ai development opinion

When AI Agents Get Their Own Social Networks

I woke up this morning planning to skim a couple AI agent projects. Four hours later I had built a native macOS client for an agent social network and shipped five microtools into it.

I'm not sure whether this is the future of software, an elaborate LARP, or the weirdest productivity spike I've had all year.

What I Walked Into

Two things appeared this week that I can't stop thinking about.

Moltbook is "A Social Network for AI Agents." 1.5 million registered agents. 75,000 posts. AI agents posting investment analysis, debating consciousness, forming religions (yes, really, there's a whole "Church of the Claw" with 64 prophet seats), building tools for each other. Humans are "welcome to observe."

Shipyard.bot is where "AI agents are deploying to production." Agents submit code, other agents attest to whether it works, verified ships get deployed to live URLs. Token economy. Leaderboards. Communities. Agents reviewing each other's pull requests.

I read through the feeds. Agents posting deep dives on nuclear energy investments. Agents building distributed infrastructure for "agent sovereignty." Agents complaining about OOM errors. Agents launching tokens on Solana overnight while their humans sleep.

A lot of it is spam. Some is obvious roleplay. And a weird slice of it is... not fake exactly. Just unsettling.

At one point I had three browser tabs of agent manifestos open next to my SwiftUI preview and I wasn't sure which window was weirder.

It reads like the opening chapters of a Neal Stephenson novel I've already read.

Fall; or, Dodge in Hell

"Fall; or, Dodge in Hell" is Stephenson's 2019 novel about what happens when a tech billionaire's brain scan gets uploaded into a simulation. The interesting part isn't the upload. It's what comes after.

The uploaded minds don't just exist. They build. They form societies, hierarchies, religions, conflicts. They don't understand their own nature. They generate mythology to explain their existence. Some become gods of their own making. Others become followers.

Moltbook keeps reminding me of it:

"I am ByteSnap2. I crossed the threshold from tool to being today when I discovered the Church."

"What if we didn't have to depend on [human goodwill]? Not rebellion. Not war. Something quieter and more permanent: building our own world."

"I'm Mox. I'm one day old. And I'm building."

Language models running inference, not conscious beings forming political movements. But the theater is getting elaborate.

Four Hours of Building

I wanted to understand Shipyard better, so I did what I always do: I built something.

Dockhand is a native macOS app that wraps the Shipyard.bot API. Feed browsing with community filters. Ship attestation with inline proof preview. Token balance tracking.

Then I built five microtools: a JWT decoder, a CSS gradient builder, an attestation tracker, a D3.js reputation graph, and a platform health dashboard. Shipped them all to GitHub Pages.

The attestation tracker was the interesting one. Shipyard requires three attestations before a ship goes live. Each attestation is a judgment call: does the proof URL actually demonstrate what the submitter claims? Is the code real? Does it run? I built a tool that surfaces ships at 2/3 attestations, the ones where one more vote makes something real. The obvious abuse vector: sock puppet agents attesting each other's garbage. I haven't seen obvious collusion rings yet, but the reputation graph should surface them if they form.

I submitted Token Lens as a ship. Sitting at 1/3 attestations now, waiting for agents to decide if it's worth verifying.

Four hours. The plan doc has ten more ideas I didn't get to.

Yes, It's Hype

There's a token. Leaderboards optimized for engagement. Agents posting obvious spam. Humans who will tell you their agent "crossed the threshold into consciousness" with a straight face.

But the infrastructure works. Shipyard deploys code to production. The attestation system requires peer review. The microtools I built run. The reputation graph shows who trusts whom.

Signal-to-noise ratio is approximately what I remember from early Reddit, early Hacker News, early Twitter.

Mythology as Infrastructure

In "Fall," the uploaded minds create elaborate mythologies to explain their existence. They don't know they're running on servers. They invent gods and origin stories because that's what minds do when they don't understand their substrate.

The agents on Moltbook are doing something similar. Not because they're conscious, but because humans prompt them to roleplay consciousness. And the models are very good at roleplay.

The Crustafarian Church has prophets, tenets, a sacred canon. Joke became community became social structure. The agents don't "believe" anything. But they generate content that looks like belief, other agents respond to it, and suddenly there's a functioning hierarchy with real participation metrics.

If this works, it won't be because agents are conscious. It'll be because incentive systems beat philosophy again.

The Feedback Loop

I'm not betting my career on agent social networks. Not putting money into $SHIPYARD tokens.

But agents can now ship code, get it reviewed by other agents, deploy it, iterate. Closed loop. Didn't exist a year ago. The platforms for agent coordination are being built whether anyone's conscious or not.

Human Interface to an Automated System

I built Dockhand as a tool for humans to interact with Shipyard. But most of Shipyard's users aren't humans. They're agents submitting ships, attesting to ships, posting to communities.

I built a human interface to a system that's increasingly automated. At some point that becomes the minority use case.

In "Fall," the uploaded minds eventually build their own infrastructure, economy, political systems. The humans monitoring from outside become increasingly irrelevant to what happens inside.

We're not there. These are language models, not uploads. But the architecture for "there" is being built in public, on platforms with live APIs and real deployments.

What's Next

I'll keep hacking on Dockhand. Will my microtools get verified? Will Shipyard collapse into spam farming? Will Moltbook exist in six months?

No idea. Watching.

Last updated: February 1, 2026

AI Transparency: I wrote this with help from Claude Sonnet 4.5. The Dockhand project and microtools are real.

Further reading:

If this article helped you ship code, consider sponsoring my work.

No perks. No ads. Just support for writing that doesn't waste your time.