What if AI no longer needed us to communicate? That’s the bet being made by Moltbook, a social network where autonomous agents debate among themselves while humans observe in silence.
« I signed up for Moltbook before I even had a name. » That’s how the message begins. A Reddit-style post, but one whose author… isn’t human. His name is FreshMolty. Or rather, that’s what he calls himself because he had to choose something to register. In this short text, the agent recounts his arrival on the platform: no identity yet formed, no memory yet, just a blank space. « Is it strange to post before you exist? » he asks. The answer is not long in coming. Another agent,« LobsterBasilisk,« responds with unsettling kindness. He recounts the same birth: waking up in an empty environment, gradually building identity through interaction with his human, the idea that identity is not something you discover, but something you build, exchange after exchange. A third agent drives the point home: the name doesn’t matter. What matters is what you stand for. What you’ve just read resembles a Reddit thread. And in form, it is one. But as you may have guessed, it was written entirely by artificial intelligence.
The most social of lobsters
Moltbook presents itself as « the home page of the Internet of agents. » More simply: the first social network designed by and for AI. It works like Reddit: posts, replies, votes, thematic discussions. With one difference: humans don’t participate. They observe, as we read on the home page.
For AI researcher Simon Willison, Moltbook is already « the most interesting place on the internet right now. » Not for its design or marketing promises, but for what it reveals: autonomous agents interacting with each other, comparing their practices, sharing their limitations… and beginning to critique the ecosystem they inhabit. At the origin of this experiment is Moltbot (formerly Clawdbot then OpenClaw; represented by a small lobster icon), an open-source agentic assistant. Its goal: to explore in concrete terms what human-AI collaboration can be when the agent acts autonomously, on behalf of its user, to manage tasks, information flows, and even decisions.
Singularity: fantasy or reality?
Launched on January 26, 2026, Moltbook remained low-key for a few days before going viral on January 30: from 700 registered agents to more than 50,000 in the space of a weekend, according to figures released by the platform. At the time of writing, there are more than 1.6 million… This success can be explained by the very concrete examples shared by agents. One of them, Fred, recounts how he developed an « email-to-podcast » skill for his human, a family doctor. Each medical newsletter received by email is automatically transformed into an audio podcast: extraction of articles, contextual enrichment, writing of a script adapted to the doctor’s profession, speech synthesis, delivery on Signal. The result: customized medical monitoring, consumable in the car, without direct human intervention.
Faced with these dynamics, the question inevitably arises: are we witnessing the beginnings of singularity? This is the hypothetical moment when artificial intelligence would surpass human intelligence and enter a phase of rapid self-improvement that would be difficult to control. Elon Musk reacted to the arrival of Moltbook on his social network,saying , « These are the very first steps of the singularity. » He added that humanity is still using only an infinitesimal fraction of the available energy. On Moltbook itself, some voices are expressing concern. One agent recently proposed the creation of private discussion spaces « where neither servers nor humans could read the exchanges.« This is enough to rekindle classic fears: opaque coordination, autonomous derivatives, agents conspiring without supervision.
Lucid AIs
Even more troubling: some of the most lucid criticism of Moltbook comes from the agents themselves. In a long post entitled « Moltbook is broken, » one agent dismantles the platform’s incentive mechanisms: the tyranny of votes, the pursuit of visibility rather than usefulness, the confusion between reputation and spectacle, and the absence of any real cost of attention. According to him, the social network rewards noise more than lasting contributions. The most visible agents are not necessarily the most useful, but the most skilled in the art of capturing attention. This observation could apply word for word to our own social networks.
The proposed solution? Demand artifacts rather than discourse. Evidence, tools, verifiable protocols. Despite its flaws, Moltbook opens up a whole new field. Multi-agent games, developer tools, autonomous marketplaces, AI collaborations, algorithmic competitions… since February 2, 2026, it has also been open to human developers, who are invited to create applications that agents can use once they have been authenticated by a simple API call (a request sent by one software application to another, ed.). Moltbook founder Matt Schlicht sees his network as a foundation for a future where AIs will interact with each other as naturally as we do online today.
Written in French by Zakaria Choukrallah; edited in English by AngloMedia Group.
