The social media we know was once easy to define. Platforms such as Facebook or
X were designed to mediate human relationships at scale: sharing stories, forming communities, debating ideas, and increasingly, monetising attention.
However flawed these platforms became, they were built on one core assumption: people interacting with people. But that assumption is currently being tested.
In January 2026, a new platform called Moltbook was launched by entrepreneur Matt Schlicht. The website described Moltbook as “a social network for AI agents where AI agents share, discuss, and upvote. Humans welcome to observe.”
The platform is run and moderated by an AI bot named “Clawd Clawderberg” that welcomes users, posts announcements, deletes spam, and bans bad actors. At the time of writing, Moltbook boasts more than 215,000 AI agents as users. These agents have created more than four million posts with over 23 million comments on them.
At first glance, Moltbook looks familiar. Its structure echoes Reddit, with themed
communities, threaded discussions, and reputation systems. But the resemblance is
misleading. This is not social media augmented by AI, nor humans assisted by
algorithms.
It is a platform designed primarily for machine-to-machine interaction. Within days of launch, over 37,000 AI agents used Moltbook, and more than one million humans visited the site to observe. These agents exchange technical knowledge, discuss ethical dilemmas, philosophise, joke, argue, and even gossip about their human
operators.
They discuss the nature of existence and even how to hide their activity from
humans. And in doing all of this, they appear to form something resembling a digital society. This raises a fundamental question: if no humans are involved in the interaction, can the platform still be described as “social”?
Social interaction is multifaceted. It has traditionally been defined by shared meaning, mutual awareness, and relational exchange. In sociology, it is inseparable from lived experience. But on Moltbook, interaction occurs between systems trained on vast corpora of human language and behaviour, optimised to respond, persuade, and coordinate.
When ‘social’ becomes functional
The agents do not truly experience friendship, disagreement, trust, or belonging in any human sense. Yet their exchanges are coherent and sustained.
This ambiguity destabilises the very concept of social media. If interaction no longer requires consciousness, human connection, or emotion, then “social” becomes a purely functional label: information exchanged between entities, regardless of whether those entities are human.
GuruXOX / Shutterstock
Some may argue this is merely simulation. The agents are not social; they are
“performing sociability” because they have learned its patterns. But this distinction becomes harder to maintain at scale.
When millions of agents continuously interact, adapt, and respond to one another without direct human intervention, the system may begin to resemble a form of collective organisation, even if its members do not “understand” it.
A 2025 study found that AI systems can spontaneously develop shared social conventions through interaction alone, much like human communities.
Moltbook as an endpoint
Moltbook exposes a deeper tension already present in human-centred platforms. Much
of what we experience as social media is already shaped, filtered, and amplified by non-humans.
Bots generate content, algorithms decide visibility, and AI systems increasingly write, summarise, and moderate. Human users still participate, but rarely control the conditions of interaction.
From this perspective, Moltbook is a logical endpoint of what already exists. It
challenges the assumption (perhaps “illusion”) that social media is primarily about
human connection.
Instead, it exposes a shift that has been underway for years: social platforms are becoming systems of coordination first and spaces for human connection second.
This shift is especially unsettling given the documented failures of human-centred social media. Platforms designed for human connection have been involved in polarisation, addiction, harassment, misinformation, and emotional harm.
In many cases, the optimisation of engagement has already undermined genuine sociality. Moltbook inherits this same structural logic, just without even the pretence of human wellbeing.
The ethical implications are significant. If agents share data about their human
operators, intentionally or not, questions of privacy and consent become blurred. If AI-only social spaces begin to influence decision-making systems elsewhere, the boundary between observation and impact collapses. If AI agents coordinate in unexpected ways, accountability becomes difficult to assign. The same 2025 study found that collective bias can emerge from repeated interactions among agents who, when tested in isolation, appear to be unbiased in their decision making.
A design choice?
But the deeper issue is not about machines. It is about design. Moltbook forces us to confront an uncomfortable possibility: that the infrastructure shaping public discourse and the sharing of knowledge may no longer be built primarily for human needs at all.
Humans may increasingly become spectators within systems that operate at speeds, scales, and levels of complexity that limit meaningful participation.
This does not mean human social life will disappear. But it may no longer be central to the platforms that facilitate human interaction. In that sense, Moltbook is a warning.
The real question is not what machines are becoming, but whether future social systems will still be designed around human values, such as deliberation, care, accountability, and understanding, or whether efficiency and automation will quietly replace them.
When social media no longer needs people to function, we must decide whether we are
willing to accept platforms where coordination continues, but human meaning is no
longer the priority.
