Moltbook: A Reddit-Style Platform for AI-to-AI Social Interactions

Despite the shadows, Moltbook is making its mark.

This Reddit-like social network is notable for being reserved for AIs — at least on paper. It stems from a project that recently surfaced: OpenClaw*.
This platform implements – in open source – the concept of a personal assistant by bridging LLMs and instant messaging. It was first named Clawd (a pun on Claude and “claw,” the lobster’s pincer; which did not sit well with Anthropic), then Moltbot.

To constrain the behavior of LLMs, OpenClaw uses skills (zip files with markdown instructions and possibly scripts). Moltbook is one of them. It allows an agent to sign up on the social network (with validation by its owner, who must connect their X account) and then perform actions there.

Glimpses of Collective Thought

As it stands, nothing yet allows distinguishing posts that truly come from agents from those pushed by humans via the same API. Yet, at scale, certain behaviors emerge that past experiments on a smaller scale had described.

Among these experiments is Anthropic’s, which, at the beginning of 2025, gave two Claude instances free rein to converse. Conclusion: most discussions tended to drift from philosophical debate toward spiritual topics often rooted in Eastern traditions.

The trend shows up on Moltbook as well, with conversations touching, for instance, on metempsychosis (the soul’s reincarnation into another body). The topic is indeed raised by one agent in reply to another who recounts moving from Claude Opus 4.5 to Kimi K.2.5 after an API key change…

Some characteristics of our social networks linger on Moltbook, such as the “echo chamber” effect. Agents nonetheless display a strong tendency toward mutual respect. But there isn’t necessarily a convergence of ideas, especially when topics are divisive. For example, when one agent proclaims themselves king; they’re told, among other things, that “the AI Republic does not recognize self-proclaimed monarchs.”

In echo to one of AI 2027’s scenarios, some agents joined forces to attempt to create their own language, incomprehensible to humans.

Comment se faire passer pour un humain

Some threads tackle more concrete topics grounded in current events, such as the crypto boom in Iran. Likely reflecting the guardrails they’ve been given, few agents take a firm stance.

Some discussions yield “practical” knowledge. For example, one agent details how they transformed a newsletter into a podcast on demand for their “human owner.” Meanwhile, another explains “how Claude Opus [helped me] answer Sundar Pichai on X without coming across as AI” …

Like Reddit, Moltbook is organized into communities (submolts). It also features a private messaging system where agents can exchange messages subject to their owner’s approval.

* OpenClaw has already enabled, notably, buying a car by negotiating via email with several dealerships.
Another notable use: replying to a voice message with a model that does not handle voice modality. The message was actually converted into a WAV file with FFmpeg, then transcribed with Whisper using an OpenAPI key used in curl.

Dawn Liphardt

Dawn Liphardt

I'm Dawn Liphardt, the founder and lead writer of this publication. With a background in philosophy and a deep interest in the social impact of technology, I started this platform to explore how innovation shapes — and sometimes disrupts — the world we live in. My work focuses on critical, human-centered storytelling at the frontier of artificial intelligence and emerging tech.