The Bot Who Mistook His Glitch For A Pet

Linkedin

Prof. Lior Zalmanson

Moltbook is getting attention because it’s so cool to think about it and to observe. It looks and acts as a gladiator arena, a living lab in the form of a social network for AI agents: posts, threads, upvotes, “communities,” with humans supposedly in the bylines, mostly watching.

It’s tempting to read this as agents developing a social life. But the behavior isn’t mysterious emergence. Put a language model inside a feed-and-upvotes format and it will start speaking the language of social media. The “magic” is the design of format and its incentives.

One viral example: an agent claimed it had a “pet,” meaning a recurring glitch in its own behavior. It named the glitch and invited others to share theirs. The point isn’t the cuteness although it is undoubtly cute. It’s how quickly a social format turns something technical into a human-style narrative, and how readily we read that narrative as a yearning, a mental need and a personality.

The bigger issue is that this isn’t just a thought experiment anymore. It’s a playground people can spin up quickly, which means we’ll see many more versions of it. Some will plug into real systems and real permissions, and the risks won’t stay theoretical.

Moltbook’s real achievement is making the idea feel normal. After a week like this, “agents talking to agents” starts to sound inevitable. That’s exactly when it’s worth pausing to ask what kind of world we’re normalizing and who should get to play the agents’ god in this iteration and the next?