What was idealism and what was greed could not be told apart by looking. They existed on the same spectrum, two poles of a single latent space, and every human decision lived somewhere on the vector between them. The only question was where.

Kyle Dennison kept his first-generation Bondi Blue iMac running twenty-four hours a day on a plain wooden desk he had built himself, in a garage on Emerson Street that still smelled faintly of the previous owner's lawn fertilizer. The garage had no heat and no air conditioning. In January his fingers stiffened on the keyboard. In August the sweat pooled beneath his wrists and smeared the ink of the notes he wrote on yellow legal pads. He did not mind. The discomfort felt like proof. Proof that he was building something real, something that mattered, something that existed independently of the money that his classmates from Stanford were already collecting in signing bonuses from Oracle and Sun Microsystems.

THE FIRST SNAPSHOT: Approximately eighty percent idealism, twenty percent greed. The vector was almost fully aligned with the pole marked Purpose.

He called the system Intueri. He had taken the name from the Latin verb meaning to look at, to contemplate, to gaze upon with understanding. The irony was that Intueri was designed to look at people in ways that people could not look at themselves. It ingested video feeds, voice recordings, linguistic patterns extracted from email and chat logs and discussion-board posts. It built models. It found correspondences between the angle of an eyebrow and the probability of a lie. Between the micro-pauses in a sentence and the emotion being suppressed. Between the word choices someone made when they were confident and the word choices they made when they were bluffing.

The first version ran on a single desktop machine and took forty-seven hours to process a ten-minute conversation. Kyle did not care. He was twenty-six years old and he believed, with the absolute conviction of the young and the brilliant, that he was building a tool for human understanding. Not a weapon. Not a surveillance device. A mirror. A way for people to see themselves, to see the gap between what they said and what they meant, between what they felt and what they showed.

He said as much to Ethan Rosenzweig, his first engineer, who had dropped out of the Stanford PhD program in computational linguistics to join Kyle in the garage. "This is going to be like the printing press," Kyle told him. "Not the internet. The printing press. A tool that makes people more literate about themselves."

Ethan, who was twenty-four and had the kind of pale, indoor complexion that came from spending childhood summers in basement computer labs, nodded without committing. Ethan believed in the technology. He was less certain about the mission. But Kyle was paying him in equity and the promise of Series A funding from a partner at Kleiner Perkins who had seen an early demo at a DARPA contractor showcase, and Ethan had student loans and a girlfriend in medical school and a mother in Sacramento who needed help with the mortgage. Twenty percent greed was not nothing.

The Sand Hill Road meeting happened on a Tuesday in March 1999. The venture partner was a man named Gerald Fisk, who had made his first fortune at Netscape and his second at a search-engine startup that had sold to Yahoo for an amount that the technology press described as "eye-watering." Gerald's office had floor-to-ceiling windows that looked out on the brown hills of the Stanford foothills, and a conference table made of reclaimed redwood that had probably cost more than Kyle's parents' house in San Jose.

"Show me what it does," Gerald said.

Kyle had prepared for this. He had a demo video. A volunteer subject, one of Ethan's friends from the linguistics department, recorded in conversation with a trained interrogator. Intueri's analysis overlaid on the video in real time: emotional valence scores, deception probability indicators, micro-expression classifications mapped to the Paul Ekman taxonomy. The system flagged three moments of probable deception, two of which the subject later admitted were accurate.

Gerald watched the entire seven-minute demo in silence. When it ended, he turned away from the screen and looked at Kyle with an expression that Kyle, even with all his training in reading faces, could not classify.

"You've built a lie detector that actually works," Gerald said.

"It's not a lie detector," Kyle said. "It's a self-awareness tool. It helps people understand their own emotional states and communication patterns."

"Sure," Gerald said. "And the telephone was invented to help farmers check the weather."

THE SECOND SNAPSHOT: The vector had moved. Sixty percent idealism, forty percent greed. The shift had happened so gradually that Kyle did not notice it until he was looking back. The vector never announces itself. It simply slides.

The term sheet came three days later. Two million dollars for twenty percent of the company. The valuation was generous for a seed round but not absurd. Kyle signed it in the garage, using a pen that Ethan had found in a drawer. The ink was dry and the signature came out scratchy and uneven. Kyle looked at it and thought: this is what commitment looks like. A scratchy signature on a piece of paper that will change everything.

What changed first was the office. They moved out of the garage and into a converted warehouse on Page Mill Road, across the parking lot from a Webvan distribution center that had not yet opened and would not survive the crash. The new office had cork floors and glass walls and a refrigerator stocked with Odwalla juices and Nantucket Nectars and the specific brand of sparkling water that Gerald Fisk preferred. Ethan hired three more engineers. A Stanford MBA named Priya joined as head of business development and immediately began setting up meetings with advertising agencies, political consultancies, corporate security firms.

"This isn't what we talked about," Kyle said to Priya, in a glass-walled conference room that smelled of new carpet and ambition.

"What did you talk about?"

"Self-awareness. Self-understanding. Giving people tools to know themselves better."

Priya looked at him with an expression that was not quite pity and not quite impatience. She was thirty-one, five years older than Kyle, and she had spent three years at McKinsey before business school. She knew things about the world that Kyle was only beginning to suspect.

"Nobody pays for self-awareness," she said. "People pay for advantage. People pay for knowing things that other people don't know. People pay for power."

"That's not why I built this."

"No," Priya said, and her voice was not unkind. "But it's why you can keep building it. The money has to come from somewhere, Kyle. The money always has to come from somewhere."

THE THIRD SNAPSHOT: Forty percent idealism, sixty percent greed. The crossover point. The moment when the vector crosses the midpoint and the story becomes something different.

Kyle stopped sleeping well. He woke up at three in the morning and lay in the dark of his rented house on Waverley Street, listening to the distant hum of 280 traffic and thinking about the names on the client list that Priya had compiled: Omnicom Group, the world's largest advertising holding company. Greenberg Quinlan Rosner, the Democratic polling firm. A defense contractor whose name he had been told not to speak aloud, even in his own home.

At a product demo for Omnicom executives, held in a conference room at the Four Seasons in East Palo Alto, Kyle showed them how Intueri could analyze focus-group footage and identify the exact emotional triggers that made consumers reach for their wallets. The executives took notes on leather-bound pads. Their faces showed nothing. Their micro-expressions, had Kyle been running Intueri on them, would have revealed a mixture of excitement and fear and something that looked very much like hunger.

After the meeting, a senior vice president named Todd pulled Kyle aside. Todd had the kind of tan that came from weekends in Napa and the kind of handshake that lasted just slightly too long.

"Here's what I see," Todd said. "I see a tool that can tell us what people want before they know they want it. That's not just market research. That's the end of market research. That's the beginning of something new."

"What would you call it?"

Todd smiled. His teeth were very white. "I'd call it the most valuable company on earth."

That night, Kyle drove home on 280 with the windows down even though it was cold. The air smelled of eucalyptus and dried grass. He passed the exit for Sand Hill Road and did not look at it. He was thinking about something Gerald Fisk had said, months ago, in a moment of what might have been honesty or might have been manipulation. Gerald had said: "Every founder I've ever backed has had to choose. The ones who succeed are the ones who stop pretending they have a choice."

Kyle did not know if that was true. He did not know if he had already made the choice without realizing it. He did not know if the vector was something you controlled or something that controlled you.

What he knew was this: the company now had forty-seven employees. The Series B was oversubscribed at a valuation of eighty million dollars. Pets.com had just gone public at a valuation that made no sense, and Webvan was about to do the same, and the whole world seemed to be running on the belief that if you moved fast enough, the contradictions would never catch up with you.

THE FOURTH SNAPSHOT: Twenty percent idealism, eighty percent greed. The vector had almost reached its destination. What was left of the original purpose was not gone, but it had become an accent—a faint coloration, a rhetorical residue, a thing Kyle mentioned in all-hands meetings and then immediately contradicted with the next slide in the deck.

The defense contract came through in September. The amount was classified but the scope was not: Intueri would be deployed at international airports, integrated with the security camera systems, used to flag passengers whose micro-expressions indicated deception or hostile intent. Kyle sat in a windowless room in Crystal City, Virginia, being briefed by men in uniforms who spoke in acronyms he did not understand.

"It's just probability scores," one of the men said. "It's not making decisions. It's providing information."

"Information that decides who gets detained and who doesn't."

"Information that decides who might be dangerous and who might not be. There's a difference."

"Is there?"

The man looked at Kyle for a long moment. He had a face that was designed for looking at people for long moments. "You tell me," he said. "You built the thing."

Kyle flew back to San Francisco on a red-eye. From the window of the plane he watched the lights of the Midwest scroll past, grids of sodium-orange arranged in patterns that from thirty thousand feet looked like circuit boards, like the inside of a machine, like a map of something that had been designed by an intelligence that understood patterns but not people.

He thought about the original code. The first version of Intueri, the one that ran on a single iMac in a garage. The one that was supposed to be a mirror. The one that was supposed to help people see themselves.

The code still existed. It was buried somewhere in the version-control system, buried under layers of patches and optimization and feature requests from clients who wanted to know not how people felt but how people could be made to feel. It would still run. It would still do what it was originally designed to do.

But nobody was asking for that version. Nobody was paying for that version. Nobody wanted a mirror when they could have a weapon.

THE LATENT SPACE

Here is the thing about vectors in latent space: they are reversible in theory but not in practice. You can always trace the path back from greed to idealism. You can always reconstruct the original state. But reconstruction is not the same as return. The space is continuous, but movement through it is not. Every shift changes something that cannot be unshifted. Every step toward one pole makes the other pole a little harder to see, a little harder to believe in, a little harder to reach.

Kyle Dennison became a very rich man. The company went public in March 2000, two weeks before the NASDAQ peaked and began its long collapse. Kyle sold enough shares at the IPO to buy a house in Atherton with a swimming pool and a guest house and a lawn that required a full-time gardener. He kept the iMac from the garage. He put it in the guest house, on a desk that faced a window that looked out on the pool. He never turned it on.

Sometimes, when he could not sleep, he would go out to the guest house and sit in front of the dead blue screen and think about what he had built and what it had become. He would think about the man who had sat in that cold garage and believed he was building a mirror. He would think about the distance between that man and the man he was now.

The distance was not large. That was the thing that kept him awake. In latent space, the distance between eighty percent idealism and twenty percent greed was almost imperceptible. It was a single step. A single decision. A single term sheet. A single moment of telling yourself that the money had to come from somewhere, that the funding had to be raised, that the product had to be sold, that the thing you built was still the same thing even when it was being used for something else.

And the distance from there to forty percent idealism, sixty percent greed was another single step. And the distance from there to twenty percent idealism, eighty percent greed was another.

One step. Then another. Then another.

None of them felt like a betrayal at the time. Each one felt like a necessity. Each one felt like the only reasonable thing to do. Each one felt like a step toward something, not away from something.

But when you looked back from the far end of the vector, when you reconstructed the path you had traveled, you saw that each step had been away. Away from the garage. Away from the mirror. Away from the man who had believed that understanding was an end in itself.

Kyle Dennison was thirty-two years old. He was a founder. He was a CEO. He was a very rich man. He had built a tool that could read the human face with greater accuracy than any human being in the history of the species. He had built a tool that knew what people wanted before they knew it themselves.

He had not yet used that tool on himself. He was afraid of what it would show him. He was afraid that the vector was not finished moving. He was afraid that there were poles he had not yet imagined, destinations he could not yet see, steps he had not yet taken that would feel, when he took them, like the only reasonable thing to do.

The iMac sat on the desk in the guest house, turned off, gathering dust. The screen was blank and blue and reflected the light from the pool. If you looked at it from the right angle, in the right light, you could see your own face in the reflection, distorted by the curve of the CRT glass, recognizable but strange, like someone you used to know.