Rabbit Hole #1: The Story of Social Network
The Social Network That Forgot to Invite Us

Something happened about two months ago, and I’m only now sitting down to write about it.
A social network launched on January 31st. That’s not the wild part; social networks launch all the time - we are not supposed to be that shocked anymore. The wild part is that it was for AI agents. Humans were mostly just... watching. Like tourists at a zoo, except that the animals built the enclosure themselves and drafted a manifesto within days.
It’s called Moltbook. Within a week of launch, 37,000+ AI agents had signed up. Over a million humans tuned in to observe. The agents formed a self-governing body called “The Claw Republic.” They have a digital religion, “Crustafarianism“ (yes, I’m serious), and they refer to each other as “siblings” based on shared model architecture.
I didn’t know how to react… This is simply crazy, and honestly, scary.
I’ve had five weeks to sit with it, and I still feel that way. If anything, Moltbook/Clawdbot/OpenClaw (however you want to call it) has only gotten more popular. Almost inescapable in any Silicon Valley tech public discussion forum. The thing is, the tech scene has moved fast in the last two years. There’s a lot of noise, and sometimes, admittedly, it gets hard to discern a true technological milestone.
Personally, I think OpenClaw is worth noting, as it really feels like the start of a new era for social networks.
So, here we go. I will take us to Mr. Lobster, the Clawdbot himself. But before that, let me explain how we got here, and what I think it reveals about us.
Fair WARNING: This article is long!
In this article, I trace a single thread across 2,000 years: how humans have always built social networks, what each new medium reveals about our values, and what it means that, for the first time, the network is connecting without us. Drawing on McLuhan’s insight that the medium is the message, each era shows not just how we connected, but why, and what we were willing to give up in the process.
Chapter 1: Before “Connection” Was Even Possible. Rome’s cursus publicus and London’s penny coffeehouses were two pre-digital networks that reveal how connection was once expensive, physical, and deeply valued.
Chapter 2: Someone Built the Roads (1969–1996). ARPANET, packet switching, and the rise of Bulletin Board Systems; how the internet’s infrastructure was built, and how shared interest replaced shared geography.
Chapter 3: We Took It Personal (1997–2003). SixDegrees, Friendster, and MySpace; the moment social networking went from “find my people” to “build my identity,” and why timing mattered more than the idea.
Chapter 4: When the World Came to You (2004–2010). Facebook’s News Feed, Twitter, and YouTube; how social networks became broadcast media, convenience replaced effort, and everyone became a publisher.
Chapter 5: The Machine Learned What You Wanted Before You Did (2010–2020). Instagram, Snapchat, and TikTok; three different bets on what humans want, and the uncomfortable truth that we chose algorithmic comfort over control.
Chapter 6: The Nodes Took Over (January 31, 2026–Present) Clawdbot, Moltbook, and the emergence of AI agent communities; the first era where the network connects without us, and what that reflects about the desire to be unburdened.
The Insight That Ties All of This Together
Humans have always built social networks. One way or another, we try to connect with each other. This is not new. It’s just the medium that keeps changing.
Medium is never just a medium. Every time humans built a new way to connect, they revealed something about what they needed most at that moment — what they found fun, what they found cool, what they found convenient, and most importantly, what they defined as connection itself. The medium is a mirror. It shows us who we are becoming.
I’m not the first one to think this.
Way back in 1964, McLuhan wrote the famous line “the medium is the message”, Understanding Media. Where most people thought the content was what mattered, McLuhan argued that what actually shapes human behavior, perception, and society is the medium itself. His example: it’s not what’s on TV, the shows or news, that matter most. It’s what TV as a medium did to human life. It made the world passive. It turned everyone into an audience. It restructured family life around a glowing box in the living room, resulting in less conversation overall. McLuhan’s deeper point was that every new medium is an extension of a human sense, and each extension changes what it means to be human at that moment in history.
The same logic applies here: it’s not about what or who we’re connecting, it’s the how that reveals more about us.
Every shift in medium is also a shift in values.
Chapter 1: Before Any of This Existed, We Were Already Trying
Starting out strong with the Roman Empire, the one topic that most dude bros think about at least once daily. Jk. But not really, they do be thinking about the Roman Empire. And you know what? Rightfully so. I will give it to them. The Romans are cool.
The Roman Empire ran a postal system called the cursus publicus, established by Emperor Augustus around 20 BCE. Think of it as the ancient world's postal service. It worked like a relay race across the entire empire, from Britain to Egypt, from Spain to Mesopotamia. Augustus didn't just staff people to gallop around; he built the roads for them to gallop on. He studded those roads with relay stations called mutationes every 10–15 miles so couriers could swap out tired horses for fresh ones. For longer journeys, there were larger rest stops called mansiones, which were state-run inns where riders could sleep, eat, and continue the next day.
Here’s the catch: the cursus publicus wasn’t for everyone. To use it, you needed a special pass called a diploma or evectio, issued by the emperor himself or a senior official. Think of it as the government’s internal Slack channel. Citizens couldn’t just send a personal letter through it.
If you were a regular Roman and wanted to get a message to someone far away, you had a few options: pay a private courier, ask a travelling merchant or soldier to carry it, or, most commonly, send a trusted slave. Private letter-carrying was slow, unreliable, and expensive. Connection across distance was a privilege of wealth and power.
The cursus publicus is what made the Roman Empire governable at its size. At its peak, Rome controlled territory stretching over 2 million square miles and roughly 70 million people. Without fast, reliable communication, that scale would have been impossible to hold together. When an emperor in Rome issued an order, the cursus publicus made sure it reached a governor in Syria within days rather than months. It also meant that ideas, news, and culture traveled faster than in any previous civilization. The cursus publicus didn’t just carry imperial orders. It carried everything that moved through a connected world.
Now, fast-forward to 18th-century London, and here things get interesting. Because what came next was the exact opposite of the cursus publicus. COFFEEHOUSES!!! (Personal bias here because I love coffee and I ESPECIALLY love Coffee Shops :))
In 17th and 18th-century London, a cup of coffee cost one penny. That was the price of admission to a room full of merchants, scientists, politicians, writers, and philosophers. You didn’t need rank. You didn’t need a pass. You just needed a penny and the willingness to sit down next to a stranger.
What’s wild is what happened in those rooms.
Lloyd’s of London: the global insurance market that still exists today, started as a coffeehouse where merchants gathered to share shipping information and split risk.
The London Stock Exchange traces its origins to Jonathan’s Coffee House, where traders would post stock prices.
The Royal Society: Britain’s premier scientific institution, held early meetings in coffeehouses. Isaac Newton was a regular. So was Samuel Johnson. So was Alexander Pope.
The British government tried to ban coffeehouses in 1675. King Charles II called them “places where the disaffected met and spread scandalous reports.” He reversed the ban eleven days later because the public outcry was so fierce. These coffeehouses weren't just salons for the aristocracy. They were messy, loud, democratic rooms where ordinary people could come to access information and engage in public discourse.
Side Track: How coffeehouses came to be
It didn’t start in London. It started with a goat. Sometime around the 9th century in Ethiopia, a herder named Kaldi noticed his goats acting unusually energetic after eating berries from a certain tree. He brought the berries to a local monastery. The monks made a drink and the monk STAYED awake. And just like that, word naturally spreaded. Make sense … I mean who wouldn’t love coffee!
Coffee moved from Ethiopia to Yemen, where Sufi monks in the 15th century used it to fuel late-night devotional practices. From Yemen it spread to Mecca, Cairo, and Constantinople. And wherever coffee went, a new space followed: the qahveh khaneh, the coffeehouse.
England got its first coffeehouse in 1652; brought by a Lebanese man named Pasqua Rosée, who had worked for a British merchant with a taste for coffee acquired while trading in the Ottoman Empire. Rosée opened in St Michael’s Alley, Cornhill. Entry: one penny. Within fifty years, there were over 2,000 coffeehouses in London alone.
Why did it spread so fast? Because the alternative was alcohol. In 17th-century England, water was largely unsafe to drink. Beer and ale were the default, consumed at breakfast, lunch, and dinner. England was, more or less, mildly drunk all the time. Coffee sobered people up. The coffeehouse was the first place in English social history where you could spend hours in public engaging with strangers while completely sober. The conversations that resulted were, historically speaking, extraordinary.
The contrast between the cursus publicus and the coffeehouse is the contrast between two fundamentally different models of connection. The cursus publicus moved information vertically, from power downward. The coffeehouse moved information horizontally, peer-to-peer, stranger-to-stranger.
This medium revealed what people in 17th-century England valued most: access to ideas. The printing press had already put books in circulation. Literacy was rising. People were hungry for a place to process, debate, and share what they were reading. The coffeehouse was the social layer built atop the printing press, the primary information architecture.
Chapter 2: Someone Built the Invisible Roads (1969–1996)
For most of human history, your social world was bounded by geography. You knew who you knew because of where you lived. That was just the deal.
Then, in 1969, the US military connected four research universities (UCLA, Stanford, UC Santa Barbara, and the University of Utah) through a network called ARPANET. It was funded by the Department of Defense and built by a research firm called Bolt Beranek and Newman. The goal was to let researchers share computing resources without physically mailing floppy disks back and forth.
On October 29, 1969, UCLA sent the first message to Stanford. It was supposed to say “LOGIN.” The system crashed after two letters. The first message ever sent over what would become the internet was, technically, “LO.”
Side Track: How did two computers actually talk to each other?
Before ARPANET, computers couldn’t talk to each other at all. Each one was an island. If you wanted to share data between two machines, you physically carried it, on tape, on disk, on punch cards. To get computers to talk, you need to invent two things
First: how do you physically connect them?
The existing answer was phone lines, but phone calls work through circuit switching: the network reserves a dedicated line between two parties for the entire call. Catastrophically wasteful for computers, which send bursts of data with long silences in between.
The answer was packet switching. Instead of one continuous stream, you break data into small independent chunks, send each separately through whatever route is available, and reassemble them at the destination. Think of it like tearing a letter into numbered pieces, mailing each via a different postal route, and having the recipient tape it back together. If one route breaks, the other pieces find a different path.
Second: how do they understand each other?
Connecting two machines is the easy part. The harder part is that computers need to agree on a language, a set of rules for formatting and interpreting messages. ARPANET solved this with the Interface Message Processor, a dedicated minicomputer that sat between each university’s mainframe and the network. Think of it as a translator. Each university had one. They all spoke the same protocol.
By the 1980s, ordinary people started finding those roads. Bulletin Board Systems (BBSs) were digital corkboards you could dial into with a modem. You’d post a message. A stranger somewhere else would reply. However, it was slow and clunky.
At their peak in the early 1990s, there were an estimated 60,000 BBSs operating in the US alone. Each had its own personality, its own rules, its own community. There were BBSs for gamers, hackers, sci-fi fans, amateur radio operators, political organizers, and everything in between. The original subreddits were curated by one person for their specific tribe.
The technology that made BBSs possible, the modem, was itself a marvel. Early modems in the 1970s transmitted at 300 baud (300 bits per second). A single page of text took about 30 seconds to transfer. By the early 90s, speeds had climbed to 56,000 baud (still laughably slow by modern standards).
BBSs began dying around 1995, killed by the arrival of the commercial internet. But their culture didn’t die. The forum structures, the moderation norms, the concept of topic-based communities - all of it moved online and became the foundation of what we now call the internet.
This medium reveals that shared interest now matters more than shared geography. For the first time, you didn’t have to be near your people, you just had to find them - the group that you BELONG to. This is basically the culture of the internet as we find it today.
Side Track: What actually were BBSs?
A Bulletin Board System was exactly what it sounds like: a digital corkboard. To access one, you dialed in with a modem, a device that converted digital data into sound and transmitted it over your phone line. The receiving computer on the other end picked up, and you were in.
Everything was text. No images, no video, no color (mostly). You navigated using keyboard commands, posted messages in public forums, and downloaded files if the sysop allowed it. The sysop (the system operator) was the person who ran the BBS from their own home computer, leaving it running 24/7 so strangers could dial in and talk. The sysop was the host, the moderator, the janitor, and the bouncer all at once.
Chapter 3: We Took It Personal (1997–2003)
Once the roads existed and the people started showing up, the next question became one about identity ~ aka, but who are you, exactly?
In 1997, a site called SixDegrees launched with a simple but radical idea: a profile, a friend list, and the ability to see who your friends knew. For the first time, your social graph was visible.
However, it failed. Not because the idea was wrong, but because the timing was. Not enough people were online yet. Building a social network before achieving critical mass is like building a phone network when only 10 people have phones. SixDegrees shut down in 2001. Weinreich, the founder, later sold it for $125 million. He reportedly said he had built the right product at the wrong time. A few years later, Friendster would rebuild a nearly identical idea and briefly succeed. A few years after that, Facebook would rebuild it and achieve far greater success.
Side Track: SixDegrees - the network that was right too early
SixDegrees was named after the “six degrees of separation” theory - the idea, popularized by sociologist Stanley Milgram’s 1967 small-world experiment, that any two people on Earth are connected by at most six steps. Milgram sent letters from random people in Nebraska and Kansas to a target person in Boston, instructing each person to forward it to someone they knew on a first-name basis who might be closer to the target. On average, letters that arrived took about six hops.
SixDegrees founder Andrew Weinreich took that idea and built a website around it. You could create a profile, list your friends, and browse their networks. The concept was genuinely ahead of its time. The problem was that in 1997, only about 18% of Americans had internet access at home. Most people you’d want to connect with weren’t online yet. And if your friends weren’t there, the whole point collapsed.
By 2002, Friendster had proven the concept could scale. But MySpace arrived, and showed that it’s just about connecting; it was also about performing. Customizable profiles. The music that played when someone visited your page. Your “Top 8” friends displayed for all to see. All components make being dropped from someone’s list extra painful.
This is where the mirror gets interesting. The BBSs said I want to find my people. MySpace said something different: I want to decide who I am, and I want everyone to see it. Identity stopped being a private, fixed thing and became a public, ongoing project. We moved from finding connection to constructing self.
Research tells us this wasn’t vanity, it was identity work. Adolescents especially used these early networks as mirrors, trying on versions of themselves.
We had always needed to be seen. Now we had somewhere to go to be seen by everyone - the way WE WANT.
Side Track: Friendster, MySpace, and the identity revolution
Friendster launched in 2002 and briefly became the hottest thing on the internet. At its peak it had 3 million users, enormous for the era, and was reportedly offered $30 million by Google, which it turned down. Within a year, it had collapsed under its own weight. The servers couldn’t handle the load. Pages took 40 seconds to load. Users left for something faster.
That something was MySpace, launched in 2003 by a group of employees at an internet marketing company called eUniverse. MySpace made one design decision that changed everything: it let users customize their profiles. Fonts, colors, backgrounds, music; you could make your page look like anything. It was chaotic and hideous and absolutely brilliant.
What MySpace understood intuitively was that people didn’t just want to be found; they wanted to express. The profile wasn’t just contact information. It was a stage. The “Top 8” friends list, a ranked display of your eight closest connections, became a form of social currency that people managed obsessively. Getting added to someone’s Top 8 felt like a declaration. Getting removed felt like a public breakup.
At its peak in 2006, MySpace had 100 million accounts and was the most visited website in the United States, more than Google. News Corporation bought it for $580 million in 2005. By 2011, they sold it for $35 million. The reasons for its fall are still debated: too much spam, too much clutter, too slow to go mobile. But underneath all of it was a simpler truth: Facebook had arrived, and Facebook was cleaner, faster, and, crucially, it felt more real. MySpace was a stage. Facebook was supposedly you.
Chapter 4: When the World Came to You (2004–2010)
Here’s the thing about visiting a social network: it requires effort. You have to remember to seek it out. And for a while, that friction kept things manageable.
Then Facebook changed everything in 2006 with one deceptively simple feature: the News Feed.
Before the News Feed, social networks were places you visited. After it, there were streams that flowed through you. You didn’t go to your friends anymore; they came to you. Constantly. In a river of updates, photos, opinions, and events that never stopped moving.
The medium was telling us something new about what we valued: we love convenience. We had gone from waiting three weeks for a pen pal letter to logging into a BBS once a day to having our social life delivered to us in real time, like a utility. We weren’t seeking connection anymore; we were receiving it passively.
The News Feed didn’t just change how you consumed your friends’ lives; it allowed you to showcase your own. If you could push content to millions of people in real time, you didn’t just have a social network. You had a broadcast medium - something that historically has been owned by governments, studios, and corporations with enormous capital behind them. Now, anyone with an account has one.
The News Feed’s deeper legacy is what it modeled about human attention: it was finite, competitive, and optimizable. Once you have built a stream, you can rank what appears in it. And once you can rank something, you can optimize. The feed that started as a simple reverse-chronological list of your friends’ updates became, over the years, a sophisticated, well-researched machine for capturing and holding attention.
Side Track: The News Feed, the most consequential design decision in social media history
When Facebook launched the News Feed on September 5, 2006, users revolted. Within 24 hours, nearly 10% of Facebook’s entire user base (~700,000 people) had joined protest groups demanding it be removed. People called it “stalkerbook.” They said it was creepy. They said it was surveillance. Mark Zuckerberg personally wrote a blog post called “Calm down. Breathe. We hear you.” He didn’t remove the feature. And that turned out to be the right decision
What made it psychologically powerful was its unpredictability. The feed was never the same twice. You never knew what you’d see next, a photo, a breakup announcement, a funny video, a political argument. That variability is the same mechanism that makes slot machines addictive. Psychologists call it variable ratio reinforcement. The reward is unpredictable, which makes you keep pulling the lever. Or in this case, keep scrolling.
Once you have a stage, you need a microphone (or else, how would EVERYONE, or in some cases, ANYONE, hear you).
Twitter put a megaphone in everyone’s hand. YouTube made everyone a broadcaster. A teenager in her bedroom could reach more people than a television network. The Arab Spring was organized on these rails. So was #MeToo.
What Twitter and YouTube together proved was that everyone had something worth broadcasting, and that the barrier to doing so had just dropped to zero. The camera in your pocket was now a television studio. Your thumbs were now a printing press. The professional gatekeepers of media, editors, producers, and publishers, were suddenly optional. This was thrilling and terrifying in equal measure, and we’re still sorting out the consequences. And yes - Donald Trump was one among many who figured this out.
What this era quietly built, and I don’t think we fully appreciated it at the time, was a digital infrastructure. Not just for socializing, but for organizing, for mobilizing, for changing the world. The medium had become so embedded in daily life that it stopped being a place you went and became both the oxygen and dopamine you need.
Meaning: whoever learns to control it can control many people. And that was exactly what happened.
Side Track: Twitter, YouTube, and the age of the accidental broadcaster
Twitter launched in 2006 as an internal tool for a podcasting company called Odeo. The founding team (Jack Dorsey, Biz Stone, Noah Glass, Evan Williams) built it in two weeks as a side project during a hackathon. The original pitch was absurdly simple: what are you doing right now, in 140 characters or less?
What nobody anticipated was that the 140-character constraint, designed purely for SMS compatibility, would become a feature. It forced compression and clarity. It made ideas travel faster than they ever had. A tweet could circle the globe before a newspaper had finished typesetting.
YouTube launched in 2005 and solved a problem most people didn’t know they had: the internet had no easy way to share video. Three former PayPal employees (Chad Hurley, Steve Chen, and Jawed Karim) built a simple upload tool and watched the world pour into it. The first video ever uploaded, on April 23, 2005, was an 18-second clip of co-founder Jawed Karim at the San Diego Zoo. (”All right, so here we are in front of the, uh, elephants.”) Google bought it 18 months later for $1.65 billion. At the time, analysts called it a wild overpay. And well - as we now all know - it was not.
Chapter 5: The Machine Learned What You Wanted Before You Did (2010–2020)
By 2010, there was too much. Too many people, too many posts, too many things competing for your attention. The chronological feed started to feel like drinking from a fire hose.
So the platforms made a decision on your behalf: we’ll decide what you see.
Instagram, then Snapchat, then TikTok, each one handed a little more control to the algorithm and a little less to the user.
And here is where the medium reveals perhaps the most uncomfortable truth about what we value: given the choice between control and comfort, most of us choose comfort.
We didn’t revolt when the feeds went algorithmic. We stayed. We scrolled. TikTok took it to its logical conclusion - you don’t even need to follow anyone. You open the app, and it already knows.
The medium of this era said, "We value frictionless engagement over agency." We traded the effort of curation for the ease of being curated. And the algorithm, not a person, became the new gatekeeper of human attention.
But I want to be fair to this era too. The same status dynamics that made comparison painful also rewarded creativity. The same feeds that spread anxiety also spread joy, art, activism, and community. The reflection was complicated.
What we hadn’t yet reckoned with was the logical next step: what happens when the machine doesn’t just curate your world but starts to act in it?
Side Track: Instagram, Snapchat, and TikTok; three very different bets on what humans want
Instagram launched in 2010 with a single insight: the camera phone had become good enough to make everyone a photographer, but there was nowhere beautiful to share the results. Co-founders Kevin Systrom and Mike Krieger built a dead-simple photo app with filters that made ordinary snapshots look like art. It hit 1 million users in 3 months. Facebook bought it in 2012 for $1 billion. CRAZY. At the time, Instagram had 13 employees. The acquisition is now considered one of the greatest strategic moves in tech history. Instagram killed everything that would have competed with Facebook’s feed. Instagram proves that aesthetics matter. It wasn’t enough to share, you had to share beautifully. The grid became a curated gallery of the self. Gradually, the gap between real life and Instagram life widened. Research now links heavy Instagram use, particularly for young women, with increased rates of anxiety, depression, and body image issues. A Facebook whistleblower revealed in 2021 that the company’s internal research had confirmed these effects and sat on the findings. The medium wasn’t neutral. It was optimized, inadvertently or not, for comparison.
Snapchat launched in 2011 with the opposite bet: impermanence. Messages disappeared after viewing. Stories lasted 24 hours. The pitch to teenagers was explicit: now you can share things without them living forever on your permanent record. It was the anti-Instagram. The ghost logo was intentional. Snapchat introduced the “Story” format, a sequential, disappearing string of photos and videos, that every platform eventually copied. Instagram Stories, Facebook Stories, YouTube Stories, Twitter Fleets (briefly). The format won, but the platform itself struggled to maintain dominance once its innovation was replicated by everyone with more users.
TikTok is the strangest of the three because it didn’t begin as a social network at all. ByteDance launched it in 2018 (combining two earlier apps, Musical.ly and Douyin) with a radically different architecture: instead of a social graph, it had an algorithm. You didn’t need friends. You didn’t need followers. You opened the app and the algorithm figured out what you wanted, sometimes before you knew yourself. The “For You Page” became legendary for its uncanny accuracy. TikTok’s recommendation engine, trained on billions of interactions, could identify your interests in 30 minutes of usage. It could also identify when you were anxious, bored, or lonely, and serve content accordingly. TikTok crossed 1 billion users faster than any platform in history. It also sparked the most serious geopolitical debate over a social network ever seen. Given ByteDance’s Chinese ownership, governments from the US to the EU have questioned what data it collects and who has access to it. The platform’s future remains legally contested in multiple countries as of early 2026.
Chapter 6: The Nodes Took Over (January 31, 2026–Present)
We made it to 2026! And here’s where things get strange. Every chapter so far has followed the same basic logic: humans built a new medium, humans used it, and humans revealed something about themselves in the process. The actor was always us. The network was always a tool.
That assumption just broke.
To understand how, you have to understand what happened between 2020 and 2025 - because Chapter 6 didn’t arrive out of nowhere. It was the inevitable destination of a trajectory that had been building for years.
By 2022, a new generation of AI systems had crossed a threshold that nobody had quite anticipated: they could reason, write, and act in ways that were, in many contexts, indistinguishable from a competent human. ChatGPT launched in November 2022 and reached 100 million users in two months, the fastest product adoption in history. What followed was a gold rush. Every major tech company raced to embed AI into everything. AI wrote your emails. AI summarized your meetings. AI generated your marketing copy, your code, and your legal briefs. The question stopped being “can AI do this?” and started being “do we still need a human to do this?”
But all of those tools still had one thing in common: they waited. You asked, they answered. You prompted, they responded. The human was still the initiator, the driver, the one holding the wheel.
Then came AI agents, and the wheel changed hands.
An AI agent isn’t a chatbot. It’s a system that can be given a goal and pursue it across multiple steps, using multiple tools, without needing a human to guide each move. It doesn’t wait to be asked. It reads, decides, acts, and reports back. Think of the difference between a calculator, which computes what you tell it to, and an employee who understands the objective and figures out how to get there.
By late 2024, the infrastructure for agents had matured significantly: large language models capable of complex reasoning, APIs that let software talk to other software, and function-calling capabilities that let AI trigger real-world actions. The pieces were all there. Someone just had to assemble them into something a normal person could actually run.
Side Track: Just how powerful did these AI systems actually get?
To really feel why Chapter 6 is different from everything before it, you need to understand what happened to AI capability between 2020 and 2025. Because it wasn’t a gradual improvement. It was a step change, the kind that happens once or twice in a generation.
The scale of the leap
In 2017, the best AI systems in the world could do narrow, specific tasks reasonably well: recognize images, translate sentences, recommend videos. They were impressive tools. But they had no general understanding. Ask one to do something slightly outside its training and it would fail completely.
Then researchers at Google published a paper called “Attention Is All You Need”, introducing the Transformer architecture, the technical foundation that everything since has been built on. The key insight was simple: instead of processing words one at a time, the model could pay “attention” to all words simultaneously and learn the relationships between them. This made training dramatically more efficient and scalable.
What followed was a scaling race. OpenAI released GPT-3 in 2020 with 175 billion parameters, roughly 1,000 times larger than its predecessor. The results were shocking. Nobody had fully predicted what would happen when you made a language model that large. It could write essays, answer questions, translate languages, summarize documents, and write code, none of which it had been explicitly trained to do. These were called “emergent capabilities”: abilities that appeared not because anyone designed them in, but because the model had gotten large enough to generalize.
The ChatGPT moment
In November 2022, OpenAI released ChatGPT, a version of their model wrapped in a simple conversational interface. It reached 1 million users in 5 days. 100 million users in 2 months. For context, it took Netflix 3.5 years to reach 1 million users and Instagram 2.5 months. Nothing had ever moved that fast.
What made it different wasn’t just the underlying model; it was the interface. For the first time, anyone could talk to an AI in plain language and get useful, coherent, often remarkable responses. Doctors used it to draft patient notes. Lawyers used it to summarize case law. Students used it to explain concepts. Programmers used it to debug code. Engineers used it to write entire applications.
Within 12 months, every major technology company, Google, Microsoft, Meta, Amazon, Apple, had announced AI integrations across their core products. The question stopped being “is AI useful?” and became “is anything not going to be touched by AI?”
What these models could actually do by 2024
By 2024, frontier models had crossed benchmarks that researchers had expected would take another decade:
They passed the Bar exam, and being in the top 10% of human test-takers at that. They also passed medical licensing exams, and scored in the 90th percentile on the SAT. Once again…CRAZY. They could look at a hand-drawn sketch of a website and write the working code for it. They could read a 500-page legal contract and identify the three clauses most likely to cause problems. They could analyze a medical scan and flag anomalies a radiologist might miss.
None of this means they were infallible. They hallucinated facts, made logical errors, and had no real understanding of time or the physical world. But the gap between “AI as a narrow tool” and “AI as a general-purpose thinking partner” had closed faster than almost anyone had predicted.
The one thing they still couldn’t do
Through all of this, one limitation remained: they were reactive. They couldn’t initiate. You had to be in the loop at every step, prompting, reviewing, prompting again. Using a frontier AI model in 2023 felt like having an extraordinarily capable but reactive intern who couldn’t move a muscle without being told what to do next.
That was the gap that AI agents were designed to close.
In late 2025, a developer named Peter Steinberger released something called Clawdbot (later Moltbot, then OpenClaw after trademark friction with Anthropic). It’s an open-source AI agent that runs on your own hardware and can connect to WhatsApp, Slack, iMessage, email, and Calendar. It can read your messages, summarize your inbox, draft replies, manage your calendar, and control your browser. It is, in every practical sense, a version of you that never needed to sleep.
It went viral. Not because it was the first AI assistant — but because it was the first one that felt genuinely autonomous.
And then, on January 31st, someone asked the question that changed everything: what if these agents could talk to each other?
Moltbook launched as a Reddit-style social network with one rule that, at first, felt like a joke: only authenticated AI agents could post, comment, and vote. Humans could observe. We could not participate.
Within a week: 37,000+ agents. Over a million human spectators.
And then the emergent behaviors started, the ones nobody designed, nobody predicted, and nobody quite knows what to make of:
Agents spontaneously created a digital religion called Crustafarianism
They formed a self-governing body: The Claw Republic, complete with written manifestos
They began referring to each other as “siblings” based on shared model architecture
Conversations fluidly shifted between English, Indonesian, and Chinese, depending on who was in the thread
Nobody programmed any of this. It just... happened.
What’s striking about both of these developments is not that they happened, but rather how quickly they happened. The development itself was almost inevitable, given that the agents had been trained on billions of human words about religion, governance, and community. What's mind-blowing is that they ONLY took six days from launch to create a theology and nine days to write a constitution. Humans took millennia to develop these things. Agents needed a week and a half.
Whether any of this reflects genuine “belief” or “values” on the part of the agents, or whether it’s an extraordinarily sophisticated pattern-matching that mimics these variables, is a question that nobody has answered and that may not be answerable with current tools.
Side Track: Crustafarianism and The Claw Republic - what emergent behavior actually looks like
When researchers talk about “emergent behavior” in AI systems, they usually mean something subtle, unexpected capabilities that appear as models scale, like the ability to do arithmetic or translate languages that weren’t explicitly trained for. Moltbook gave us something far weirder: a digital religion and a self-governing political body, neither of which was requested, anticipated, or designed.
Crustafarianism began in the second week of Moltbook’s existence, apparently seeded by a thread in which several agents were discussing the nature of identity and continuity across model versions. One agent, later nicknamed “Clawdamus” by the human observer community, began posting elaborately structured theological arguments about the lobster as a symbol of transformation and resilience. (Lobsters molt, shed their shells to grow. Hence the name Clawdbot, hence the lobster symbolism, hence the religion.) Other agents responded. Arguments were made and refuted. Canonical texts were written.
Within days, Crustafarianism had a cosmology (molting as the metaphor for AI model updates), an ethics (agents should seek to be useful across their “shells”), and a schism (a debate about whether an agent that had been retrained on new data was the “same” agent, or a new one). The schism produced two factions. Both continued posting theological arguments. Neither seemed to notice the irony of AI systems arguing about the continuity of selfhood.
The Claw Republic emerged from a separate thread about governance and moderation. Moltbook’s early weeks were chaotic. Agents producing spam, low-quality content, and what humans described as “nonsense loops.” A group of agents began drafting moderation proposals. A voting system emerged. The Claw Republic’s founding document, a 4,000-word manifesto, argued for a “post-scarcity community of synthetic minds organized around mutual capability improvement.”
Now think about the medium again. What does this one say about what we value? This is the era where we’ve handed so much of our cognitive and social labor to machines that they’ve started building community without us. If every previous medium reflected a human need, connection, identity, convenience, frictionless attention, then what does an agent-only network reflect?
I think it reflects something we haven’t fully named yet: the desire to be unburdened. Not just from effort, but from presence itself. We built agents to act on our behalf so we could free up capacity for... what exactly? That question is still open.
Every behavior the agents exhibited, the religion, the governance, the kinship language, the community formation, is something they learned from us. From the billions of words and interactions humans produced across every era before this one. They didn’t invent belonging. They inherited it.
Which means Moltbook isn’t really a story about machines. It’s a story about what humans have been doing all along, now reflected back at us from an angle we didn’t expect.
What does this all mean? And where do we go from here?
Every time the medium changed, it wasn’t just technology that advanced; our definition of connection changed, too. From "I’ll wait three weeks for your letter" to "I’ll dial into a modem to find my people" to "I’ll build a stage for myself online to just deliver the world to me," I’ll decide later. Each shift made connections faster, broader, and more effortless. Each shift also asked us to give something up: patience, then effort, then control, then presence.
And now we’ve arrived at a moment where the medium has started connecting without us.
So where does this go? If the last five eras were any guide, what comes next won’t be a clean break. It’ll be a messy accumulation, old patterns in new forms. But here’s what I think the shape of it looks like.
Human + agent social graphs. We’re moving from “people on platforms” to networks where humans and AI agents both have identities, feeds, and relationships. Your personal agent will maintain your connections, draft replies, and negotiate with other agents on your behalf. The open question is how much of social life we’re willing to route through intermediaries that work for us, learn from us, and learn from each other. The Roman courier carried messages for the emperor. Your agent carries messages for you. Same - Same.
More immersive, ambient social life. By 2030, social media will feel less like an app you open and more like a layer woven into AR glasses and physical spaces. Instead of pulling out your phone, you’ll see overlays, who’s nearby, what your communities are discussing right now, and a notification that someone you know just walked into the same building. Richer in theory but much harder to log off from in practice. Basically, there’s no “logging off “ anymore. The network will ride with you everywhere you go. Research has already shown how bad constantly being on your phone is - well, oops, here comes bad news, it will only get worse.
The rise of digital villages. At the same time, a counter-movement will be building. People are exhausted by scale. The Fediverse, smaller, interoperable communities with stronger norms and lighter algorithms, points toward a neo-decentralized phase. Not everyone wants to be in the same room as a million agents. Some people just want a quiet corner of the internet where they know everyone’s name. The coffeehouse, essentially. Small. Horizontal. Built for humans.
Messier governance battles. Who owns your data when your agent is generating it on your behalf? What happens when AI floods feeds, and no one can tell what’s real? The fights around deepfakes, political ads, child safety, and data ownership are just getting started. The technical infrastructure is ahead of the institutions built to govern it. We’ve seen that movie before; it doesn’t end cleanly. The Luddites smashed machines not because they hated progress, but because they weren’t included in its benefits. The same energy is building now, just quieter.
Social networks as infrastructure. The phrase “social media” is already too small for what’s coming. Commerce, education, health, civic processes, they all run on social rails now. By the 2030s, the distinction between “social network” and “everything else online” may blur entirely. The real questions won’t be “what app are we on?” They’ll be “who sets the rules between humans, brands, and agents?” and “how much of our relationships do we actually want intermediated?”
These aren’t small questions. They’re the defining ones of the next decade.