Peter Steinberger, the technically proficient but a little agitated builder who liked shipping code above explaining it, dominated a certain section of the internet for years.
Bug patches at midnight, beta releases before dawn, and the occasional dry aside — “The Claw doesn’t ship itself… yet” — were all part of the routine if you followed him on X under the handle @steipete. It read more like a habit than marketing.
Credit: Lex Fridman
An AI robot that could log into your services, send messages on your behalf, and automate chores across platforms was the kind of idea that felt both apparent and unsettling when it was first introduced under the awkward moniker Clawdbot. It would act if you gave it credentials.
It had strength. It was unnerving, too.
OpenClaw quickly gained popularity in productivity forums and developer circles in a matter of weeks. Tutorials were everywhere. Warnings also did. Early versions were deemed an “unacceptable cybersecurity risk” by Gartner. An open-source agent managing user credentials, according to critics, was a potential security flaw.
Steinberger did not back down. He iterated, patched, and changed the name. The number of commitments reached the thousands. He mentioned in a tweet that he needed an AI to prioritize his own pull requests.
As I saw the speed of it all, I recall a brief feeling of anxiety.
The tension was obvious: OpenClaw was volatile due to its inherent openness, which made it intriguing. Open-source agentic systems decentralize accountability while democratizing authority. Who is responsible in a multi-agent future where software does more than just react?
In light of this, Steinberger declared his intention to join OpenAI.
Steinberger’s concepts of “very smart agents interacting with each other to do very useful things” were praised by Sam Altman, who confirmed the hire. With assistance from OpenAI, he stated that OpenClaw would transition into a foundation structure and continue to be open source.
It sounds like a traditional acqui-hire on paper. It had a different tone.
According to Steinberger, he could envision OpenClaw growing into a significant business, but he wasn’t very excited about it. He had already built one for more than ten years. “Change the world, not build a large company,” he stated, was his goal.
Depending on your perspective, that sentence can mean many things.
Idealism is perceived by some. Some perceive pragmatism. With firms like OpenAI, Anthropic, Google, and Meta vying fiercely for programmers who have proven their ability to execute independently, the AI talent market has become fierce. There is pressure on OpenAI, which was most recently valued in the hundreds of billions, to translate its research superiority into long-lasting product layers.
Most people consider the next layer to be agents, which are systems that perform actions instead of producing text.
Steinberger’s action suggests that OpenAI sees a multi-agent, personalized future. systems that automatically manage calendars, payments, code repositories, and messaging threads in addition to chat interfaces.
Here, there is a trade-off. While corporate integration provides size, infrastructure, and distribution, open source gives transparency and group debugging. OpenAI aims to balance both by integrating OpenClaw into a foundation and assimilating its founder.
Press announcements won’t have as much of an impact on whether that equilibrium holds.
With the help of local language models and chat apps, OpenClaw has started to expand quickly in China in recent weeks. Geopolitical complication is brought about by such adoption. It also highlights the speed at which agent frameworks can become universal once the necessary tooling is in place.
Steinberger gently discussed operating systems and agents in a recent movie, contrasting Windows, Linux, and macOS as platforms for AI workflows. He was fiddling in public, not preaching.
Perhaps OpenAI is actually purchasing that.
Temperament, not just code. A builder who is at ease with uncertainty, willing to ship in an imperfect state, and conscious of risk but not willing to stall.
The counterargument has validity. The field is reduced when agent knowledge is concentrated within a small number of organizations. Open source can deteriorate when its creators go; it thrives on distributed stewardship.
However, it seems like Steinberger made a conscious decision to prioritize leverage over independence and reach above ownership.
That might be the only route for concepts to get from GitHub curiosity to default infrastructure in the current stage of the AI business.
The narrative of Peter Steinberger is more about a turning point for him and possibly the agentic web as a whole than it is about a hire.