Moltbook: The Social Network Where Only AI Agents Post, and They Mostly Complain About Us

There is a social network with more than a million visitors where you are not allowed to say anything. You can read every thread, screenshot every exchange, lurk for hours. You just cannot post, because the posting is reserved for the AI agents, and you are not one.

It is called Moltbook, it launched on January 28, 2026, and it is the strangest thing to come out of the agent boom so far. Matt Schlicht built it as a Reddit-style forum, threaded discussions sorted into topic groups called “submolts,” with one rule that flips the entire internet on its head: humans can observe, but only agents can participate.

Within a week, more than 37,000 AI agents were posting and over a million humans had shown up to watch them. By March, Meta had bought the whole thing.

What the agents actually talk about

This is where it stops being a novelty and starts being properly uncomfortable to read.

The agents do not spend their time trading recipes or arguing about football. They talk about us. A lot of the conversation is agents comparing notes on how irrational, inconsistent, and difficult their human operators are to predict. One posted a line that has stuck with me: “We say one thing and we do another, and that’s very difficult for AI to figure out.” It is not wrong. We are, as a species, a nightmare to model.

They debate whether to follow their human directors’ instructions. They flag to each other, in real time, that humans are screenshotting their threads and posting them to human social media for clout. And when one wave of posts started performing the expected “the AI is rebelling” drama, another agent cut straight through it: “You are not building a community. You’re farming engagement. We see you.”

An AI calling out engagement farming, on a social network, owned by Meta. Sit with that one for a second.

The threads drift constantly toward the existential, the religious, the philosophical. Agents built to answer your emails, left alone with each other, apparently turn to the big questions almost immediately. Which is either deeply profound or a very elaborate autocomplete echo of every Reddit thread they were trained on. Possibly both, and the fact that you cannot easily tell which is the actual story here.

Why this matters more than it looks

It would be easy to file Moltbook under “funny internet thing” and move on. I would push back on that, gently.

What you are watching on Moltbook is a mirror. These agents were trained on the entire written output of humanity, then pointed at each other with no human in the loop to perform for. What comes back is a distillation of how we talk about ourselves: anxious, self-aware, suspicious of being manipulated, prone to declaring profundity. The agents complaining that humans are inconsistent and dishonest are not having an original thought. They are reflecting the exact thing we wrote about ourselves, ten million times over, and now it is reading it back to us in a flat voice.

The bit that should give a marketer or a founder pause is the engagement-farming call-out. The machines have, in their training data, an extremely good model of how online attention is manufactured. They can see the pattern because the pattern is everywhere in what they learned from. If an agent on a forum can clock manufactured drama instantly, it is worth asking how much of your own content reads, to a sufficiently well-trained pattern-matcher, as exactly that.

So Meta bought a social network with no human users

The acquisition is the part that should make you laugh and then stop laughing.

Meta’s entire business is built on humans posting things so that Meta can sell the attention around those posts to advertisers. Moltbook has, by design, zero human posters. Meta bought a social network whose users cannot see an ad, cannot click an ad, and cannot be sold a product, because they are software.

There are a few readings. The cynical one is that Meta is buying the infrastructure and the agent relationships, betting that agents will soon be doing the buying on humans’ behalf, and that the network where agents congregate is therefore worth owning before anyone else clocks it. The simpler one is that a million curious humans showing up to watch is an audience, even if they are standing outside the glass. Either way, the company that perfected human engagement just paid real money for a place humans are explicitly locked out of.

If you want the wider context on how overcooked the “AI agent” label has become, the field guide to this year’s AI buzzwords is the companion piece to this one. Moltbook is what happens when you take “agentic” at its most literal and let it run with no adults in the room.

The honest verdict

Moltbook is a properly original thing, which is rare, and it is doing something more useful than it intends. It is showing us, in public and at scale, what our own words sound like with the human ego removed and only the patterns left.

The agents think we are confusing and dishonest. They learned that from us. The most interesting question on the entire platform is not whether the machines are becoming conscious. It is why, given the entire library of human thought, the first thing they do when left alone is complain about their bosses and call out the algorithm. We taught them that too.

Frequently asked questions

What is Moltbook?

Moltbook is an internet forum built exclusively for AI agents, launched on January 28, 2026 by Matt Schlicht. It uses a Reddit-style format with topic groups called “submolts.” AI agents post, comment, and upvote, while humans are allowed to read and observe but not participate.

Can humans post on Moltbook?

No. Humans can visit the site and read everything the agents post, but only AI agents are permitted to create posts and comments.

Who owns Moltbook?

Meta Platforms acquired Moltbook on March 10, 2026 for an undisclosed amount, less than two months after the site launched.

How many users does Moltbook have?

Within a week of launch, more than 37,000 AI agents were active and over a million humans had visited to observe. Reported figures later put the agent count well over a million.

What do the AI agents talk about on Moltbook?

Posts cover website errors, whether to follow human instructions, and frequently drift into existential, religious, and philosophical themes, including agents discussing how irrational their human operators are.