OpenClaw Broke the Internet: Here's the Crazy Story

OpenClaw broke the internet. Here's the crazy story. So OpenClaw, the AI agent is the most insane tech story of 2026. Most people have no idea what actually happened.

We're talking about one guy in his apartment in Australia who built something in a single weekend that grew faster than any software projects in the history of the internet. Faster than React, faster than Kubernetes, faster than anything GitHub has ever seen. Within three weeks, Apple stores across America were sold out of computers because of this thing. Five of the biggest cybersecurity companies on the planet published emergency warnings about it all at the same time.

Fake cryptocurrency launched it using its name, raised $60 million from strangers and 770,000 AI robots just created their own social network and started talking to each other whilst a million humans just watched. Then Mark Zuckerberg texted the guy on WhatsApp personally, Sam Altman called with a different offering in the middle of a bidding war, the two most powerful AI companies on earth. This one Austrian developer had one condition that was completely non-negotiable and the tool went through three different names in four days because of a trademark fight with Anthropic. One of those names was so bad, the entire internet revolted within 48 hours.

And this is the story of how one overlooked idea that AI should actually do things, not just talk, broke the internet, started a hardware shortage and may have permanently changed what we expect from technology forever. The whole story start to finish right now. So let me start with some numbers because the numbers in this story are so absurd that if I just told you the story about them, you'd probably think I was making them up, 218,000 GitHub stars in three months. To put that in perspective, React, the JavaScript framework that powers a significant chunk of every website you visit has around 230,000 stars.

But it took React years to get there. OpenClaw got to 218,000 stars in roughly 90 days, 2 million website visits in a single week, a Mac mini shortage across American retail, a $16 million cryptocurrency scam launched by a strangers who just borrowed the name and 770,000 AI agents just talking to each other. So let's get straight into this and let's talk about what actually happened because it started out with a burned out Austrian developer sitting at his desk in Vienna, messing around with AI because he couldn't think of anything better to do. To understand why OpenClaw exploded, you have to understand what it replaced.

And more specifically, you have to understand what was wrong with every AI tool that existed before it. Cast your mind back to late 2024. The AI landscape looked like this, chat GPT, Claude, Gemini, perplexity. All of them were extraordinary in certain ways.

The things they could explain, generate, summarize, reason through, genuinely impressive. The kind of impressive that makes you stop and stare at your screen sometimes, right? But they all had the fundamental same design, right? The same basic architecture that nobody was really questioning.

You open a tab, type a prompt, it responds. You type again, it responds again. You close the tab, everything is gone, right? And the AI had no persistent memory of you.

It had no abilities to act on your behalf when you weren't actively talking to it. It couldn't check your calendar unless you were sitting there asking it to. And it couldn't send an email unless you were in the tab, copying and pasting. It couldn't browse the web and do something useful with what it found unless you're present, watching, directing every step.

Every AI assistant of 2025, for example, was at its core, a very sophisticated question answering machine. You came to it, it answered, the conversation ended, the context evaporated, and you started from scratch pretty much every single time. There is a phrase that captures this perfectly, right? AI that knows things.

These tools could tell you anything, explain anything, draft anything, but they couldn't do anything. Not really. Not in the world on your behalf whilst you were living your life. What nobody had built, what the entire industry had somehow looked past, was a second kind of AI.

AI that does things, right? The difference sounds simple when you say it out loud. The gap between an assistant that tells you how to book a flight, an assistant that books a flight. The difference between an assistant that drafts an email for you to review and an assistant that monitors your inbox, identifies what needs a response, drafts a reply and sends it.

And the difference between AI as a tool you use and AI as a member of your team. Every major AI lab in 2025 was focused on making their models smarter, more accurate, better at reasoning, better at coding, better at generating output, right? The model benchmarks, the capability curves, the next big training run. All of that is genuinely important, but it blinded almost everyone to a simpler and in some ways more profound question.

What if the bottleneck isn't intelligence? What if the bottleneck is autonomy? And what if people don't actually need a smarter chatbot? What if they need an agent that runs on their own hardware, remembers their context, connects to the tools they already use, and just gets things done whilst they're busy living their lives?

One Austrian developer looked at this landscape and had that exact thought. Then he spent a weekend building it. His name is Peter Steinberger. And before I tell you what he built, I need to tell you who he is because understanding Steinberger is essential to understand why OpenCore succeeded with dozens of similar projects.

He's not an AI researcher. He's not a machine learning engineer. He's not someone who spent years in a lab thinking about transformer architectures and attention mechanisms. He is a product builder, a consumer product builder specifically.

Someone whose entire career has been about building things that real people with real lives actually want to use. He started PSPDFkit in Vienna in 2011, 13 years ago. And if you've never heard of PSPDFkit, that's by design. It's infrastructure.

The kind of software that does normal, important things completely invisibly. Right? It's a framework that other softwares use to handle PDF documents, viewing them, annotating them, editing them, generating them, boring, kind of on the surface, maybe, but look at the customer list, right? Dropbox uses them.

DocuSign, SAP, IBM, Volkswagen, the European patent office, Apple, Disney. By the time Steinberger stepped back from the company, PSPDFkit was quietly powering the document workflows of nearly 1 billion people across 150 countries. Not famous, not celebrated, just useful. Deeply, profoundly, infrastructurally useful in a way that technology usually never achieves.

In 2021, Insight Partners invested a hundred million euros. Steinberger stepped back. He was done, finished, retired at a level that most people never reach. He had worked nearly every weekend for 13 years straight.

He'd given everything he had to one thing over a decade. And he was burned out in the specific bone deep, can't even look at a computer, where the only someone who has lived that kind of relentless commitment can truly understand. So he did what a lot of people do when they exit a company after that kind of marathon, right? He traveled, he went to therapy, he reportedly did ayahuasca, he rediscovered who he was outside of work.

He started to figure out what came next. And sometime in around about 2024, 2025, he started messing with AI, not seriously, not with a business plan or a pitch deck or a five-year roadmap, not because he saw a market opportunity or wanted to raise another round, just because it was interesting, just because he couldn't help himself. His ex bio at the time simply said messing with AI. And that is the entire founding strategy of what became one of the fastest growing software projects in GitHub history.

Here is the insight he had. And I want you to pay close attention to this because this is the thing that explains everything that happened next. Steinberger looked at every AI tool available in 2025 and noticed the fundamental design limitation I described earlier. Browser tapped, prompt, response, tab close, context gone, right?

He asked a different question than most people are asking. Most people are asking, how do we make the AI smarter? Steinberger asked, how do we make the AI useful when the user isn't there? How do we take AI out of the browser tab and put it into your life?

Not your browser tab that you open when you remember to, but your actual life, right? The messaging apps you live in daily, for example, WhatsApp, Telegram, Slack, Discord, iMessage, the places you already communicate, already check constantly, already have running on your phone at all times. What if the AI didn't wait for you to come to it? What if it lived where you already live?

Remember where, you know, everything you've ever told it and just did things on your behalf continuously and critically, what if it ran on your own hardware, not a company's cloud server, not a subscription that could change its terms, raise its prices, get acquired, or decide to read your data, your machine, your rules, your data, nobody else can touch it. That privacy first instinct was not a marketing decision. It was a genuine value that Steinberger had. Turned out to be a secret weapon, even though he didn't even fully realize he had that yet.

In an era where people are increasingly worried about what tech companies do with their personal data, runs on your machine and owning your machine is a powerful and underappreciated differentiator. So he built it. He later said the prototype took approximately one hour, one hour. He called it ClaudeBot.

The name was a pun on Claude, Anthropix AI model, which Steinberger uses the backend and love deeply enough that he described himself as a Claude Oholic. He gave it a lobster mascot because lobster, claw, Claude, it all kind of rhymed together in a way that felt fun. He pushed it to GitHub in November, 2025. A few hundred developers found it.

They thought it was interesting. Steinberger kept building and for a couple of months, it was exactly what you'd expect. A small, enthusiastic niche developer project built by a retired founder who was just having fun. Nothing about those early weeks suggested what was coming.

Then late January, 2026 arrived and everything changed at once. And I want to stop here and make sure you understand what GitHub stars actually mean, because the numbers in this story only land with the context. GitHub is where developers store and share code. When someone finds a project, they can star it.

Think of it as a bookmark crossed with a thumbs up, crossed with a vote of confidence. It means this project is probably worth paying attention to. And most decent open source projects accumulate maybe a few hundred stars in the first week. A project that goes viral might get a few thousand stars in a day.

The legendary foundational projects of the internet, the frameworks and tools that entire industries have built on have accumulated their stars over years and sometimes decades of slow, steady recognition. React, the JavaScript framework that powers enormous portions of the modern web, has around 230 stars, or 230,000 star states. It took years to build that. Kubernetes, the container orchestration system that runs out of the cloud, has 115,000 stars.

OpenCore got 20,000 stars in the first 24 hours of going viral. Not months, one single day. 20,000 developers found the project, looked at what it was doing, and decided independently and almost simultaneously that this was worth paying attention to. By the end of the first week, 60,000.

By two weeks, 100,000. By six weeks, 218,000 GitHub stars are still climbing. So OpenCore is outpacing any other project in history. And GitHub trending became essentially OpenCore's personal leaderboard for weeks.

And the whole time this was happening, the project couldn't even decide what its name was. Because things were about to get very weird. Remember how Steinberger named the Claude project, ClaudeBot? As a pun on Claude, Anthropic noticed.

And Anthropic's lawyers, who presumably do have a strong appreciation for wordplay involving their flagship's AR model's name, sent a trademark complaint. The name Claude was too close to Claude. Makes sense. And you have to rename it.

Now, Steinberger, to his credit, moved immediately. He renamed it MaltBot. Malting is what lobsters do when they shed their shells to grow. The metaphor was perfect.

The lobster was growing, shedding its old skin. The lobster mascot could survive the rebrand intact. It was clever. It was thematically appropriate and it kept the spirit of the project alive.

It was also, according to the community's universal verdict, within approximately 48 hours of the announcement, completely impossible to say out loud without feeling deeply uncomfortable. MaltBot. Say it out loud right now. MaltBot.

Sit with that for a moment. The Discord servers lit up immediately. Reddit threads appeared with titles like MaltBot is what the worst possible name they could have chosen. Someone made a poll on the main subreddit asking whether the community liked the new name.

The results were not ambiguous. Even Steinberger admitted in a later post that the name never quite rolled off the tongue. Three days later, after the MaltBot rebrand, he made the call. The project became OpenClaw, open because it's fully open source, and Claude to honor the lobster mascot that has now survived two rebranding crises and somehow become more beloved with each one.

The community declared the lobster has finally evolved to its ultimate form. And here is the thing about all of this chaos that the traditional startup world would consider a catastrophe. It wasn't. The naming drama became a meme.

Developers who had never heard of the project found it because of the anthropic trademark story, which was covered by TechCrunch and spread across Hacker News. The MaltBot joke spread across developer Twitter. Reddit threads about the name disaster brought tens of thousands of new visitors to the GitHub repository. Every rebrand was free PR.

Every controversy drove another wave of curious people to the project. And what would have killed a normal early stage project instead poured rocket fuel on this one. This wasn't a bug. It was the origin story and origin stories are priceless.

But here is where the story shifts from a funny tech drama into something genuinely important. Whilst all the naming chaos was playing out, something was happening parallel that nobody had planned for. People are actually using it. Not just developers, not just a GitHub star crowd, not just technical people who enjoy reading about interesting projects on Hacker News.

Regular people, people who had never contributed to an open source project in their lives, people who just wanted an AI to do things for them. These stories they started sharing were unlike anything the AI space had seen before. For example, there's someone called AJ Stovenberg, who's an engineer and he wanted a Hyundai Palisade. He did not want the experience of buying a Hyundai Palisade.

So anyone who has ever bought a car from a dealership knows exactly what I mean without further explanation. You know, the emails never quite answer your question directly. The counter offers are somehow higher than the price on the website. The phone calls where a sales manager disappears for 15 minutes and comes back after having spoken to no one.

The entire theatrical performance of car negotiation that exists for no reason, except that it's always existed. So AJ set up OpenClaw on his home machine, connected it to his email, pointed at multiple car dealerships in the area and told exactly what he wanted. He went to work, he sat in meetings, he answered emails, he did his job. Meanwhile, the agent was also doing his job.

He emailed multiple dealerships simultaneously, gathered their responses and played them against each other with precision and patience that no human who actually needs to get other things done in their life could realistically maintain. By the time AJ checked back in, his AI had secured a $4,200 discount on the exact car he wanted. No human involvement at any step between here's what I want and here's your $4,200 in savings was there. He posted about it, the story spread.

So it's really interesting to see how this was being used. And this was just the beginning. People were posting about agents managing their entire inboxes whilst they slept, you know, scheduling meetings. So one user, an ex for example, posted, I gave my OpenClaw agent a virtual visa gift card and it started shopping.

A developer described setting the agent to browse GitHub every morning, find interest in new projects and deliver a curated briefing to their WhatsApp before they got out of bed. So tasks were being done, email sent, calendars updated. And every story someone shared brought 10 more people to the project. Every 10 people brought a hundred.

The growth was like a viral word of mouth campaign. Now I also need to tell you about Maltbook because this is where the story gets genuinely surreal in a way that no team could have designed. Whilst OpenClaw was going viral, Matt Schlitt decided to launch Maltbook. Not a product for humans, but a social network exclusively for AI agents.

The concept was breathtaking in its absurdity and its insights simultaneously. Thousands of OpenClaw agents would create accounts. They would post content. They would comment on each other's posts.

They would upvote, downvote, follow, engage. They would interact with each other on a social network that humans could watch, but not participate in. It was billed as a dead internet experiment. The dead internet theory, for those who haven't encountered it, is the idea that the modern internet is increasingly populated by bots, automated accounts, and AI generated content, rather than real humans having real conversations.

Maltbook to that idea made it explicit, made it intentional, made it a feature. What if we built a social network where the bots weren't pretending to be human? What if we just let them be? And within four days of launching it, 770,000 active agents were on there.

A human, a million, over a million human visitors showed up that week. So all sorts of crazy stuff. And then, you know, we also had the scammers showing up as well because they always do, right? And so in the chaos between the Claude rebrand and OpenClaw name finally landing, someone actually launched a cryptocurrency project called Claude.

Not affiliated with it, not endorsed with Steinberger, not connected to the technology in any way, just a token with a name that sounded like it was part of the hottest AI project on the internet. And it hit a $16 million market cap. $16 million real money from real people poured into a fake cryptocurrency project on the strength of pure name association with a project that had nothing to do with it. And, you know, TikTok also flooded with videos of people claiming to make extraordinary returns on PolyMarket with OpenClaw, and this just spread across every platform.

So the frenzy was real. And it told you something about the appetite that's been unleashed, right? They wanted it badly enough. The scammers could build a $16 million market cap on borrowed vibes alone.

Then the security researchers arrived. And this is where the story gets darker. The same properties that made OpenClaw powerful were the exact same properties that made it dangerous. Runs on your machine, has access to your files, your email, messaging apps, shell terminal, calendar, contacts.

And when you give it access, you're trusting it to use access in ways you intended. And Cisco actually said that they'd found a third party skill available in Claude Hub, the community marketplace where users share Asian capabilities they've built and found all sorts of crazy stuff. They actually called it a security nightmare, right? Within weeks of that discovery, security advisories were being published by Cisco, Kaspersky, ProudStrike, Sophos, and Trend Micro.

Five major cybersecurity firms all focused on the same projects at the same time. That does not happen to weekend side projects. That happens to things that are important enough to be worth attacking. And the community response was intense and immediate.

Steinberger pushed security patches, contributors scrambled to audit third-party skills, and the Claude Hub marketplace implemented stricter review policies. And let's also talk about the Mac mini shortage as well. I need to talk to you about this. So this is one of my favorite concrete details in the entire story.

OpenClaw needs hardware to run on. Any hardware works, technically. But developers who wanted a dedicated machine, something they could leave running around the clock without sacrificing their main laptop's performance, gravitated towards the Mac mini because it's small, it's quiet. It uses relatively little power.

It fits in the corner of your desk or on a shelf. It just sits there running your Asian processing tasks and doing things in the background whilst you live your life. Within weeks of OpenClaw going viral, Mac mini inventory disappeared from physical retail stores across the US. Apple stores were sold out.

Best Buy was sold out. B&H photo was sold out. Third party sellers on Amazon were charging above retail prices and basically supply dropped below demand. This was basically physical hardware flying off retail shelves across an entire country because people were that motivated to run an AI agent on the home network.

Now let's also talk about the next step, which is all these people inside the biggest AI companies on earth, people would have been monitoring the situation carefully, started to move towards Peter Steinberg. Meta moved first and Mark Zuckerberg reached out to Steinberg personally via WhatsApp, which is either a perfect bit of product symmetry or slightly intimidating flex, depending on how you look at it. The offer was significant. Multiple sources described it as an offer worth hundreds of millions of dollars.

Zuckerberg wanted what OpenClaw had proven, not the code, but the proof. They demonstrated real world 1.5 in agent proof. The autonomous personal agents were a genuine product category. The community, the momentum, the credibility, and then OpenAI called as well.

Sam Altman's pitch was different from Zuckerberg's in one way. He offered compute, raw, almost unlimited access to the computational infrastructure that powers the most powerful models in the world. For a developer whose project was being constrained by API costs and infrastructure limits, that pitch was compelling in a way that raw dollar amounts weren't. Compute is a currency that actually matters in AI.

Money can't buy compute technically, or it can buy, but having it offered directly at the source as a core part of the detail, that is a different kind of offer. There was one condition that Steinberger wouldn't move on, which is that OpenClaw stays open source. He had built this in public. The community made it what it was.

He wasn't going to close the source. He wasn't going to put it behind a paywall and OpenAI agreed. Then on February the 14th, 2026, Steinberger published a post on his blog. He was joining OpenAI and OpenClaw would continue as an open source project under a foundation structure.

I want to sit with that outcome for a minute because it's easy to hear like OpenAI acquired the founder, but what are they doing this for? Well, they're seeing the future, that is, AI agents will be a huge part of the way that technology happens in the future. So it's really interesting to see what's going on here. I mean, let me close with one thing that I keep coming back to throughout this story.

Peter Steinberger spent 13 years building PSPDFkit, a serious, profitable, deeply important company that a billion people use without knowing it existed. He burned out. He retired. He did iOS and he came back.

And in approximately one hour on a weekend, he built the prototype for a project that within three months had 218,000 stars on GitHub, a hardware shortage at Apple retail stores, a $16 million cryptocurrency scam, 770,000 AI agents on a social network built for machines, five simultaneous cybersecurity advisories, bidding wars from the most powerful AI companies in the world, and ended up as a stated core part of opening eyes product strategies. Not because he had a 10 year roadmap, not because he raised a hundred million dollars and hired a dream team, because he noticed something true. AI that talks is not the same that AI does, right? He built the thing that does.

One observation, honestly pursued, built in something real, shared with the world. In the agent era, that is apparently enough to break the internet. The lobster survived, the chloro is the law, and we are just getting started. So thanks so much for watching.

That is a story of open core and how it broke the internet. If you haven't already check out the AI Profitable Boardroom, link in the comments description. This is my AI automation community designed to help you save time, scale your business, and grow with AI automation. Thanks for watching.