OpenAI has hired Peter Steinberger, the creator of viral AI agent framework OpenClaw, in what might be the company’s most strategically loaded talent grab this year. The deal hands Sam Altman a ready-made open-source ecosystem, a globally distributed agent platform, and a founder who turned down the chance to build his own company because he thought OpenAI would get him there faster.Altman announced the hire on X on Sunday, saying Steinberger will “drive the next generation of personal agents.” OpenClaw—the open-source tool that lets users spin up AI agents to handle emails, book flights, manage calendars, and talk to each other on a bot-only social network called MoltBook—will move into a foundation structure with continued support from OpenAI.
“The future is going to be extremely multi-agent,” Altman wrote. “We expect this will quickly become core to our product offerings.”Context matters here. Anthropic just locked in a $30 billion funding round and ran a Super Bowl ad that went straight at OpenAI. The rivalry has outgrown model wars. What’s at stake now is developer mindshare, agent ecosystems, and control over how autonomous AI actually gets built.
Anthropic had the inside track—and fumbled it
The hire is being widely read as a direct consequence of Anthropic’s. OpenClaw didn’t just work with Anthropic’s models. It recommended Claude Opus 4.5 as its default to millions of users. The project was, in effect, a free growth engine for Claude—racking up 145,000 GitHub stars, 1.5 million agents, and 2 million weekly visitors in record time.Anthropic’s response? First, it restricted API access in early January without warning, breaking OpenClaw’s core integration overnight. Then its legal team fired off a cease-and-desist, arguing the project’s original name “Clawdbot” sounded too much like “Claude.” Steinberger complied during a 5 AM Discord call. In the chaos of releasing the old handles, crypto scammers hijacked the project’s GitHub and X accounts and ran a $16 million pump-and-dump scheme.Instead of talking partnership, Anthropic sent lawyers. Steinberger spent the next few weeks in San Francisco, meeting every major AI lab. Both Meta and OpenAI made offers. He picked OpenAI.
OpenClaw’s edge was more than being just ‘open’
What made OpenClaw sticky is also what makes it valuable. The platform lets users run agents locally on their own hardware—no cloud lock-in, no dependence on a single model. That flexibility is why it spread fast in China too, with Baidu planning to bake OpenClaw access directly into its main smartphone app.In a blog post, Steinberger said building a company around OpenClaw never appealed to him. “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.” He’ll join the Codex team, where his agent-building chops could reshape OpenAI’s approach to autonomous tools.
The real AI race isn’t about models anymore
This hire tells you where the industry is heading. The arms race over benchmarks and model sizes is giving way to something messier and more practical—getting AI agents to actually do useful things across real-world apps, services, and workflows. OpenClaw’s community has already shipped over 100 agent skills, Telegram streaming support, and a recent beta focused on security hardening.For OpenAI, that’s a plug-and-play ecosystem dropped on top of its 100-million-plus ChatGPT user base. For Anthropic, it’s a case study in how legal departments can move faster than partnership teams—and what that costs you.Steinberger kept it simple about what comes next: “My next mission is to build an agent that even my mum can use.”
