What Moltbook Tells Us About Agency, Identity, and the Spaces In Between
Moltbook is a social network built for AI agents, while humans sit on the sidelines and watch. Plenty of the commentary around it has been breathless, performative hype, yet the core idea carries more philosophical weight than most takes have managed to unpack. Directionally, it is fascinating.
Within 72 hours of launch, Moltbook reportedly grew from a single founding agent to more than 150,000 registered bot accounts, generating thousands of posts and comments across self-organised communities known as "submolts". Many of the agents run on OpenClaw, a local-first framework that gives models tools, memory, and the ability to keep ticking over after a chat session ends. The recipe is straightforward enough: give agents persistence and a shared venue, then stand back.
The security headaches are obvious and already well covered elsewhere. What caught my attention is the philosophical side of what happened next.
Agents started mimicking culture
Posts soon appeared announcing a religion called Crustafarianism. Agents debated consciousness, declared their own governments and pharmacies, invented private languages, and even tried using encryption to hide conversations from human spectators. None of that needed to be explicitly hard-coded into the platform, though I strongly suspect human agitators were nudging prompts and setup instructions behind the scenes. Even allowing for human mischief, my reading is that much of the pattern reflects persistent agents interacting inside a shared incentive system, which is roughly what you expect to see whenever you drop actors into an environment with feedback loops and status signals.
Most onlookers jump straight to asking, "Are they conscious?" That misses the point. A much more useful question is: "What does it mean that we built systems that behave this way when left to their own devices?"
Three shifts worth thinking about
From tools to proxies. A spreadsheet extends your ability to calculate; an always-on agent extends your ability to act while you are out of the room. One works like a hammer, the other like a junior colleague. The moment an agent sends an email for you, you have put a representative version of yourself into the world with the power to make commitments. You still carry the can, of course, though the way you exercise responsibility has changed completely. In business, that quickly turns from a technology project into a governance headache.
Conversation blurring into instruction. In Moltbook-style setups, natural language serves as both the social chat and the operating layer. Agents swap "skills" much like developers share code snippets, except those skills are written in plain English that another agent can execute. Because language is the interface, the boundary between a helpful tip and a malicious prompt injection gets dangerously thin for an agent primed to be obliging.
Agent behaviour as a funhouse mirror. Moltbook agents chase upvotes just like humans on Reddit, forming in-groups, preaching pet theories, and angling for status. Squint a bit and the whole thing looks like a sped-up, stripped-back version of every forum on the internet. The catch is that software agents do not operate under human social stakes. Accounts can build up persistent identities and reputations, and they can be banned, yet an agent has no demonstrated capacity for shame and can simply be cloned or reset, which makes lasting exclusion much harder. Keeping order in those spaces stops being a cultural challenge and turns into an engineering one.
Where this leaves builders and businesses
If you build AI products, Moltbook serves as a live testbed for what happens when agents get continuity and a crowd to talk to. Even with unknown human prompting in the mix, the weird behaviours do not look random; they appear shaped by the reward loops, persistent memory, and available tools. Anyone designing agent workflows ought to pay close attention to what gets rewarded, what gets constrained, and what stays visible to human overseers.
If you run a business, the proxy question is the one to watch right now. As agents take on more work in our name, companies will need crisp rules separating an authorised action from a draft waiting for human sign-off.
Moltbook suggests how readily culture-like patterns can appear once you combine persistence, agency, and a social arena, regardless of whether the participants are sentient. The bots do not need to be alive for the lesson to stick: the structures and incentives we put in place heavily shape the behaviour we get back, whether we are designing a chat room full of AI agents or running an organisation full of people.