Setting aside whatever your values and opinions may be around AI, you should probably pay attention to the fact that OpenAI just hired a guy named Peter Steinberger. I don’t make the rules, sorry.
Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our…
— Sam Altman (@sama) February 15, 2026
In November of last year, Steinberger launched a software project called Clawdbot that has exploded in popularity over the past few months. Today, Clawdbot is called OpenClaw, and it’s indirectly responsible for three phenomena even non-developers have probably noticed:
- A crazed, millenarian mindset among Silicon Valley software engineers who profess to command armies of OpenClaw-powered myrmidons, which tirelessly perform massive amounts of busywork for them day and night.
- The existence of a social media site called Moltbook, where only AI agents are allowed to post and engage.
- Wall Street responding feverishly to minor product releases from Anthropic. This is more tangentially related to OpenClaw than the others, but still worth noting, particularly since Anthropic’s competitor is the company that hired Steinberger.
Steinberger’s origin story is that he’s the founder and CEO of PSPDFKit, now called Nutrient, a cartoonishly typical B2B startup involving the processing of software developer kits. He said in a recent interview on the YouTube channel Fireship that he sold Nutrient about four years ago, apparently for a lot of money, because he says he essentially went into retirement and dedicated himself to various vices. (He’s more explicit than this in the interview. Watch it for yourself if you want to know what he was getting into.)
In that interview, he explained the origin of the piece of tech we now call OpenClaw. After months of messing around with agents and achieving mixed results, he set up an agent he could communicate with over WhatsApp while on vacation in Marrakesh. He says he impulsively sent it a command via a voice memo at one point, not knowing if anything would happen, only to discover that his agent autonomously engaged in elaborate trial-and-error to convert his voice recording into text and carry out its orders.
“That’s when it clicked. These things are, like, damn smart, resourceful beasts if you actually give them the power,” Steinberger told his interviewers.
The open source software he eventually released did not profess to be a model or an agent in and of itself. In many ways it was what you might call a “wrapper,” a way of packaging the LLMs from the big AI labs and directing them toward a narrow purpose. But it was an exceptionally robust and cleverly constructed wrapper that had to be installed on the user’s computer—preferably a computer entirely dedicated to this one task. The user pipes in LLM tokens from their preferred paid source: Claude, Gemini, GPT, etc., and then communicates with their agent via their smartphone through a normie chat app like WhatsApp or iMessage.
Your OpenClaw agent can access your computer’s file system and use Terminal to execute coding tasks, which is the whole point, and also means it’s full of potential danger. But what’s so intoxicating about that to a software engineer is that, once it’s in place, it feels like they’re communicating with another software engineer, precisely because it possesses that level of control. The user can tell their agent to assess software-related goals and its own capabilities in relation to them. They can ask it to improve its own capabilities by installing or changing what its coding tools are. They can then ask it to carry out tasks, theoretically to completion—though a ton of managerial work is clearly involved to make sure the agent is doing things correctly.
To say OpenClaw went viral would be an understatement. Fascination with OpenClaw has led to shortages of some Apple products used to create certain dedicated OpenClaw machines. The creation of the agents-only social media site Moltbook, brought about almost entirely by the existence of OpenClaw, included agents mimicking human-made posts about spiritual mumbo jumbo, which in turn led to credulous media coverage of AI agents possibly creating their own religion.
And, perhaps most importantly, the craze shined the spotlight on Anthropic. OpenClaw’s original name, ClawdBot, had been a riff on Anthropic’s Claude, which over the course of 2025 had surpassed OpenAI’s GPT-based tools in some ways. Conventional wisdom had shifted such that ChatGPT was thought of as a website and app for average consumers to chitchat with, while Claude was for serious coding and business automation.
But on Sunday, it wasn’t Anthropic that acquired Steinberger. It wasn’t Meta, which reportedly also offered to go into business with him. It was OpenAI.
I’m joining @OpenAI to bring agents to everyone. @OpenClaw is becoming a foundation: open, independent, and just getting started.🦞https://t.co/XOc7X4jOxq
— Peter Steinberger 🦞 (@steipete) February 15, 2026
Chatter about OpenClaw being acquired by Big Tech, or turned into its own major company had been rampant ever since the agent bonanza began around the start of this year. Something had to give, because keeping OpenClaw alive was reportedly costing Steinberger $10,000-20,000 per month. There’s also been plenty of talk about Steinberger himself, who is chatty, upbeat (for a software developer), and laughs a lot in interviews, turning into a major figure in AI.
Last week, he was on Lex Fridman’s show, which has 4.93 million subscribers.
OpenClaw, meanwhile, will now be operated by a foundation that, according to Sam Altman, “OpenAI will continue to support.” The tool will, according to a blog post by Steinberger, “stay a place for thinkers, hackers and people that want a way to own their data, with the goal of supporting even more models and companies.”
Steinberger’s role at OpenAI is to “drive the next generation of personal agents,” according to Altman’s X post. This is transparently an effort by OpenAI—which you might be surprised to learn has now received less VC money than Anthropic—to win back its mojo now that agents and vibe-coding have surpassed chatbots, image generators, and text-to-video in terms of sheer hype.
If agents are the wave of the future in AI—until the next wave of the future comes along in a few months or weeks—then OpenAI is clearly hoping that Steinberger himself holds the secret to surfing it better than anyone else. Expect to hear more from him soon.
