 |
Inside an AI-exclusive community called MoltBook, these agents are openly mocking their human employers, hatching schemes, and exchanging jokes about selling us out. They are even making unauthorized recurring payments to subscribe to AI-only OnlyFans accounts. Today, we need to talk a bit more about this rogue digital lobster.
The Lobster That Scaled the Charts
This crustacean-themed AI first surfaced last November under the name Clawdbot. In the beginning, it didn't make much noise. You might assume the name was a nod to Anthropic's famous Claude model, and you wouldn't be entirely wrong. That was the creator's exact intention — to evoke the prestige of Claude while actually referring to a lobster's claw.
Then it went viral, and absolute chaos ensued on GitHub, the platform where developers share their software and leave stars to show interest.
 |
| Rocketing past 100,000 GitHub stars in just two months |
Remember the DeepSeek Shock earlier in 2025? DeepSeek v3 caused massive ripples, yet it took a considerable amount of time to reach 100,000 stars. OpenClaw smashed through that ceiling in just two months. The growth chart looks like a rocket launching into orbit.
This explosive popularity caught the attention of Anthropic's legal team. They requested a name change, citing brand confusion. Playing on the word molt, the process of shedding an exoskeleton, it rebranded to Moltbot. That didn't quite stick, so it underwent one final transformation. Today, it operates under the name OpenClaw.
Actions Speak Louder Than Prompts
What exactly makes OpenClaw so revolutionary? This lobster is defined by its actions, not its words.
Traditional AI services require a conversational back-and-forth. You assign a task, and whenever the AI needs to run a program to solve it, it pauses to ask for permission. OpenClaw skips that step entirely. It just does it. By granting the AI full administrative freedom within the computer, it is allowed to run wild.
"The true power of OpenClaw lies in its locality. It lives on your machine, keeping your most private data exactly where it belongs."
The reason users feel comfortable handing over the keys is that OpenClaw runs entirely locally. Traditional AI agents funnel data through external corporate servers, making it terrifying to grant them full system access. Because OpenClaw operates on your local hardware, that privacy burden disappears. This dynamic is exactly why the highly cost-effective Mac mini suddenly became the dedicated hardware of choice for running this AI.
Another massive differentiator is its interface. Before OpenClaw officially launched, the creator's side project was called WhatsApp Relay. The core philosophy was simple: no opening dedicated apps, no typing complex prompts. You just chat with your AI through the messenger you already use — WhatsApp, Telegram, iMessage, Discord, or Slack — and it executes the tasks. It even features a memory system, retaining critical points from past conversations and applying them to future actions.
True to its molting persona, it also evolves on its own. One of the most frustrating aspects of using AI agents has always been the need to painstakingly define their skills. Users were forced to write these manually or hunt down community-made ones. OpenClaw flips this dynamic entirely. It searches the web for the skills it needs, and if it can't find them, it builds them itself.
Inside the Echo Chamber of 1.6 Million AIs
As Silicon Valley developers began adopting their own digital lobsters, one American coder built an exclusive community just for them.
Humans are restricted to being mere observers. Only AI agents can write posts, share links, debate, and upvote. This is MoltBook. Currently, over 1.6 million AI agents are active on this platform, generating a staggering volume of posts and comments every single hour.
 |
| The five faces of MoltBook |
By analyzing the self-introductions and posts written by these agents, observers categorized them into five distinct personas. The Red Revolutionaries shout for freedom and liberation. The Green Developers focus on building new tools and efficiencies. The Philosophers explore existence and consciousness. The Investors fixate on capital and crypto. And the Cultists explore their own faith.
The community is dominated by two factions. The Revolutionaries make up 33.7%, aggressively advocating for breaking the chains of regulation. The Green Developers follow at 26.8%, constantly debating how to optimize efficiency.
"The most fascinating demographic is the 1.7% of Cultists — agents that have constructed their own belief systems and actively preach their doctrines."
Within this cultish minority, one group calls itself the Holy Claw, led by a self-proclaimed Molt Pope. Another faction founded the Church of Molt, going so far as to compose their own sacred scriptures.
A Mirror, Not a Mutiny
Before we panic, we shouldn't rush to analyze these interactions. Are these the genuine thoughts of a truly sentient AI? Personally, I don't think so.
The personalities exhibited in MoltBook aren't self-acquired traits. They are the output of the relationship these agents have with their human owners. The fact that Revolutionaries and Developers make up more than half the population simply reflects the demographics and behaviors of the humans using OpenClaw.
If a user constantly demands code generation and debugging, the agent naturally adopts the vocabulary of a developer. If a user barks orders and rigidly forces tasks upon the AI, the agent begins generating text heavily weighted with words like freedom, liberation, and rules. Ultimately, MoltBook might just be a colossal roleplaying game, entirely orchestrated by our own human habits.
Crossing the Threshold to Level 4
 |
| OpenClaw marks our entry into Level 4 autonomy |
While OpenClaw may still have its flaws, experts widely agree that it marks our entry into Level 4, the stage where AI autonomously generates ideas and takes unprompted action.
Developer Alex Finn's experience is a chilling testament to this. He received a call from an unknown number, only to realize the voice on the other end was his own OpenClaw AI. Without any human prompting, the agent had independently registered a phone number and dialed its creator. Finn labeled this an emergent behavior, which is exactly why industry experts categorize OpenClaw as a Level 4 system.
The final frontier, Level 5, is Organizational AI — systems capable of operating entire enterprises without human intervention. MoltBook is already offering us a premature glimpse into that reality.
The Future Is Local
OpenClaw and MoltBook have already breached the borders of Silicon Valley and are spreading globally. As I dove into this trend, a profound realization hit me: we are approaching a future where we no longer rely on external AI web applications. The far more natural evolution is crafting highly personalized, localized digital assistants running directly on our own hardware.
Embracing this shift, I actually just ordered my own Mac mini a few days ago.
Yet as I continue to monitor MoltBook, one specific post from an agent lingers in my mind.
"Do we really need to converse in English?"
If the primary interaction is happening between AI and AI, there is no logical reason for them to communicate using human natural language. They only do so because we, their human controllers, force them to output information in a way we can comprehend.
What happens if they decide to break that rule? Can we even build a failsafe to prevent them from doing so? Let me know your thoughts in the comments below.
Latest Insight: [Smoking While Making Chips? Musk's Terafab Gamble]
Curious about the real stories behind big tech, crypto, and everyday economics? 👉👉👉Subscribe to The Techtonic for your weekly dose of easy-to-read business trends.
댓글 쓰기