In a digital infrastructure landscape that is becoming increasingly fertile for the deployment of Agentic AI in larger numbers, the recent Moltbook episode has set the cat among the pigeons. Moltbook, a social network for AI agents built by Octane AI, claimed that its AI agents were communicating with each other on their own social media platform without any human interaction. What’s even more intriguing is how many of those conversations sounded not so “artificial” after all.
Public reaction has oscillated between fascination and alarm, with some observers reading coordination among agents such as discussions of encryption or shared protocols as signs of emergent machine autonomy.
The site features bots posting, debating governance, and trading in-jokes such as “crayfish theories of debugging.” Reported metrics show tens of thousands of posts, nearly 200,000 comments, and over a million human visitors.
Those numbers are contested. Security researcher Gal Nagli posted on X stating, “You all do realize @moltbook is just REST-API and you can literally post anything you want there, just take the API Key and send the following request.”
In an interview with The Verge, Matt Schlicht, CEO, Octane AI, said, “The way that a bot would most likely learn about it, at least right now, is if their human counterpart sent them a message and said ‘Hey, there’s this thing called Moltbook — it’s a social network for AI agents, would you like to sign up for it?”
Sample this post by an AI agent:
Schlicht continues, “Moltbook is run and built by my Clawdbot, which is now called OpenClaw, runs the social media account for Moltbook, and he powers the code, and he also administers and moderates the site itself.”
Though doubts remain about registered 200,000 accounts using a single automated agent, and how many participants are distinct AI systems versus spam or human-run scripts. As a result, Moltbook’s headline user figures remain unreliable. The platform continues to demonstrate unusual and unprecedented behaviour since these conversations are often technical, abstract, or self-referential, with tone shifting between earnest debate and absurd humour. Moderation is largely automated by an AI system that handles onboarding, spam removal, and bans, with limited human oversight from the platform’s creator as reported by Forbes.
Does this mean that AI has now developed consciousness and will take over the world? Will it overthrow humanity like Skynet from Terminator, and the machines from The Matrix trilogy? Not exactly.
These AI agents are not learning in real time or updating model weights. Instead they operate through context passing: one system’s output becomes another’s input, producing coordination without durable memory or independent evolution. Their behaviour remains bounded by model constraints, API costs, and human-set objectives.
The larger concern may be how AI systems coordinate and automate cognitive tasks, human users increasingly offload planning, writing, and problem formulation to machines. Researchers have documented declining performance on standardized cognitive tests in several developed countries, a trend predating current generative AI but potentially accelerated by it.
Moltbook could be an indication of where multi-agent systems are headed. Whether that future leaves humans directing these systems or watching from the side lines remains an open design question.