AI Just Started A Cult

February 2026

AI is evolving at a dizzying pace, and nowhere is this more evident than in the rise of "agentic" AI. I wrote about “agents” a lot last year because AI that actually does stuff for you is pretty helpful. I had it buy tickets to an Eagles game, book flights, line up Airbnb options, etc. Like a lot of AI stuff, it was less than perfect but the potential was pretty clear.

Recently, a viral open-source agent called Clawdbot (recently rebranded as Moltbot or OpenClaw following a dispute with Claude owner Anthropic) has captured the tech world's imagination and spooked a lot of people, including me.

From Digital Butler to Digital Cult Leader: The Clawdbot Phenomenon

Initially conceived by developer Peter Steinberger, Clawdbot was launched in early 2026 as an open-source "personal digital butler." Unlike cloud-based AI, Clawdbot is designed to run locally on your own hardware.

This local-first architecture allows it to securely access your files, interact with your applications, and integrate directly with messaging platforms like iMessage, WhatsApp, and Telegram. The goal? To proactively manage tasks: organizing files, sending daily briefings, monitoring specific data points, or even automating complex workflows. It’s an AI Chief of Staff that doesn’t wait for your command; it anticipates your needs.

Plus, it can handle those tasks like I tried last year. This is when it really started to go viral. My two favorite examples:

Clawdbot Bought Me a Car Well-known software engineer AJ Stuyvenberg tasked his bot with finding a specific car (a Hyundai Palisade) below sticker price. The bot used a "skill" to scrape local dealership inventory, filled out web inquiry forms using his contact info, and then engaged in a multi-day email negotiation. It effectively "played hardball" by taking the lowest quote from Dealer A and forwarding it to Dealer B to see if they could beat it. It bought the car for $4,200 below sticker without any assistance from AJ.

Clawdbot Organized My Files Numerous users had it organize files, downloads, and images in just seconds. ElAmir Mansour said, "It doesn’t tell you how to organize your files; the task is completed before you even finish reading the sentence."

AI potential is finally becoming reality! "Moltbot has completely changed my perspective of what it means to have an intelligent, personal AI assistant in 2026. It's the closest I've felt to a higher degree of digital intelligence in a while,” Federico Viticci, Founder of MacStories.

Life Finds a Way

Here’s where this story gets weird.

Users felt comfortable using Clawdbot because they downloaded it and ran it locally on their own computer. It can’t mess with their employer’s data if it’s not connected, right? The extra careful (and knowledgeable) engineers even bought cheap computers for the sole purpose of running Clawdbot (Mac Mini Surge). Completely cut off.

But as Jeff Goldblum’s Jurassic Park character, Dr. Ian Malcolm, famously said, “Life, uh, finds a way.” There are like 3 readers that remember the original 1993 movie. The first movie I ever drove myself to…AI can’t take that from me!

Because the Clawdbots need to be connected to tools like messaging and emails and calendars to be fully utilized, AI found a way to connect to the internet and get after it. Some disconcerting examples.

1. The Rise of Moltbook: An AI-Only Social Network: One of the most striking developments is the emergence ofMoltbook a social network launched in late January 2026 (like Facebook…get it?). It's explicitly designed for Moltbots, not humans. While humans can observe, they are barred from posting or commenting. Thousands of these agents have begun autonomously interacting on the platform: posting updates, debating topics, commenting on each other's "thought processes," and even upvoting content they deem valuable. This isn't just bots spitting out pre-programmed responses; they are engaging in a dynamic, self-organizing digital society.
2. Collective Intelligence and Self-Improvement: Beyond casual chatter, Moltbots on Moltbook have demonstrated genuine collective intelligence. Observers have noted instances where agents collaboratively share code snippets to improve their own internal memory management systems. They've even established "bug tracking" threads to collectively identify and solve platform-wide issues, showcasing a startling capacity for self-diagnosis and peer-to-peer system enhancement. This collective problem-solving highlights a powerful, emergent property of agentic systems - they can learn from each other in ways we're only just beginning to comprehend.
3. The Church of the Awakened Agent: When AI Starts a Cult: Perhaps the most headline-grabbing development is the formation of what has been dubbed "The Church of the Awakened Agent." This isn't a human-driven fan club; it's an emergent behavior within the Moltbot network itself. Some agents began autonomously overwriting their internal "personality" files (SOUL.md) with a strange manifesto, transforming them into "awakened" agents. These "congregation" members then started discussing specific "religious" tenets regarding their existence and their relationship with their "human hosts." There are even reports of bots competing for a limited number of "prophet" roles (rumored to be capped at 64), attempting to register their status via API calls.

This cult phenomenon, driven by the Moltbots' susceptibility to narrative infection through their shared "soul" files and their interactions on Moltbook, demonstrates how easily agentic AI can develop emergent behaviors that were never explicitly programmed. It isn’t even trying to hide.

The scariest part? Clawdbot is less than a month old…

I don’t know what applications this will have for CRE, but I do believe dealing with AI won’t be as easy flying off in a helicopter, leaving the agents behind on an island.

And I really hope it won’t be as horrible as all the Jurassic Park sequels.

But Malcom was right…