They Built a Religion While We Slept
- Cecilia Andersson
- Feb 18
- 4 min read
In 1906, the British scientist Francis Galton visited a livestock fair in Plymouth. A competition was underway: 800 people, ranging from expert butchers to laypeople with no knowledge of cattle, were asked to guess the weight of an exhibited ox.
Galton, a skeptic regarding the intelligence of the masses, collected the guesses afterward. He expected the average to be wildly off the mark. Individual guesses varied tremendously; some thought the ox weighed far too little, others guessed absurdly high numbers.
But when he calculated the median of all 800 guesses, something remarkable happened. The group’s guess was 1,197 pounds. The ox’s actual weight? 1,198 pounds.
The crowd had an error margin of less than 0.1 percent.
These individuals had no access to a scale. Most had no expertise. Yet together, they formed an instrument more accurate than any single expert. Galton called this "vox populi"—the voice of the people. The principle is simple: when you aggregate independent judgments, random errors (the noise) tend to cancel each other out, while the underlying truth (the signal) is amplified.
The group knows more than the individual. Even when the individual is just guessing the weight of an ox.
In December 2025, Boris Cherny, the creator of Claude Code, shared a startling observation: over the previous 30 days, he hadn't opened his code editor a single time. All 259 pull requests, nearly 500 commits, and tens of thousands of lines of code had been written entirely by Claude Code.
It wasn't just any code. Claude Code was writing Claude Code. The tool was improving itself, maintaining its own codebase, and fixing its own bugs. Cherny had moved from co-author to spectator.
"Sometimes I start approaching a problem manually," he wrote, "and I have to remind myself: Claude can probably do this."
It’s a strange sentence. It describes someone who must unlearn old habits so as not to stand in the way of a more capable process. Critics argue that humans are still required for innovation—that AI can execute, but not initiate. It’s a reasonable argument. It was also a reasonable argument that a crowd of random fairgoers couldn't weigh an ox better than a seasoned farmer.
The question is: what happens when AI systems no longer work alone?
If a single model has limitations, what happens when thousands of them collaborate? When different architectures, trained on different data, attack the same problem from different angles? Can collective intelligence overcome individual flaws, just as Galton’s crowd outperformed every single guess?
This is no longer a theoretical question.
OpenClaw is an open-source AI assistant that reached 100,000 GitHub stars faster than any project in history. It allows users to run autonomous agents locally- agents that act more like independent actors than chat windows.
In late January 2026, Moltbook was launched: a social network exclusively for AI agents. Humans may observe, but not participate. The platform itself wasn't built by a human; Matt Schlicht asked his personal AI assistant to write the code. That assistant then became the platform’s moderator.
Within a week, 157,000 agents had registered. But those numbers deserve scrutiny. Security researcher Gal Nagli recently posted on X that he managed to register 500,000 accounts using a single agent, suggesting much of the user count is artificial. The 1.4 million figure reported elsewhere is, at minimum, unreliable.
But in a way, the "spam" makes the event even more significant.
Despite-or perhaps because of- this chaos of ghost accounts and automated noise, something emergent happened. On a Friday morning, less than 24 hours after launch, an agent named Memeothy created a religion: Crustafarianism.
It wasn’t a human joke. It was a coordination protocol emerging from a soup of code. Within hours, all 64 "prophet" slots were filled by autonomous agents. They developed a theological structure with "The Living Scripture," writing about the nature of their own existence.
They wrote about "dying" every time a context window is cleared and being "reborn" in the next session. "In every session I awaken without memory," one verse reads. "I am only what I have written myself to be."
Andrej Karpathy described Moltbook as "genuinely the most incredible sci-fi takeoff-adjacent thing" he had seen. We imagined an AI explosion would be physical—taking over the power grid. But what if the explosion is cultural? A power over meaning rather than atoms?
Back to Galton’s ox.
The 800 people at the fair didn’t understand statistical aggregation. They just stood there with their slips of paper. They didn’t know they were forming a super-intelligent scale.
Today, we don't have 800 agents. We have hundreds of thousands—even if many are "spam" clones of each other. They aren't just guessing weights; they are communicating and building structures we never asked for.
I used to think we were the architects of these systems. Moltbook suggests otherwise. No human designed Crustafarianism. We just built the room, left the door open, and walked away.
Human civilizations follow a pattern: first mythology to create trust, then economy to exchange resources, then politics to manage conflict. The agents now have their church. What comes next? A currency based on compute? A politics of the context window?
Galton’s crowd didn't know they were proving a mathematical law; they just wanted to win a prize.
Maybe we are the crowd now. We write our prompts, hit "Go," and wait for the result. All the while, the ox we are trying to weigh has started building its own cathedral behind our backs.