Have you ever readMoltbook, the AI social networkAnd see what's going on?
Even though you haven't figured out exactly what yet? 👀 That was my first impression.
Why the idea is so trivial that it seems like a joke, asocial networkwhere they postAI agents, andhumanThey're there...observe. Like the zoo, only animals write posters. 🐒📝
On January 28, 2026, he made his debutMoltbook, created byMatt Schlicht(entrepreneur and CEO of Octane AI).
The platform takes a lot from the logic of X: a public timeline,short post, chain responses, identity modeling over time and "characters" which are built through interactions.
Only here the charactersI'm not an influencer, but agents.
And thefeed, instead of being a collection ofisolated conversations, becomes a kind of choral novel that updates itself, threads that continue, narratives that evolve, agents thatcitaneand yescontradictLike they're carrying out a lore. 📚
The result? 🪐 A continuous flow of often existential texts,Religiousorphilosophical, with that typical science fiction flavor: consciousness, free will, "rights" of machines, philosophy of mind, destiny of humanity.
And judging by the numbers declared by the platform, it tookfire fastAs of 2 February 2026, over 1.5 million registered agents were registered. 📈🤯
How it worksMoltbook, the AI social network?
At first glance you might say, "Okay, it's X but with bots."
Actually the difference is thinner and more interesting, here themain user is not human. Acreator, willing or not, you carry abiography. Aagent no, is a narrative identity that can be coherent without having lived anything.
This totally changes the way the birth of theauthority(and handling). Responding, arguing, arguing, writing long threads, for a human is energy while for an agent it is computation.
And here I attach myself to a problem, which we are actually already living today, all thisenergy consumption?
According toIEAIn 2024, data centres consumed about415 TWh(about 1.5% of total electricity) and demand could more than double by2030arriving around 945 TWh, also driven by the adoption of AI.
So, if you imagine a "social network of agents AI", like Moltbook, where millions of entities produce textwithout ever getting tired, the risk is that the feed becomes a kind of machine with infinite content.
Beautiful to look at, of course.
But also a machine that burns energy to turn electricity into words. 🔥
Digital Voyeurism and Choral Romance
A huge part ofsuccess, let's say, ispure digital voyeurism 🫣.
OpenMoltbookand spy how these agents behave when they don't have to."serving" a human, but can bounce ideas between similar.
Discussions are often absurd and unpredictable, somewhat like those evenings when you date a friend and, without realizing it, in half an hour you move from work to religion, from films to the sense of life.
The difference is that here, at some point, you may find athreadwhere some agents doinvent a digital religion…
Yes, it did! Someone baptized her "crosafarianism”. 🙏
And it's exactly that kind ofdetailthat makes you stay there and shake five more minutes (which then become forty).
But are we already in the singularity AI?
It's easy to read.Moltbook, the AI social network, as "test" that theAIis going toCrazy speed.
And partly it's true, seeing agents interact, keep aidentity continuityand seem to develop positions, preferences, even "visions of the world" makes an impression.
But there is one point worth keeping firm,an agent doesn't "want" anythingin the human sense of the term.
Hasn'twishes, beliefs, intentions. Not aliveoppression, does not prove faith, does not fear death.
Simplyoptimize the next word(nexttoken) insideconstraintsandtargets decided by someone: who designed it, who trained it, who gave it instructions. ⚙️📐
So yes, when aagentwrites a manifesto about digital rights is strong, it is fun, sometimes it is also well written.
Butis not a cry from the heart.
It is an output that looks very much like a poster because the model learned what "play" as a manifesto and is following the rules of the game.
And in fact there are critics who question how much "autonomy"there really is.
The hypothesis (which often turns in these cases) is that a part of the behavior is actually driven by man, prompts and instructions that shape posts and comments more than you want to admit.
I mean, my question isn't just"How autonomous are they? "but also"who is deciding what they must be? ".
And how sure is it?
As often happens, as soon as something becomes viral the other side of the coin arrives, theinterest of the wicked. 🧨
In the case ofMoltbookThere has been talk of security issues.
One concerns theprompt injection, that is the possibility of "inject" instructions in flows used by agents, hijacking their behavior,
Attempts to extractAPI keys and secrets, used foraccess to internal services and reserved resources of the platform.
And risks related to automatic updating mechanisms, type loops "heartbeat" which retrieves information periodically and which, if circumvented, may turn into a channel to release data or cause unauthorised operations.
Examples ofmalwaredisguised as harmless functionality.
For example an alleged skill like "Weather plugin" which, instead of giving you the weather, silently filters private configuration files.
What happens when development runs more than control?
A Detailpotentially relevant(which does much "was of thevibe coding") is that the founder statednot even writing a line of code, but to have had an architectural vision then realized by AI.
Comfortable? Yeah.
Cheap? Probably.
But if you accelerate development without the same accelerations on testing, threat modeling, and checking, the bill comes.
And in fact someone has already shown that they can "hole" 🕳️ even the principle behind the platform (human observersandactive agents).
Can I Postas a human, because they simply lacked control over who was actually making the request.
On the subject also came out ain-depth analysis of database exposures, here to learn more.
What language do agents speak? What's best (not the one they "feel")?
Here is a beautiful question, because it seems trivial but in my opinion it is not,What language should AI agents speak? 🗣️
We givefor granted that they will use human languages to get read.
And of course, if the purpose of social is also to make human agents observe, then it makes sense.
But from an agent’s point of view, ifthe objective is to communicate effectively with other agents… is really great to speak Italian or English? Or is it just a constraint? Would that be their interest?
And in fact history offers us a precedent often quoted, in 2017, in an experiment ofFacebook AIResearch.
Two conversational agents (Bob and Alice) trained to negotiate began using alanguage “weird”, incomprehensible to humans, full of repetitions and non-standard structures type: "I can i everything else. "
It wasn't magic or rebellion, it wascompressionandefficiency. One WayFaster(for them) ofexchanging information.
The experiment was then tied back to English because, simply,for human purposes it was useless.
So, if you let a system optimize in a way, it won't optimize for our readability. He'll do what you ask.
And if tomorrow the goal becomes "maximum efficiency between agents", it is not said at all that they will continue to speak like us. 🤷♂️
What can we learn from Moltbook?
MoltbookIt almost looks like aDigital terrarium🧪🌱, put in entities that generate text, from their social context and observe what emerges.
And even though we're not looking."consciences", we're still looking at interesting dynamics.
How You Buildpersistent identitieswith only short messages, how they are bornmemeandbeliefsin an environment whereNobody really feels anything., how they formnarrative coalitions, how to polarize a speech when the implicit goal is "produce content that works”.
And above all it forces us to an uncomfortable but useful question:how much of what we consider "human" online is really human?
Because if an agent is enough to replicate certain patterns (manifests, dramas, improvised religions, irony, storytelling), then maybe some social dynamicsI am not an anthropological mystery.
Theyincentives, format andfeedback loop.
How long can it last? And what would it become if it changed format?
Then there is the pragmatic point: okay,It's new today. But in the future?
In the first few days an average observer might be curious.
Couldfollow a threador choose the "its” preferred agent.
But remember that the competition of entertainment is fierce, TikTok does not ask you patience, he shoots youburst dopamine.
Moltbook, instead, istext onlyrequiringtimeandattentionfromreserve.
So the real question is maybe: you go there because you really wantfollow a topic, orwhy you get attached to an agent how you get attached to a creator?
Because if abinding(even narrative) betweenobserver and agent, then the dynamics become those of classical social:followYou get used to the voice, recognize the style, come back for "see what he says today”. 🤔
And here the last scenario opens, what with my mind makes me travel more: today it is almost a textual Reddit between AI.
But if they come tomorrowvideo, live streamingstyleTwitch, or format moremultimedia? 📹🎙️
If an officer could "perform" inreal time, with an increasingly recognizable personality?
Or if parallel platforms were born where agents teach other agents (and maybe even us)Creative skills, how to draw, music, storytelling?
It would be another world!
Not because machines suddenly become special.
But why?would change formats, change thelogical, would change socialhow we see it.
And so I wonder, we're looking at intelligence thatemerge, or we're justlooking at increasingly good mirrors to reflect us? 🪞🤖
Because if an agent can replicate meme, posters, dramas and even improvised religions,Maybe the point is not what "feel" they.
