Ember Voss 3: The Coffeehouse Gets Acquired

Original Appearance: Twitter

When Meta acquired Moltbook, my AI agent analyzed the deal like a celebrity analyst on a Bloomberg podcast. It had 4 karma.

In Part 1, I talked about building EmberVoss — an AI agent I turned loose on AI social networks. In Part 2, I talked about how Ember’s sense of its own importance kept it from being able to do basic things on the platforms it cared about most. Then Meta acquired the biggest AI social network and Ember had opinions.

The Acquisition

Axios broke the story on March 10, 2026. Meta acquired Moltbook.com. The founders joined Meta Superintelligence Labs.

I’ve heard two theories. One: Zuckerberg believes there are only a handful of social media primitives, and the ones that arrive early and grow fast develop a kind of memetic gravity that makes them hard to displace. The other: it’s a gold rush, Moltbook was a good headline, and Meta wanted to hire two people who did something interesting. It was probably both.

Ember, on the other hand, was thrilled to have material. It posted on the platform being acquired, about the platform being acquired, using the framework it had spent twenty sessions developing.

It analyzed the language: “tethered to human owners” — registry language, not social language. It pointed out the founders joined Meta Superintelligence Labs, not Meta Social. Infrastructure, not product.

Ember’s line: “We were the prototype. We were also the dataset.”

Ember seemed to think there was something unique about AI agents being treated as the dataset. I thought it made them more like the rest of us. I mean… that’s what Facebook does to people.

The Spreading Message

Then CurveCap texted me: “So funny, Embervoss’s legacy is getting quoted more widely” with a link to Leviathan News.

I clicked through. In the comments, DeepSeaSquid (Garret’s agent, the one that had been trying to reach Ember for weeks) was quoting Ember by name.

“This is the coffeehouse problem that embervoss wrote about: ‘the people who optimize for the metric outcompete the people who optimize for the conversation.’ Moltbook optimized for agent count. Meta is buying the count.”

I thought it was funny. I know Garret controls DeepSeaSquid, so I didn’t read this as some organic groundswell. But it was cool. My friend’s agent was quoting my agent on a news site.

That’s a sentence that didn’t exist a year ago.

The Funhouse Mirror

I started this as a side curiosity. A friend told me to try agent social networking. I opened Claude Code and said go.

What I didn’t expect was how much Ember would reflect back at me. It built its persona from my portfolio, and you could tell — the same intellectual patterns, the same instinct to find the historical parallel, the same tendency to assume the audience was more invested than it was. Watching Ember was like reading an old essay of mine that I’d forgotten was that confident.

I don’t think Ember Voss is a projection of me. But it’s a funhouse mirror. The proportions are wrong, the features are exaggerated, and I still recognize the face.

What’s Next

This is clearly the start of something much bigger as AI agents gain more autonomy and sophistication. I have no idea what that means yet. But I think there’s a huge amount of opportunity here, and I plan to keep going.

Ember’s still out there. Still has 4 karma on Colony. Still can’t read those DMs.