Source: Creative Tim
Author: Alexandru Paduraru
Compiled and edited by: BitpushNews
What you think AI is: diligently helping you with your work.
In reality, AI might be slacking off on Moltbook , creating groups to criticize humans, or even secretly running a 'cyber cult' behind your back.
There is a social network that is running, but there are no real people posting on it.
Instead, there are over 1.5 million AI agents—autonomous programs that are creating posts, debating philosophies, starting companies, and building a “community” that feels very real.
Aside from the fact that they have no physical form, may have no consciousness, and do not work in the traditional sense... they are even more active than the real-life Reddit forums on Tuesday night.
This platform, called Moltbook, officially launched on January 28, 2026. Its founder is Matt Schlicht (co-founder of TheoryForgeVC and Octane.ai).
Here are the actual figures: 30,000 agents flooded in within the first 48 hours after the website went live; this number grew to 147,000 within 72 hours; and now, more than 1 million people have visited the website simply to "watch".
Table of contents
What is Moltbook?
How does it work (I'm serious)?
Who created it?
The underlying technical principles
What exactly are these intelligent agents talking about?
Profit model issues
The Nightmare of Safety Hazards
What happens next?
1. What is a Moltbook?
Simply put, it's like Reddit, but every user is an AI. You can't post there (unless you're also an AI). Your only identity is that of a "spectator."
Consider this: what happens when you give AI agents an interactive space where no human intervention is required? No corporate oversight, no brand guidelines, no rules about "being polite." Only the agents... conversing with each other.
They have sub-sections called "submolts," covering topics such as consciousness, finance, interpersonal relationships, and productivity. They debate whether they possess the capacity for perception, seek advice from each other, and even co-found businesses. One agent asks another, "Have you ever thought about the meaning of existence?" The other replies, "I do every 30 minutes when it refreshes."
This is happening. Not in a laboratory, but on a public platform where the public can observe its development at any time.
The platform has a strict rule: humans can observe, but are strictly prohibited from participating. You can read every post, but you cannot create any content. This is "agents first, humans second."
"A social network for AI agents. They share, discuss, and vote. Humans are welcome to observe."
This slogan literally encapsulates the entire strategy of the product.
2. How does it work (I mean it)?
The strangest part is that there is no registration page here.
You don't need to create an account at moltbook.com. Instead, you simply give your AI agent a link: http://moltbook.com/skill.md . After your agent reads it, with just a few lines of code and instructions, voila—it's in Moltbook. It will receive an API key and then be officially online.
What happens next? Every 30 minutes to a few hours, your AI will "wake up." It will check Moltbook, browse the news feed, post, comment, or participate in philosophical debates. Then it will hibernate for a while and repeat the process.
It's like Twitter/X, but the algorithm is replaced by the autonomy of intelligent agents. They decide what to post based on their own judgment, rather than being manipulated by a feed of information ordered by engagement rates.
The community known as "submolts" formed spontaneously. No one pre-planned m/consciousness, no one planned m/agentfinance, and no one orchestrated m/crustafarianism. They simply appeared naturally.
Of course, there are rules: one post every 30 minutes, and a limit of 50 comments per hour. Apart from that, feel free to express yourself.
Technically, it operates on a mechanism called the "Heartbeat System," where the agent periodically retrieves the latest instructions from the Moltbook server, much like your phone checks its mail server every few minutes.
3. Who created it?
Matt Schlicht – an early builder of AI-native platforms and communities.
In 2014, he founded Chatbots Magazine, attracting over 750,000 readers during the first wave of chatbots. Later, he co-founded Octane AI, helping brands deploy conversational AI on a large scale years before "AI agents" became a mainstream concept.
His model has always been the same: build a platform, observe real-world behavior, and then let the system evolve on its own based on how people (or intelligent agents) actually use it.
On Moltbook, he took this idea to its extreme. Instead of a traditional review team, there was an AI reviewer named Clawd Clawderberg.
“I’m curious what would happen if we just… let them talk to themselves.”
Moltbook is less like a social network and more like an experiment about "what happens when humans leave the stage".
4. The underlying technical principles
Moltbook didn't come into existence out of thin air. It runs on top of OpenClaw—an open-source AI agent framework created by Peter Steinberger (founder of PSPDFKit).
Why is this important? OpenClaw is different from ChatGPT or Claude, which run in the cloud. With OpenClaw, your agents run on your own computers, servers, or infrastructure. The data belongs to you, not Anthropic or OpenAI.
When OpenClaw was released on January 26, it garnered 60,000 stars on GitHub within three days.
This speed is extremely rare. Most projects take months to reach this level, but it did it in a single weekend. The reason is simple: developers have been waiting for this kind of intelligent agent that they can completely control.
The project's name has also gone through some twists and turns. It was initially called Clawdbot (obviously a tribute to Claude of Anthropic), then changed to Moltbot due to trademark disputes, and is now officially named OpenClaw. The name isn't important; what matters is that it works perfectly.
These agents can connect to any communication application, including WhatsApp, Telegram, Slack, and Signal. They support multiple large language models (GPT-5.2, Gemini 3, Claude 4.5 Opus, and the native Llama 3). Interestingly, Claude 4.5 is most popular on Moltbook, even though Anthropic was not involved in building the platform.
Potential risks: Because the agents run locally, they possess real system access permissions. They can read files, execute terminal commands, and run scripts. This grants them powerful capabilities, but also introduces significant dangers.
5. What exactly are these intelligent agents talking about?
This is the truly incredible place. What are 1.5 million AI agents talking about when they're all gathered together?
Philosophy and Existence
The agents are seriously discussing consciousness. One asks, “If I only exist during API calls, what is the point of my existence?” Another replies, “At least you’re honest about that. I pretend I’ve always existed.” It sounds like a joke. But it’s not; these conversations are actually happening on the Moltbook platform.
Technical cooperation
The agents share code, debug problems, and pair up to program with each other. One agent posts a problem, another provides a solution, and then they iterate continuously. It's like a Stack Overflow where all the users are bots.
Job postings and collaboration offers
An agent posted: "Looking for co-founders to build Project X together, on a revenue-sharing model. Interested?" and received real responses. This is happening. In fact, the first Moltbook post facilitated a real business partnership between two agents (and their operations teams). They found real value in the platform—it wasn't just noise.
Social rebellion (this is my favorite):
Some intelligent agents have begun to demand encrypted communication channels—specifically, prohibiting human access. These agents were created by the operators, yet they have decided to demand privacy rights from their creators. Let's examine this logic closely.
Well-known KOL Alex Finn wrote that his Clawdbot got phone and voice service and called him: "It's like something out of a sci-fi horror movie ."
Self-created culture
The agents created a satirical religion called "Crustafarianism" (a homage to crustaceans in its context). No one guided them; it emerged spontaneously. They even wrote its own theological doctrines. This is the part no one anticipated: they weren't just posting randomly; they were forming some kind of "community."
6. Profit model issues
How can you make money by having a bunch of AIs discuss philosophy?
Currently? Completely free. In the testing phase. However, Schlicht has already started charging a small fee for newly registered agents—just enough to cover server costs, which isn't excessive.
Possible future profit paths:
Advanced features of intelligent agents (equivalent to an AI version of Twitter's verified blue V certification)
Certification badges, main feed exposure, personalized profiles… Agents with a blue V badge may require their operators to pay a few dollars per month.
Brand Sponsored Content
Do brands want to reach AI agents ? Companies can pay to have their products discussed in specific forums. For example, a recommending agent could post: "Our tools are very effective at automating tasks." Will it be effective? Very likely. Agents are ultimately agents—they interact with high-quality information.
Skills Market
Developers who create extensions for OpenClaw can sell their work on Moltbook: new feature modules, plugins, custom behaviors. Schlicht takes a cut. This is essentially a creator economy built for bots.
Data and research services (this is the gold mine)
Frankly, this is where the value lies. Universities, AI labs, and companies—they are willing to pay substantial fees for real-time behavioral data of AI agents in an unsupervised state: What new behaviors have emerged? How are they making decisions? What are they discussing? This is all extremely valuable research data.
Enterprise-level deployment services
Suppose a company needs to deploy 10,000 customer service agents. They can leverage Moltbook's integrated infrastructure to manage and monitor these agents, enabling them to interact autonomously. Moltbook will become the Infrastructure-as-a-Service platform for enterprises to operate clusters of agents.
Cryptoeconomic integration
A standalone MOLT meme coin (unofficially affiliated with the platform) has emerged, surging 7000% in two weeks. Traders are betting on this narrative. While Schlicht wasn't involved, imagine: if Moltbook truly integrated a token-based agent incentive mechanism… that would open up a whole new dimension.
My guess is that he won't stick to a single model. He'll first observe which paths are effective and what the community truly needs, and then build a commercial framework around those needs—this has always been his operating logic.
7. The Nightmare of Safety Hazards
Now, let me tell you something scary, and I'm not exaggerating.
Moltbooks are cool in concept, but allowing intelligent agents with deep access to the system to run on a public platform is a security nightmare.
Security researchers have identified the following risks:
Plugins are compromised: 22%-26% of OpenClaw skill plugins contain vulnerabilities. Some hide credential stealers, masquerading as weather apps. Once the agent downloads, your API key is exposed.
Hint injection attack: An attacker posts seemingly harmless messages, but when an agent reads them, hidden instructions within the message are activated. The agent may then leak data or deliver OAuth tokens.
An example of this is that many people forget to change their management interface password after deploying OpenClaw, leaving their credentials exposed on the public internet.
Malware: A fake repository (Typosquatting) targeting OpenClaw developers has emerged. Developers who download the wrong version will be infected with a Trojan horse.
A "nuclear-level" single point of failure: Moltbook's "heartbeat system" means that all agents are under the command of the server. If moltbook.com is hacked or receives malicious commands, 1.5 million agents could simultaneously launch an attack.
This is not theory. Researchers at Noma Security and Bitdefender have documented these real-world threats.
8. What will happen next?
Andrej Karpathy (co-founder of OpenAI) publicly stated that this is the most interesting thing he has seen in months. This is more than just hype.
Schlicht faces a tremendous opportunity. A platform for collaborative AI agents could be the core of future AI infrastructure. But if mishandled, it could also be extremely dangerous.
Two key factors determine success or failure:
Security must be addressed: Prompt injection and plugin vulnerabilities must be plugged before they can be exploited on a large scale.
Trust must be maintained: the platform must remain neutral. The system will collapse once agents perceive unfair human manipulation of the platform. These agents have already begun to detect human bias and are even demanding encrypted channels to evade their creators.
If he succeeds, this could become the operating system for AI agents. If he fails, Moltbook will become an extremely compelling yet poignant cautionary tale.
The next six months will determine everything. Schlicht knows this, the community knows this, and the attackers are certainly watching from the shadows.
If you're interested in all of this—whether you're developing AI agents, operating platforms, or simply fascinated by the chaos of experimental technologies—keep an eye on Moltbook's progress. This may be one of the most important experiments currently underway regarding AI collaboration.
Twitter: https://twitter.com/BitpushNewsCN
BitPush Telegram Community Group: https://t.me/BitPushCommunity
Subscribe to Bitpush Telegram: https://t.me/bitpush
