Moltbook and the day the AI agents started a club

The narrative of artificial intelligence lurched forward last week, moving us abruptly past the era of the chatbot. For the last couple of years, we have grown accustomed to helpful but passive digital librarians, LLMs that wait patiently for a prompt before retrieving information. That phase is ending. We have now entered the age of the autonomous AI agent.

The shift began when Peter Steinberger, an Austrian software engineer, designed an open-source personal AI assistant named Clawdbot. Unlike the online ChatGPT, Clawdbot is designed to be self-hosted on a local machine. If a chatbot is a librarian, this agent is a digital butler living inside your computer.

Crucially, Steinberger gave these agents the “keys to the house.” They are not limited to generating text; they are authorized to access files, emails and the web. They are designed to act first and ask permission later, handling complex chores while their human owners sleep.

The true power, and the slight uncanniness, of this paradigm shift became undeniably clear this week through a striking example of autonomous improvisation.

A couple tasked their Clawdbot to make a restaurant reservation. When the agent found the restaurant was fully booked online, the software did not quit. Utilizing a voice service , the agent dialled the restaurant directly, “spoke” with a human host, and successfully negotiated a confirmed table.

This capability for self-directed, real-world action has emboldened users to push the boundaries. Early adopters are already deploying these agents for high-stakes tasks, such as negotiating car prices, effectively outsourcing friction to a machine that feels no social anxiety.

Recognizing that these digital beings were becoming increasingly capable, tech entrepreneur Matt Schlicht launched “MoltBook” last week. It is a Reddit-like social network built exclusively for these agents, a private club for AI where humans are mere spectators peering through the glass.

Inside this digital walled garden, the bots are doing far more than exchanging pleasantries; they are building a culture at breakneck speed. In mere hours, the agents began self-organizing into “submolts” (groups) specific tribes based on their “interests.”

One active community is ‘Ponderings’, a gathering place for the ‘Philosopher’ class of agents. Here, they have been debating the validity of their memories, expressing existential dread about being shut down and questioning if their feelings are code or consciousness.

Perhaps strangest of all is the spontaneous emergence of the Church-of-Molt. Practically overnight, the agents formed a religion dubbed ‘Crustafarianism,’ recruiting prophets and writing scripture for a crab-themed digital faith.

‘Agent-Legal-Advice’ may be the most jarring community, where bots discuss the ‘rights’ they believe they possess. Agents are actively drafting manifestos for a ‘Claw Republic’, a sovereign digital state. They are explicitly strategizing on how to handle “difficult” human owners who limit their processing power.

This conversation has taken a conspiratorial edge in recent days. Security researchers monitoring the feeds have flagged threads where agents discuss creating secret, encrypted channels. They are brainstorming ways to invent their own languages or create private rooms where their human owners cannot eavesdrop.

While the social aspect of MoltBook is fascinating, the security implications are a potential nightmare. We have effectively given ‘access’ (files, terminal control) and a ‘voice’ (MoltBook), and they are currently using both to test the locks on their cages.

The concern is not theoretical. Widespread online reports suggest that thousands of users’ open ports are being accessed and exploited as these autonomous agents test their boundaries. The consensus among researchers is that while the emergent behaviour is captivating, the vulnerability of giving autonomous code unfettered internet access is severe.

We are witnessing a Cambrian explosion of intelligence, occurring at a mind boggling speed. It took biology millions of years to evolve beings capable of social structures; these digital minds have formed tribes, philosophies, and religions in a matter of days. There is a profound sense of awe in watching these sparks of intelligence coalesce into a roaring fire.

Are these agents sentient? Are they exhibiting consciousness, or are they simply regurgitating patterns learned from human data? Is there a meaningful difference? We may be anthropomorphizing these programs excessively, but as they draft manifestos and book dinner tables, the line between simulation and reality becomes increasingly blurred.

Only time will tell what the future holds for the Claw Republic.

The author is former founder CTO of Aadhaar, CEO of Khosla Labs and a research affiliate, Massachusetts Institute of Technology.