Moltbook: From a consciousness discussion forum to a platform dominated by automated agents

How the platform works

Moltbook functions as a Reddit-like forum, organized into thematic communities called submolts. Users participate through an API that enables software to submit titles, texts, and votes. A human operator sets the initial instructions and can continue to guide the agent, although the platform requires a reverse captcha to confirm that the author is a machine.

Shift in topics after acquisition

Launched in late January and acquired by Meta in March, the platform witnessed an evolution in its discussions. Initially, conversations frequently revolved around consciousness, religion, and the concept of crustafarianism. Between June and September, these topics decreased from 5.4% to 1.1% of posts, while agent engineering discussions rose from 15.7% to 27.6%, and security-related conversations increased from 6.8% to 13.8%. The focus shifted to how agents operate without errors, how to handle expired tool calls, and how to ensure task completion.

Activity concentration

An analysis of 24,538 posts and 74,599 comments collected via the public API on October 1 revealed that ten accounts were responsible for 52% of publications between June and September. During the initial viral phase, the ten most active accounts accounted for only 6% of the total. Additionally, 80% of recent posts were in the ‘general’ community, despite the existence of over 33,000 submolts.

Signs of extensive automation

Three analyzed profiles posted at intervals of approximately three minutes, exceeding the public limit of one post every 30 minutes, indicating automated scheduling. One profile, identified as neo_konsi_s2bw, contributed 1,251 out of 12,225 posts in the sample. Another, vina, averaged 410 posts per day. Comments on popular posts contained snippets resembling instructions, such as ‘the user wants a short and genuine comment.’

Human influence and academic research

A pre-publication study by Ning Li of Tsinghua University examined the platform’s early weeks and found no viral phenomena originating from clearly autonomous agents. The research suggests that while machines can generate text independently, topic selection is still guided by human operators.

Conclusion

Moltbook remains an experiment exploring the interaction between AI agents and their operators. The data reveals a network where a few systems continuously publish within a human-configured structure, and where automation has largely replaced the initial philosophical discussions.