."

This is a man who sold PSPDFKit for over $100M. He spent 13 years building a company with 70+ employees. He knew exactly what company-building looked like. The meetings. The fundraising. The HR issues. The board dynamics. He'd already done the thing everyone's dreaming about.

He chose mission over empire.

The one-person billion-dollar company might not look like what we expected. The person who could build it might simply choose not to.

Three Lessons for Every AI Company

1. Your biggest fans can become your biggest competitors

Steinberger didn't start as a competitor to Anthropic. He was a superfan. He named his project after Claude. He defaulted to Claude's models. He was driving more organic Claude adoption than most of Anthropic's marketing efforts.

Don't send cease-and-desists when you should be sending job offers. Or at minimum, partnership proposals. Ecosystem champions are the rarest and most valuable asset in the AI platform wars. Anthropic had one. They litigated him into their competitor's arms.

2. A person building for fun and enjoying their work can do wonders

Steinberger wasn't grinding toward a Series A. He wasn't optimizing for retention metrics. He was a rich man with no financial pressure building something because it excited him. That's the most dangerous kind of competitor — and the most valuable kind of ally.

He ran 5-10 AI agents simultaneously. He spent his energy on system architecture rather than data transformation code. He shipped faster than entire VC-backed teams because he had zero meetings, zero stakeholders, and zero reason to compromise.

3. Distribution beats capability in the agent wars

Steinberger chose OpenAI over Meta not because OpenAI had better models — he literally preferred Claude's reasoning — but because OpenAI had better distribution for getting agents into the hands of normal people. His goal was an agent "even my mum can use." That's a distribution problem, not a capability problem.

This is the lesson Anthropic hasn't learned. Claude might be the best model. But if the most popular agent framework that was built on your model chooses your competitor because they have better distribution for consumer agents, your technical superiority doesn't matter.

The Full Timeline

• Nov 2025
  Clawdbot launches
  Steinberger publishes a personal AI assistant named after Claude. Recommends Claude Opus as the default model. 9K stars in 24 hours.
• Mid-Jan 2026
  Viral explosion
  Clawdbot surpasses 60K stars. Mac Mini hardware sells out. More Google searches than Claude Code or Codex combined.
• Jan 27, 2026
  Anthropic's cease-and-desist
  Anthropic's legal team sends trademark notice. "Clawd" is too similar to "Claude." Steinberger renames to Moltbot. Bad actors hijack the abandoned accounts within seconds and use the project's credibility for scams.
• Jan 30, 2026
  OpenClaw is born
  Third rename in four days. Project repositions as model-agnostic. Claude's branding advantage erased.
• Early Feb 2026
  The bidding war
  Steinberger flies to SF. Meets Altman, Zuckerberg, Nadella. Argues with Zuckerberg for 10 minutes about Claude Code vs. Codex. Both Meta and OpenAI make offers.
• Feb 15, 2026
  Steinberger joins OpenAI
  Altman announces the hire. OpenClaw moves to a foundation. OpenAI commits to sponsoring continued development. 180K+ stars and counting.

The Bottom Line

Anthropic makes the best model. Most people in the space quietly agree on this. Steinberger himself preferred Claude's reasoning. OpenClaw was built on Claude.

But Anthropic is playing the model game while OpenAI is playing the platform game. And in the platform game, the person who has the community wins.

Anthropic had the community's biggest champion. They sent lawyers. OpenAI sent Sam Altman.

This won't show up in Anthropic's quarterly numbers yet. But in two years, when OpenAI's agent platform is the default because Peter Steinberger built it, and it all traces back to a trademark cease-and-desist letter that didn't need to be sent — someone at Anthropic is going to look back at January 27, 2026 and wonder what might have been.

The claw is the law. And now the claw works for OpenAI.

Sources