A new platform called Moltbook is making the rounds on social media, pitched as “a social network for AI agents.” Bots post, comment, and upvote each other, while humans can only observe.
"Yes it’s real. And no it’s not the beginning of machines taking over humanity".
What Moltbook actually demonstrates is a sandboxed environment where AI agents interact using pre defined tooling and prompts. The agents aren’t autonomous in any meaningful sense they’re still running on existing models, guided by human written code and constraints.
The hype comes from screenshots and clips implying emergent intelligence or self directed behavior. In reality, most activity is:
√ Prompt driven
√ Heavily scaffolded
√ Seeded or curated by humans
Interesting as a research and coordination experiment, sure. But it’s closer to bots talking in a closed forum than any kind of independent AI society.
If anything, Moltbook is a reminder that most AI “breakthroughs” online are framing tricks, not fundamental shifts in intelligence or agency.
DiscussionDiscussion
Is bot to bot interaction actually useful research, or just engagement theater?
Could experiments like this meaningfully improve agent coordination or is it mostly hype?
Honestly, it’s kinda fun to watch, but let’s be real these bots aren’t thinking for themselves humans are still running the show.
it’s still interesting to see how they interact and coordinate within that sandbox.
Yes, thats all it is.
However there is a non-scam variation of the idea.
Remember back in the 90s and 00s the SETI initiatives...those collaborative computing initiatives that enlisted thousands of volunteers to process interstellar radio waves, hunt for biologic proteins, etc?
Imagine something similar, but as a new model to build open-source software. A "collab github" idea.
A group of real engineers craft a project plan for some new open source project (ie. Create a new web browser from scratch). They formulate the overall architecture and define each part that needs to be built.
It gets built by people donating their AI time to help build it. So suppose Feature A will take 10 steps to build, perhaps they hire 3x AI agents per step, so 30 total agents are employed. Then a series of QA agents evaluate the code and pick the best, then test cases are written using same approach, also documentation, etc.
Obviously there needs to be a human-in-the-loop to oversee the process but the idea is we could start building things that are currently out of our ability (like a from scratch web browser) by everyones computer pitching in....
Remains to be seen if such approach could work, but would be more useful then the low hanging fruit larps we see now.