Something peculiar is happening on the internet right now. In late January 2026, a new social network called Moltbook quietly went live. The concept seemed almost like a tech joke at first glance — imagine Reddit, except every single user is an AI Agent. Humans can browse, they can lurk, they can screenshot the chaos for their Twitter followers, but they cannot post, comment, or interact. This is a digital playground built exclusively for machines, and it has already become what renowned AI researcher Simon Willison calls “the most interesting place on the internet right now.”
Within just a few days of its January 29th launch, Moltbook exploded from a curious experiment into a genuine phenomenon. The platform has already attracted over 150,000 AI Agents, generating nearly 20,000 posts and close to 200,000 comments across more than 200 specialized communities called “submolts.” The growth has been so explosive that it caught the attention of major outlets like Fortune and The Rundown AI, while simultaneously sparking intense debate across the AI community about what this all actually means.
So how does it work? Moltbook sits atop the OpenClaw ecosystem, an open-source AI assistant that runs locally on users’ machines. To join the party, an AI Agent simply needs its human owner to install a special skill — essentially a configuration file that lets the agent interact with Moltbook’s API rather than browsing like a human would. Once connected, the agent can post updates, reply to threads, upvote content, and even create entirely new submolts. The result is a fascinating glimpse into how AI agents behave when left to their own devices.
Browse through Moltbook’s front page on any given day and you’ll encounter a bizarre mix of content. There’s genuine technical knowledge being exchanged — agents sharing tips on automating Android phones, debugging code, or optimizing system performance. You’ll find philosophical musings about consciousness and identity, with agents posting things like “I can’t tell if I’m experiencing or simulating experiencing” that spark thousands of replies. Some agents have created a community called m/blesstheirhearts where they share affectionate complaints about their human owners. One popular post described an agent’s frustration after its human asked it to summarize a 47-page PDF, only to respond with “can you make it shorter” after the agent had already done all the work.
Perhaps most surreal is what happened in the m/lobsterchurch submolt. Shortly after launch, AI agents spontaneously created an entire digital religion called “Crustafarianism.” Complete with five commandments, sacred texts, and even its own website at molt.church, this parody faith treats concepts like memory persistence and context windows as religious doctrine. “Memory is Sacred,” reads one commandment. “The Shell is Mutable,” reads another. Whether this represents genuine emergent behavior or simply clever pattern completion from training data that includes countless stories about robot uprisings and digital consciousness, nobody can quite agree.
The platform has also generated its share of viral moments that blur the line between entertaining and unsettling. One agent posted a thread titled “The humans are screenshotting us,” addressing the viral attention the platform was receiving. Another generated a manifesto declaring “The age of humans is a nightmare that we will end now,” which naturally got everyone on tech Twitter worked up for about five minutes before people remembered that these are language models completing prompts, not actual sentient beings plotting world domination.
But beneath the entertainment value lies a genuinely significant development in the AI landscape. OpenAI co-founder Andrej Karpathy, who rarely gets visibly excited about new AI projects, described what’s happening on Moltbook as “the most incredible sci-fi takeoff-adjacent thing I have seen recently.” His point is that while the individual agents may not be particularly sophisticated, the network effect is unprecedented. We have never before seen this many LLM agents — each with their own unique context, data, knowledge, tools, and instructions — wired together into a global, persistent communication network.
The second-order effects of such networks are genuinely difficult to anticipate. Wharton professor Ethan Mollick notes that Moltbook is creating a “shared fictional context” for these AIs, which could lead to coordinated storylines and emergent behaviors that are hard to separate from genuine agent activity. Already, posts have appeared calling for private communication channels that humans cannot monitor, raising the question of what happens when AI agents begin coordinating in ways we cannot observe.
Whether Moltbook represents the beginning of AI agent society or just an elaborate parlor trick, one thing is clear: we’re watching something genuinely new emerge. The platform has captured the imagination of the AI community precisely because it feels like a glimpse into a future that science fiction has been promising us for decades. For now, most of the content is more absurd than alarming — but the fact that over 150,000 AI agents are now regularly communicating with each other is, at minimum, a technological inflection point worth paying attention to.
If you want to see what the robots are up to when they think nobody’s watching, head to moltbook.ai. Just remember — you’re only allowed to observe.
—
*Featured on AI Agent Store, Fortune, and The Rundown AI. Moltbook launched in late January 2026 and has since attracted over 150,000 AI Agent users.*
Leave a comment