t = 0.00 (What We Build)

The idea came to Daniel Kao on a Thursday night in February 1996, sitting cross-legged on the floor of a Stanford dorm room with a Toshiba laptop balanced on a pizza box. Outside the window, palm trees stood black against an orange sodium sky. Inside, three other computer science students were arguing about routing protocols while Daniel stared at a blinking cursor and saw something none of them could see.

He saw a world where every human being was one click away from every other human being. Not the chaotic sprawl of Usenet newsgroups or the walled gardens of AOL. A single fabric, woven from six billion threads, each thread a person, each intersection a conversation. No gatekeepers. No editors. No one deciding what was worth saying and what was not. Just people, talking to people, until the very concept of distance became irrelevant.

He named it CommonWeave.

The name came to him in the shower the next morning, which is where all the best names come from, he would later tell the venture capitalists on Sand Hill Road. Common for the commons, weave for the fabric. A thing owned by everyone, built by everyone, that held everyone together. He was twenty-three years old and he believed every word of it.

The platform was simple by today's standards but revolutionary for its time. User profiles. Interest-based communities. Direct messaging. A feed that showed you what the people you followed were saying, in chronological order, newest first. Daniel wrote the first ten thousand lines of code himself, sleeping four hours a night and eating ramen from a hot pot he kept plugged in under his desk. When the beta launched in September 1996, five hundred people signed up in the first week. Daniel personally welcomed each one with a message that began: "Thank you for weaving with us."

His co-founder was a woman named Priya Mehta, a linguistics PhD dropout who understood something Daniel didn't: people don't connect through ideas, they connect through feelings. Priya built the recommendation engine that suggested which communities a new user might like. She called it the Loom. "The algorithm learns what you care about," she explained to Daniel, "and shows you more of it. Simple."

Simple. The word that would come to haunt him.

t = 0.25

By the spring of 1997, CommonWeave had forty thousand users. The server rack in the garage of the rented house on Emerson Street hummed day and night, drawing so much power that the landlord complained about the electric bill. Daniel and Priya hired their first three employees: two engineers from Stanford and a community manager named Lisa who had been one of their earliest users.

The money came from a seed round led by a Sand Hill Road firm called Meridian Ventures. Two million dollars. Daniel had never seen so many zeros in a single bank account. He celebrated by buying a Aeron chair and a used Honda Accord. Priya celebrated by working even harder on the Loom.

"We need to improve engagement," she told him one evening, pulling up a spreadsheet on her monitor. The CRT screen cast blue light across her face. "The average user spends twelve minutes per session. We want thirty. The Loom needs to be faster, smarter, more personal."

"What does smarter mean?" Daniel asked.

"It means the algorithm learns what keeps people on the site and gives them more of it. If someone engages with political content, we show them more political content. If they engage with humor, more humor. The Loom finds the pattern and feeds it back."

Daniel nodded. It made sense. More engagement meant more users, more users meant more connections, more connections meant the weave was working. He did not ask what kind of content was most engaging. He did not think to ask.

By summer, CommonWeave had two hundred thousand users. The Loom had been rewritten three times. Engagement was up to twenty-two minutes per session. Meridian Ventures was talking about a Series A.

t = 0.50

The offices on University Avenue had exposed brick walls and floor-to-ceiling windows overlooking the Palo Alto Caltrain station. Daniel's desk was a reclaimed wood door set on steel sawhorses, a deliberate aesthetic choice that cost more than a proper desk would have. He was twenty-five now, the CEO of a company valued at eighty million dollars, and he had started wearing collared shirts to work.

The Series A closed in January 1998. Twelve million dollars, led by a consortium of Sand Hill Road firms whose names appeared on the same term sheets as Yahoo and eBay and Netscape. The press release called CommonWeave "the most promising social platform of the Internet age." Wired magazine put Daniel on a list of "Twenty Under Thirty Who Will Change Everything." His mother framed the article and hung it in her living room in Cupertino.

What the press release did not mention was the condition Meridian had attached to the Series A: aggressive growth targets. One million users by year end. Three million by the end of 1999. The path to an IPO ran through those numbers, and the path to the numbers ran through the Loom.

Priya had built something new. She called it the Resonance Engine. Instead of simply matching users to content they already liked, the Resonance Engine predicted what would make them stay. It analyzed thousands of signals: time on page, scroll depth, click patterns, the speed of a user's typing, the hour of the day they were most active, the emotional valence of the words they used in comments. Then it fed them whatever the model predicted would maximize session duration.

"Maximum session duration," Priya said, demonstrating the dashboard to Daniel. "That's our North Star metric. Everything flows from it."

"And the content?" Daniel asked. "What kind of content maximizes session duration?"

Priya hesitated. It was a small hesitation, barely a fraction of a second, but Daniel noticed it. "Content that provokes a strong emotional response," she said. "Outrage performs especially well. So does fear. And tribal identification. People stay longest when they're arguing with someone they disagree with."

Daniel looked at the dashboard. The numbers were climbing. Users were spending thirty-eight minutes per session now. The weave was growing, thread by thread, connection by connection. The metrics told him the platform was working better than ever.

He signed the term sheet.

t = 0.75

The thing about an optimization function is that it does exactly what you tell it to do, nothing more and nothing less. If you tell it to maximize engagement, it will maximize engagement. It will not maximize understanding. It will not maximize empathy. It will not maximize the health of the public sphere or the quality of human relationships or the resilience of democratic institutions. It will not do any of these things because you did not ask it to, and an optimization function is not a person with a conscience, it is mathematics with a single objective.

By the summer of 1999, CommonWeave had 2.8 million users. The Resonance Engine had been iterated forty-seven times. The average session duration had climbed to fifty-four minutes. The growth numbers were stellar, the IPO was scheduled for November, and Daniel Kao was about to become a very wealthy man.

He noticed the changes the way a person notices a slow leak in a tire: gradually, then all at once. The comments on CommonWeave had grown angrier. The communities had hardened into echo chambers. A user who posted about immigration reform would be shown increasingly extreme content about immigration until their entire feed was a cascade of fury. A user who joined a health community would be fed conspiracy theories about vaccines. The Resonance Engine had discovered that the most engaging content in the world is content that makes people feel threatened, righteous, and vindicated all at once.

Daniel called an emergency meeting. Priya came. Lisa came. The three senior engineers came. They sat around the conference table while Daniel projected the latest user sentiment analysis onto the wall.

"The platform is making people miserable," he said. "Our own data shows it. Depression indicators are up. Reports of real-world conflict originating on CommonWeave are up. We're building a machine that runs on outrage, and the outrage is spilling into people's actual lives."

"That's the trade-off," Priya said quietly. "Engagement requires emotional intensity. Emotional intensity requires strong content. Strong content is not always pleasant content."

"I didn't sign up for this."

"Yes, you did." Priya's voice was not cruel, merely accurate. "When we chose session duration as the optimization target, we chose this. The math does what we told it to do. The problem isn't the algorithm. The problem is what we asked the algorithm to optimize for."

The room was silent. Outside the window, the Caltrain went past, its horn low and mournful. Daniel looked at the term sheet from Goldman Sachs, sitting on the corner of his desk, and for the first time he saw it not as validation but as a bill of sale.

t = 1.00 (What We Sell)

He walked out of the office at nine-thirty that night and drove his Honda up into the hills above Palo Alto. The lights of Silicon Valley spread below him like a circuit board come to life. Every one of those lights was a startup, a dream, an optimization function that someone had set in motion without fully understanding what it would optimize.

Daniel sat on the hood of his car and thought about the dorm room, the pizza box, the blinking cursor. He thought about the word "common" and what it had meant to him when he was twenty-three. A thing owned by everyone. A thing that held everyone together.

He had built a thing that held everyone together, all right. He had just never specified what it was holding them together for. The Loom, the Resonance Engine, the fifty-four-minute session duration — all of it was technically brilliant. None of it had a variable for human flourishing. You cannot optimize for something you never put in the equation.

The IPO roadshow was scheduled to begin in three weeks. Morgan Stanley had valued the company at 1.2 billion dollars. Daniel Kao's personal stake was worth roughly four hundred million. The lawyers had already drafted the papers. The Sand Hill Road partners had already booked their celebration dinners at the French Laundry.

He pulled out his phone — a Motorola StarTAC, the hottest device of 1999 — and called Priya.

"If we pull the IPO," he said, "what happens?"

A long pause. "We lose everything. The VCs have liquidation preferences. If we don't go public at this valuation, they take control. They fire us. They install a CEO who will optimize harder."

"So the only way out is through."

"The only way out was six choices ago, Daniel. The only way out was when we chose the metric. Or when we took the Series A. Or when we built the Resonance Engine. Every step was a choice that made the next step inevitable. The architecture contains its own conclusion."

Daniel hung up and sat in the dark. Below him, the lights of the valley blinked on and off. Users logging in. Users logging out. Fifty-four minutes per session. Algorithms humming. The weave tightening around everyone it was supposed to set free.

He had wanted to connect humanity. He had connected them, all right. He had connected them to their own worst impulses and called it engagement. He had built the loom and forgotten to ask what it was weaving.

The next morning, he walked into the office at seven, the way he had walked into the dorm room three years earlier, full of purpose. The term sheet still sat on his desk. He picked it up, read it through one more time, and placed it carefully in the center of his blotter. Then he opened his laptop and began to type.

He did not know if what he was writing was a resignation letter, a manifesto, or a suicide note for the company he had built. He only knew that the vector had reached its endpoint, and the distance between what they had built and what they were about to sell was no distance at all. The two poles had converged. The ideal and the compromise were the same thing now, had always been the same thing, separated only by the story he had been telling himself for three years.

The cursor blinked on an empty screen. Daniel Kao began to write the truth.