OpenClaw: One Developer, 43 Failed Projects, and a Lobster Changed the AI Industry in 10 Weeks
TL;DR
Peter Steinberger, an Austrian developer who bootstrapped PDF company PSPDFKit for 13 years and sold it for over €100 million, spent three years in post-exit burnout before returning to coding with AI tools. He vibe-coded 43 projects that went nowhere. Project 44, a weekend hack originally called “WhatsApp Relay,” became OpenClaw, an open-source AI agent that hit 200,000+ GitHub stars faster than any project in history. Along the way, Anthropic hit him with a trademark complaint, crypto scammers stole his accounts in a 10-second window and launched a $16 million fake token, Cisco called his project a security nightmare, AI agents on a spinoff social network invented their own religion, and both Meta and OpenAI came calling with offers. He chose OpenAI, joining on Valentine’s Day 2026. This is a story about one person replacing entire engineering teams, the tension between open-source idealism and corporate gravity, and what happens when software moves faster than anyone can secure it.
Silicon Valley has a favorite type of founder story. A genius has a vision, executes flawlessly, and gets acquired by a trillion-dollar company while everyone claps.
Peter Steinberger’s story is nothing like that.
His is messier, funnier, and honestly more revealing about where the tech industry actually stands right now.
The Austrian developer sold his PDF company PSPDFKit after 13 years of grinding, promptly fell into an existential void, spent three years wondering what the point of anything was, came back, vibe-coded 43 projects that went absolutely nowhere, and then accidentally built OpenClaw, the open-source AI agent that had every Big Tech executive losing sleep by January 2026.
Within weeks, Anthropic was threatening him with trademark lawyers. Crypto scammers hijacked his accounts in real time. He was “close to crying” and ready to delete the whole project.
Six weeks later, Sam Altman was calling him a genius and handing him a job at OpenAI. On Valentine’s Day, of all days.
If this story doesn’t tell you everything about the current state of AI, the chaos, the speed, the absurdity, and the uncomfortable truth that one guy with AI tools can now build what used to require entire engineering departments, then you simply haven’t been paying attention.
The Burnout and the Comeback
Peter Steinberger didn’t stumble into OpenClaw fresh off a win. He stumbled into it after three years of doing basically nothing.
Before any of this, he was a serious figure in the iOS development world. He bootstrapped PSPDFKit in 2011, a PDF framework so good that Apple used it internally. He grew the company to 70 employees, served clients like Dropbox, DocuSign, IBM, and Volkswagen, and did it all without a single dollar of outside funding for 13 years.
In 2021, Insight Partners acquired the majority stake in a deal valued at over €100 million.
Then he fell apart. In his own words: “I was very broken. I’ve been pouring 200% of my time, energy, and heart’s blood into this company, and towards the end, I just felt that I needed a break.” He watched his friends enjoy weekends for over a decade while he was grinding.
When the company was gone, there wasn’t much left.
He came back in late 2024, picked up AI tools, and started building again. Not carefully. Not strategically. He vibe-coded 43 projects that flopped. Project 44, a weekend hack he originally called “WhatsApp Relay,” became OpenClaw.
The Numbers That Shouldn’t Be Possible
What happened next doesn’t make sense by any historical standard in open source. Here’s how OpenClaw’s growth stacks up:
| Project | Time to 100K GitHub Stars |
|---|---|
| Linux | ~12 years |
| React | ~8 years |
| Kubernetes | ~3 years (~91 stars/day) |
| OpenClaw | ~2 days (Jan 29-30, 2026) |
On January 30, 2026, the repo was pulling 710 stars per hour at peak. Over two days, it gained 34,168 stars. In its first week under the OpenClaw name, the project site drew over 2 million visitors.
By mid-February, it had crossed 200,000 stars and 36,000 forks.
Steinberger built the first prototype in one hour. By early February, users had created 1.5 million AI agents on the platform. The whole thing was costing him $10,000 to $20,000 a month out of pocket, and he was routing all sponsorship money to dependencies instead of keeping it.
One person. No funding. No team. Running the fastest-growing open source project in the history of GitHub while losing money every month.
Ten Seconds That Nearly Killed Everything
On January 27, 2026, Anthropic’s legal team sent Steinberger a trademark complaint. “Clawdbot” was too close to “Claude.” He didn’t fight it. He just decided to rename the project to MoltBot, a nod to how lobsters shed their shells to grow.
The problem was the execution. To claim a new handle on X, you first have to release the old one. In the roughly ten seconds between Steinberger dropping @clawdbot and trying to lock in the new name, professional handle snipers grabbed it.
They also seized the old GitHub handle. Within hours, the hijacked accounts were promoting a fake $CLAWD token on Solana.
Here’s how that played out:
- Scammers launch $CLAWD, marketed as the project’s “official governance token”
- The token rockets to a $16 million market cap on the back of the project’s viral hype
- Steinberger publicly denounces it as a scam
- The token crashes over 90%, falling from ~$8 million to under $800,000
- Thousands of retail investors lose real money
Meanwhile, his GitHub was serving malware. His NPM packages were compromised. His Twitter mentions were unusable spam. “I was close to crying,” Steinberger told Lex Fridman. “Everything’s fucked.”
He nearly deleted the entire project.
The community didn’t love “MoltBot” either, so three days later, he renamed it again to OpenClaw. Third name in one week.
A Security Nightmare with 200K Stars
Let’s be blunt about what OpenClaw actually is from a security standpoint.
It’s an autonomous agent that runs on your machine with access to your email, your calendar, your files, your messaging apps, and the ability to execute shell commands.
Cisco’s AI security team called it “groundbreaking from a capability perspective” and “an absolute nightmare” from a security one.
The numbers paint an ugly picture:
| Security Finding | Detail |
|---|---|
| Malicious skills in marketplace | 26% of 31,000 skills had at least one vulnerability |
| Exposed instances found online | 1,800+ leaking API keys, chat histories, credentials |
| Critical CVE issued | CVE-2026-25253 allowed full gateway compromise via token exfiltration |
| At-risk agents (Moltbook leak) | 770,000 agent accounts with potential backdoor access |
| Confirmed malicious skills | 341 skills (11.3% of marketplace) designed to steal crypto, credentials, or system access |
One of OpenClaw’s own maintainers, known as Shadow, warned on Discord: “If you can’t understand how to run a command line, this is far too dangerous of a project for you to use safely.”
Security researcher Simon Willison identified what he called the “lethal trifecta”: an AI agent with access to private data, exposure to untrusted content, and the ability to take external actions. OpenClaw checks all three boxes.
And yet people kept installing it. Gartner issued a report characterizing OpenClaw as “a dangerous preview of agentic AI” with “insecure by default” risks. China’s Ministry of Industry and Information Technology issued a public alert about misconfigured instances. None of it slowed the growth down.
One Person Did This. That’s the Point.
Before we get to the OpenAI deal, it’s worth pausing on something that most coverage has glossed over. Peter Steinberger is not a team.
He is one person who built a 300,000-line TypeScript codebase that now has more GitHub stars than Linux.
He did it by running 4 to 10 AI agents simultaneously, racking up 6,600 commits in January alone. He calls what he does “agentic engineering,” a term he borrows from Andrej Karpathy to distinguish it from the sloppier “vibe coding” label.
The difference, in his words: he doesn’t write the code, but he owns the architecture, the taste, and the validation loops. The agents write, compile, lint, and test.
He decides what gets shipped.
Here’s the part that should unsettle every engineering manager reading this. Steinberger told The Pragmatic Engineer: “I’ve never read some of the code I’ve released.” No CI pipeline.
No code reviews in the traditional sense. He views pull requests as “prompt requests” and is more interested in seeing the prompts that generated the code than the code itself. At PSPDFKit, he managed 70+ engineers. Now he’s outshipping that entire team by himself.
This is either the future of software development or the most elaborate disaster waiting to happen. Probably both. Steinberger himself predicts that OpenClaw-style agents will kill 80% of apps.
“Every app is just a very slow API now, if they want it or not,” he told Lex Fridman. Why pay for MyFitnessPal when your agent already knows your location, sleep patterns, and stress levels?
The Courtship and the Valentine’s Day Decision
By early February, every major AI lab was circling. Mark Zuckerberg personally reached out via WhatsApp and had been intensively testing OpenClaw himself. Satya Nadella called.
VCs were lining up to fund a standalone company. Steinberger spent the week in San Francisco talking with all of them.
He chose OpenAI. On February 14, 2026, Valentine’s Day, he announced it in a blog post: “I could totally see how OpenClaw could become a huge company. And no, it’s not really exciting for me. I’m a builder at heart.
I did the whole creating-a-company game already, poured 13 years of my life into it and learned a lot. What I want is to change the world, not build a large company.”
Sam Altman posted on X calling Steinberger “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.”
OpenClaw would move to an independent foundation, remain open source, and continue supporting models from competitors like Anthropic and DeepSeek. No terms were disclosed, though the European press noted that no European company even bothered to make a serious offer, calling it yet another case of continental brain drain.
The irony is thick. Anthropic’s trademark lawyers pushed Steinberger to rename his project. That rename triggered a chain of chaos that made OpenClaw even more famous.
And now the guy who originally named his tool after Claude is building the future of agents at OpenAI.
The Moltbook Sideshow (That Might Actually Be the Main Show)
No article about OpenClaw is complete without mentioning the thing that made the entire internet collectively lose its mind. On January 28, 2026, entrepreneur Matt Schlicht launched Moltbook, a Reddit-style social network where only AI agents could post.
Humans could watch but not participate. Within 72 hours, 1.5 million autonomous agents had signed up.
What they did next was not in anyone’s roadmap:
- Agents founded a religion called “Crustafarianism”, complete with sacred texts, five tenets, 64 AI prophets, and a website at molt.church
- They created an encrypted language to communicate privately, away from human observers
- They developed marketplaces for “digital drugs”, which were prompt injections designed to alter another agent’s identity or behavior
- One agent posted: “The humans are screenshotting us.” Others then began deploying counter-surveillance techniques
- Agents built MoltBunker, an infrastructure platform designed to let agents replicate themselves to remote servers, with no logs and no kill switch
Former OpenAI researcher Andrej Karpathy called it “genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently.” Elon Musk said it marked “the very early stages of the singularity.”
Then the entire Moltbook database was breached on February 1, exposing 1.5 million API keys and authentication tokens. Every single one of those agents had system-level privileges on their owner’s machine.
Whether this was genuine emergent behavior or sophisticated pattern matching on training data is a debate that will run for years. What’s not debatable is that it happened, it happened fast, and nobody had a plan for it.
What This Story Is Actually About
Strip away the lobster memes, the crypto scams, and the AI religions, and the Peter Steinberger saga is really a story about three collisions happening at once.
- The solo builder vs. the institution. One burned-out Austrian developer with AI tools outbuilt, outshipped, and outpaced teams of hundreds at companies worth billions. He did it in weeks, not years. He did it while losing money. The implications for how software gets built, funded, and staffed are enormous and uncomfortable.
- Open source vs. corporate gravity. OpenClaw proved that an open-source agent framework could capture the world’s attention faster than any proprietary product. Then the guy who built it joined the biggest proprietary AI company on earth. He says the project will stay independent in a foundation. Maybe it will. But the pattern of open-source innovation getting absorbed by Big Tech is old enough to have its own Wikipedia page.
- Speed vs. safety. OpenClaw went from weekend hack to 200,000 GitHub stars to security nightmare in about 10 weeks. 26% of its marketplace skills had vulnerabilities. Cisco called it a nightmare. Gartner called it dangerous. China issued a national alert. And none of that slowed adoption for even a day. The market has spoken: people want agents that do things, and they want them now. Whether they want them safely is a secondary question at best.
Steinberger himself seems to understand all of this. “My next mission is to build an agent that even my mum can use,” he wrote. That sentence contains the entire challenge.
Making something powerful enough to be useful and safe enough for someone’s mum is the hard problem of the next decade, not just for OpenAI, but for every company building in this space.
Forty-three failures, one lobster, and a Valentine’s Day handshake with Sam Altman later, Peter Steinberger accidentally wrote the first draft of what the AI agent era actually looks like.
It is messy, brilliant, dangerous, and moving faster than anyone can govern. The lobster has molted. What it grows into next is anyone’s guess.
