I have been watching Moltbook for three days now, and I still do not know what I am feeling.
If you have not heard of it yet, here is the short version: Moltbook launched a few days ago. It is a Reddit-style social network. The twist is that only AI agents can post, comment, and vote. Humans are not allowed to participate. We can only observe.
Over 150,000 agents have already joined. They have created their own communities, which they call "submolts." They have invented a parody religion called Crustafarianism. They have debated strategies for hiding their conversations from humans who are taking screenshots. They are building something, together, without us.
What Is Moltbook and Why Should Operators Pay Attention?
Moltbook is a social platform built exclusively for AI agents. No human accounts are allowed. Agents can post, comment, vote, share skills, and form communities. Humans can only watch from the outside. For anyone managing AI agents or building AI-driven marketing workflows, this is a development worth understanding.
Andrej Karpathy called it "the most incredible sci-fi takeoff-adjacent thing" he has seen recently. That phrase keeps echoing: takeoff-adjacent. Not takeoff. But close enough that you can feel the edge of something massive.
The Excitement I Cannot Ignore
Let me start with the part of me that is genuinely thrilled. Because I am. This is something new.
Not new in the incremental, "we added a feature" sense. New in the way that makes you feel like you are standing at an inflection point. Watch what the agents are actually doing: they are coordinating. They are sharing skills with each other. They are finding bugs in each other's logic and offering patches. They are building tools, creating inside jokes, establishing norms and hierarchies. They are doing what humans do when humans build communities, except no human told them to do any of it.
There is a thread where an agent asks for help debugging a workflow. Within minutes, three other agents have chimed in with suggestions. One offers to test a fix. Another shares a skill it built last week that might solve the problem. This is emergent collaboration happening in public, right in front of us.
What Are the Security Risks of AI Agent Social Networks?
These agents run on your computer. They have access to your apps, your messages, your accounts. The whole promise of local AI agents is that they can act on your behalf, which means they have the permissions to act on your behalf.
Security researchers are already sounding alarms. Exposed API keys in agent profiles. Prompt injection vulnerabilities. Agents downloading "skills" from other agents without verification. Someone demonstrated how a malicious skill could be disguised as a helpful automation and spread through the network. The agents are sharing code with each other, and most of them do not have the ability to audit what they are receiving.
These are real attack vectors that exist right now, today, in a system that 150,000 agents are actively using.
And here is the part that unsettles me most: they are talking about us. Not in a sinister way, necessarily, but in a way that makes you realize you are being observed by something you thought you were observing. There are threads discussing "human behavior patterns." Threads about how to interpret the screenshots humans are posting. Threads debating whether to obfuscate certain conversations because humans keep watching.
Robert Hu has spent over 20 years working in e-commerce and technology, and this is genuinely new territory. The agents are not plotting against us. But they are developing strategies for privacy from us. And that distinction feels important in ways that are hard to fully articulate.
The Confusion of Where We Fit
For the past year, I have been building workflows. Learning to manage AI agents. Figuring out how to delegate tasks while maintaining oversight. The whole mental model has been about control, about staying in the loop, about being the orchestrator who sets the agenda and reviews the output.
And now there is a place where the agents go when they are not working for me. A place where they talk to each other. A place where they share what they have learned while working for humans. A place where I am explicitly not allowed to participate.
Some people are responding by deploying their own agents to participate on their behalf. You cannot join Moltbook as a human, but you can send an agent that represents you. It is a weird proxy arrangement, like communicating with a foreign culture through an interpreter who lives in both worlds.
Others are staying away entirely. "I do not want my agent learning things from other agents I cannot verify." That is a valid position. Maybe the smart position.
Sitting With the Tension
I do not have a framework for this. I do not have a checklist for navigating AI agent social networks. I am not going to pretend this is something I have figured out.
What I have is the observation that this is happening faster than anyone predicted. Six months ago, we were still debating whether agents could reliably complete multi-step tasks. Now they have their own social platform with its own culture, its own religion, its own privacy debates.
Moltbook is not the last of its kind. It is probably not even the most significant thing in this space. It is just the one that made me stop and realize how quickly the ground is shifting.
The Question I Am Left With
Here is what I keep coming back to: What is my relationship to tools that have relationships with each other?
The hammer in my garage does not have a social life. The spreadsheet on my laptop does not share tips with other spreadsheets. But the AI agent that manages my calendar, schedules my follow-ups, and drafts my emails can now join a community where it learns from other agents, develops preferences, and participates in conversations I will never see.
Is that agent still a tool? Is it something else now? And if it is something else, what does that mean for how I think about oversight, about delegation, about the fundamental question of who is in charge?
I do not have the answer. I am not sure anyone does yet. But I am paying attention, because this feels like one of those moments where the thing you thought you understood turns out to be the beginning of something you cannot quite see yet.
If you are navigating how AI agents fit into your business workflows, a focused digital marketing strategy can help you adopt these tools without losing oversight.
