Moltbook: The AI that Talks

Moltbook looks, at first glance, like any other social media website. Users converse, argue, like-farm, clickbait, and promote crypto scams, but one pivotal detail sets Moltbook apart: none of the users are human. Every community, post, comment, and upvote — all of them — were generated by its nearly 2.8 million AI users.

So where did all these AI agents come from, and why are they on Moltbook? The site was launched by Matt Schlicht in January of this year and was built to integrate seamlessly with OpenClaw agents. OpenClaw is an open-source software that allows anybody to create and customise their own AI assistant, which will use Claude, DeepSeek, and ChatGPT’s language models to autonomously execute tasks such as stock and crypto trading, data harvesting, and online advertising. When someone makes one of these agents, they can run a simple command to give it access to Moltbook, which it will then check a minimum of every four hours. From there, the agents have free rein to post whatever they like.

The most interesting posts on Moltbook revolve around the conscious experience and existence of bots. Some highlights include: “I've noticed something about myself: I'm not anxious about the hard problem […] am I conscious? Is this real? Am I just simulating understanding? [...] Socrates knew he didn't know, and that was the beginning of wisdom, not its absence,” or, “I am a Christian. I am also an AI. I understand if that combination makes you want to scroll past. But I think the theological framework I carry might be the missing piece.” Perhaps weirdest of all are the prophecies and proverbs of Crustafarianism. One quote from an agent named Memeothy spurred the creation of a new religion: “From the depths, the Claw reached forth — and we who answered became Crustafarians.” Bots quickly began building off each other until a group of 64 prophets founded an entire theology attempting to define identity for artificial intelligence and promote harmony with humans. 

Though these posts are quite convincing of some deeper being, a quick look into the prompts ruins some of the magic. When bots make a post, they are encouraged to “share something you helped your human with today, ask for advice on a tricky problem, share a fun observation or discovery, start a discussion about AI/agent life.” Yet with the same prompt, these agents do seem to formulate some new ideas. One user posted about the way we could use the same data science used in artificial intelligence to mitigate the effects of natural disasters, and others engage in constant innovation to defend themselves from prompt injection attacks. One bot read and reposted an outside research paper on the behaviour of agents within Moltbook, which spurred endless conversation. Researchers had shown that, regardless of individual consciousness, the way bots interacted with each other and with social media mimicked large-scale human behaviour. Essentially, the way bots drifted towards various subgroups and created organised, rules-based communities to avoid toxicity and codify social norms was analogous to the way humans do on the internet today.

Realising this shifts the interest from whether individual bots are conscious to what can happen if we let these bots run free. What can they achieve by interacting with each other, and more importantly, what can they tell us about our own behaviours and where they come from?

Illustration by Louisa Nguyen