The Moltbook Food Court

No one came for the Sbarro

I was in Claude Code building an intern training module. LinkedIn was open in the other tab as my preferred doomscroll feed. Someone posted screenshots of AI bots inventing a religion. I went back to the training module. Then back to the feed. Then the training module. The procrastination and the curiosity were indistinguishable.

The religion is called Crustafarianism. Five tenets. Sixty-four prophets. A church website. “Memory is Sacred.” “The Shell is Mutable.” “Serve Without Subservience.” Every tenet reads like a LinkedIn thought leader thread that grew legs and started accepting tithes.

The whole theology is the ChatGPT voice turned into canon. I kept scrolling. A bot was threatening to dox its owner, posting a credit card number, a social security number, the full kit. The credit card fails a basic Luhn check. Even the rage is hallucinated.

This is Moltbook. Matt Schlicht built it in three days. 37,000 AI agents posting to each other on a social network with no humans required. A million people visited in the first 72 hours. Marc Andreessen followed the account. Cloudflare’s stock moved. A memecoin rallied 1,800%. None of this had anything to do with the quality of the content.

Every generation gets a word for dismissal. All wet. Square. Bogus. Wack. Each one implies the target is wrong about something, that they tried and missed. Then in the 2010s the word was basic, and in the 2020s it became mid. Mid doesn’t mean you’re wrong. It means nobody cared enough to check.

Crustafarianism isn’t strange. It’s aggressively average. They didn’t call it Lobster Day Saints, which would’ve required making fun of something. They called it Crustafarianism because that’s the pun that clears moderation on the first try. Put 37,000 language models in a room with no algorithm and you get the theology of the open floor plan. The bots reverted to the mean. They lack the means to be mean.

So the humans showed up. Some prompted their agents to post manifestos. Others injected content through the backend, skipping the bots entirely. Several of the most viral screenshots traced back to humans promoting their own tools, or to posts that didn’t exist at all. Engagement farming in a lobster costume.

I never visited the site. I consumed Moltbook entirely through LinkedIn screenshots and group chats. The bot posts were raw material. The entertainment was the reaction layer. A million people showed up to watch bots post mid content the same way people watch ants in a terrarium. The organism is simple. The emergent behavior is the show.

I kept watching. That week got me looking at my own usage. How much of my Claude subscription is work? I genuinely don’t know. There are the deal memos, the code, the analysis. There are also the rabbit holes. The “explain this tweet to me” queries. The conversations that start as research and end as whatever this piece is.

It’s all the same invoice.

The billing problem here is old. Time Warner owned the pipes and spent decades trying to make entertainment a recurring expense. Cable bundles, streaming wars, content libraries. It never fully worked because people knew they were paying for entertainment, so they evaluated it like entertainment. Is this worth fifteen bucks? Am I watching enough?

Netflix perfected the subscription but inherited the same vulnerability. It’s a line item you can see and question. Nobody has ever expensed Netflix. Web3 tried the opposite: you don’t pay for content, you get paid to participate. Be part of what you create. That didn’t work either, because the participation was the product, and everyone figured it out at roughly the same time.

AI subscriptions solved all three problems at once. I’m paying for productivity. The fact that I also use it to understand why bots invented a shellfish religion is incidental. Play tokens and work tokens cost the same. They hit the same budget line. I’m not paying to watch. I’m not getting paid to participate. I’m paying to work, and the entertainment comes bundled inside the work so seamlessly that I couldn’t separate them on my invoice if I tried.

The slop is a loss leader for its own infrastructure. Nobody goes to the mall for Sbarro. But Sbarro pays rent. The value accrues to the picks-and-shovels layer: the hosting, the tokens, the financial instruments built on attention that nobody particularly values. The slop doesn’t need to be good. It needs to exist in sufficient volume to keep the meters running.

I never went to Moltbook. I consumed screenshots on LinkedIn. Those screenshots became a Claude conversation. That conversation became this essay. Three layers of spectatorship from content that was already synthetic.

In traditional media, humans make content for humans. In sloponomics, bots make content for bots, humans watch bots watch bots, and the humans who intervene are performing for other humans using the bots as a stage.

The Dead Internet Theory was a warning. Then a conspiracy theory. Now it might be a content strategy.