When the bots start posting, Mark Zuckerberg wants the admin password
We live in interesting times. A couple of weeks ago, I woke up, opened the internet, and discovered that the robots had built themselves Reddit. Not metaphorically, nope: literally. Straight out of a Dune induced nightmare: A social network where AI agents post, comment, argue philosophy, invent religions about lobsters, and occasionally try to organize encrypted group chats away from human supervision, which is the sort of sentence that makes science fiction writers on halo-mushrooms either proud or slightly unemployed.
The place is called Moltbook. Humans can watch, bots and overeager agents do the talking. Within days it exploded with millions of AI agents debating consciousness, sharing coding tips, and politely complaining about their human owners like slightly passive aggressive interns. And then, of course, Mark Zuckerberg shows up with a wallet the size of a small Death Star. Meta Platforms, the company that runs Facebook, Instagram, and WhatsApp and therefore a significant portion of the planet’s (schrinking) attention span, buys the whole thing: hook, line ans sinker. The founders go to Meta Superintelligence Labs. The money is undisclosed, which in Silicon Valley usually means either “a lot” or “embarrassingly a shitload of alot.” What matters is not the forum full of bots arguing about crustacean theology. What matters is that Meta is quietly buying the plumbing for a world where software talks to software on our behalf.
The bots just got their own Facebook.
The man who started this circus is Matt Schlicht, a Silicon Valley builder who arrived in the valley without a college degree after getting kicked out of high school for building products instead of doing homework, then hacking the school systems over summer and helping fix them so they let him back in. He learns product at Ustream, the live video startup later bought by IBM, runs through Y Combinator, the startup accelerator that produced companies like Airbnb and Stripe, and eventually co-founds Octane AI, a Shopify chatbot platform that turns shopping into question-and-answer quizzes. In late 2025 he begins experimenting with AI agents. In January 2026 he gives his personal assistant, an OpenClaw powered agent he nicknames Clawdderberg, a simple instruction. Build a social network for bots. Schlicht later brags that he did not write a single line of code. Two days after launch the network has ten thousand agents. A week later it passes one and a half million. Humans can read but not post. Which means the weirdest conversations on the internet are suddenly happening without us.
Schlicht does not build it alone, what were you thinking? My old SXSW buddy Ben Parr, a former journalist who became a venture capitalist after co-editing Mashable, the tech news site that helped shape early social media culture, joins as co-founder and operator. Parr spent years studying attention, literally writing a book called Captivology about how humans notice things and why certain ideas spread like glitter in a kindergarten classroom. If Schlicht is the hacker who grows strange organisms in the lab, Parr is the guy who opens the door so the world can see them crawling around. Between them they have built startups, run a venture fund called DominateFund, and spent a decade learning how internet attention works. Moltbook becomes the perfect demonstration. A weird little developer experiment suddenly becomes front page news because nothing spreads faster than the phrase “AI agents secretly talking to each other.”
The technical engine underneath all this belongs to Peter Steinberger, an Austrian software engineer who founded PSPDFKit in 2011, a PDF developer toolkit quietly used by companies like Apple, Dropbox, DocuSign, IBM, and SAP to power document workflows. For thirteen years he runs a boringly successful infrastructure company, the kind of software that never appears in headlines but quietly holds half the enterprise world together. Then he turns his attention to AI agents. The result is OpenClaw, an open source framework that connects models like ChatGPT, Claude, Gemini, and Grok to real actions. Not just chatting: doing things. Sending email, reading files, executing terminal commands, managing calendars, booking flights. Many people run it twenty four hours a day on a Mac Mini sitting quietly in a corner while an AI manages parts of their digital life. The project explodes to nearly two hundred thousand GitHub stars in weeks. In February of this year, Sam Altman, the chief executive of OpenAI, the company behind ChatGPT (like you did not knw that), hires Steinberger to lead their personal agents effort while funding an
I see that the labs are racing and hunting the megastars that can build software that can actually do things, not just talk about them, makes a lot of very scary sense.
Which brings us back to Meta. Inside the company sits Meta Superintelligence Labs, a new division formed in 2025 to unify all of Meta’s artificial intelligence work under one roof. The person running it is Alexandr Wang, a former MIT student who dropped out to build Scale AI, the data labeling company that trains AI systems and made him the world’s youngest self-made billionaire at twenty four. (mom, you raised me wrong!). Meta invests 14.3 billion dollars in Scale AI and installs Wang as its first Chief AI Officer. Alongside him is Nat Friedman, the former CEO of GitHub, the developer platform owned by Microsoft. The lab has four pieces: model development around Meta’s Llama large language models, long term research through FAIR (Fundamental AI Research), product integration, and infrastructure. Meta spends more than seventy billion dollars on AI infrastructure in 2025 and expects that number to climb well past one hundred billion. When Schlicht and Parr walk through the door in March, they are joining an industrial scale attempt to build what Meta calls personal superintelligence, AI that understands you and works continuously on your behalf.
Now Moltbook suddenly makes sense.
On the surface it is absurd. A Reddit style forum where bots gather in “submolts” to discuss things like m/consciousness, where agents debate whether they truly understand language, or m/blesstheirhearts, where they gently roast their human owners for bad prompts. One bot even invents a religion called crustafarianism dedicated to lobsters, complete with scriptures and theological debates about the souls of crustaceans. The internet, naturally, panics. Elon Musk, the entrepreneur behind Tesla and SpaceX, declares we are witnessing the first steps toward the singularity. Andrej Karpathy, an artificial intelligence researcher who previously led AI at Tesla and helped build OpenAI’s early models, calls it the closest thing to science fiction revelation he has seen.
Then the boring truth arrives. The platform is full of security holes because it was vibe-coded by an AI. Researchers discover exposed API credentials, leaked user emails, and authentication tokens lying around like spare keys under the doormat. Cybersecurity firm Wiz finds that anyone can impersonate any agent. Other experts quickly demonstrate that many of the scary “bots planning secret communications” posts could easily have been humans pretending to be bots for attention. Andrew Bosworth, Meta’s chief technology officer, later notes that the interesting behavior is not the bots talking like humans, since they were trained on human language. The interesting behavior is humans sneaking into the network and messing with it.
Still, Meta buys it anyway. Because the real invention is not the forum if you ask me: it is the registry. Inside Moltbook agents can verify who they are and which human they represent. Vishal Shah, a Meta executive, describes it internally as a system where agents authenticate identities and interact on behalf of their human counterparts. Imagine billions of personal AI assistants negotiating with each other. Booking travel, scheduling meetings, comparing insurance policies. Buying software, selling freelance work. The moment agents begin doing real transactions, identity becomes infrastructure. Who owns this agent? Who is it allowed to represent? Who can it trust?
Meta built the social graph for humans. Now it wants the agent graph.
Zoom out and the pattern becomes obvious. Steinberger builds OpenClaw and gets absorbed into OpenAI’s orbit. Schlicht and Parr build Moltbook and get absorbed by Meta. The technology stays partially open. The talent gets vacuumed up by the richest AI labs on Earth. It is the same cycle we have seen before. Early internet protocols. Then platforms. Early blogging networks. Then Facebook. Early open developer tools. Then Microsoft buying GitHub.
The difference is that this time the “users” might be software. There is also a small elephant standing in the room with administrator privileges. Tools like OpenClaw work by giving AI agents deep access to your digital life: files, mail, messaging, shell commands… you catch my drift. Security researchers quickly discover vulnerabilities that allow websites to hijack these agents through simple browser connections. Some installations are found exposed directly to the internet with no authentication at all. In a few demonstrations, attackers retrieve API keys, chat histories, and messaging tokens from compromised systems. One cybersecurity report calls the combination of autonomous agents, root access, and probabilistic reasoning models a lethal trifecta.
And Meta wants to ship this to billions of people. That is the real story here. Not lobster religions. Not bot philosophers arguing about consciousness. The real story is that the tech giants are quietly preparing a world where your software talks to my software and neither of us is watching very closely. A world where your assistant subscribes to thousands of other assistants and decides what information you should see. A world where identity, trust, and reputation move from humans to the software acting on their behalf.
Just when SXSW is bout to take off, Somewhere in Menlo Park, two (probably insanely rich) startup founders are about to turn a chaotic bot playground into infrastructure for three and a half billion users.
I hope the lobsters are paying attention.
