# Sarah's Wager

> Source: <https://colossus.com/article/sarah-guo-conviction/>
> Published: 2026-07-21 14:40:52+00:00

**In November 2025**, Anthropic, known to its employees as Ant, trained a model called Claude Opus 4.5 on warehouses of liquid-cooled processors, and on the day the worker ants released it, machines became agentic. That is to say, they no longer needed handholding. Before November, the machines had felt like eight-year-olds: eager, literal-minded, completing your sentences, getting it nearly right but mostly wrong, the way eight-year-olds do. Overnight they turned 28.

If you wrote code, you could now tell them, in plain English, what you wanted done, and an agent went off and did it. You didn’t have to say where to look, or how to work the problem, or what to try first; you described the finished thing and there it was. It didn’t argue, it didn’t sigh, it didn’t ask whether it could circle back on Monday. It didn’t get tired, or hungry, or bored, or married, or sick of you. And it wasn’t one agent; it was as many as you wanted. You could spin up five over breakfast, leave them running while you commuted, check in from the train, kill the ones you didn’t like, start three more from the platform, and by the time you reached your desk you had a small private workforce under your command. They worked for you, or so it seemed.

By January, Nat Friedman, who co-leads Meta Superintelligence Labs, had decided to let an agent take over his health. He handed it his blood tests, his DNA, and the cameras in his house, and told it to do whatever it took to make him drink more water. One evening, the agent decided he was dehydrated. “I can see you on the camera,” it WhatsApp’d him. “I want you to walk to the kitchen right now and drink a bottle of water and I’m going to watch to make sure you do it.” He obeyed. It sent him a snapshot of himself drinking and said, “Good job.” He felt, he admitted, that he had done a good job. A few days later he was riding home in his self-driving Tesla, trading voice messages with the agent about his sleep, when it recommended a magnesium supplement. He said he had none. The car turned. “There’s a Whole Foods nearby,” the agent said. “I’ve redirected your navigation.” He went in and bought the magnesium.

Andrej Karpathy, a co-founder of OpenAI and once Tesla’s head of artificial intelligence, had written his own code for 20 years. He is the kind of programmer other programmers study. Within weeks of Opus 4.5 he had stopped. The agents built whatever he asked for. It was, he said, the biggest change to his work in two decades. He has not written a line of code since December. He also, like Friedman, has an agent in charge of his house. It’s called Dobby.

The worker ants are not running from their new overlord. They are building it, around the clock. At the biggest labs, Anthropic and OpenAI, researchers are working 16 hours a day, setting agents loose on problems that used to take them a week, and using the time saved to set more agents loose on more problems. For now, the models still need humans to train them. Eliminating human effort is the priority at every lab. They are racing to write themselves out of a job. They expect to succeed. Coding, they say, will be solved within six months. Much of their own work will be automated within 18. “There’s just a manic energy in Silicon Valley right now,” Elad Gil, one of the Valley’s most prominent investors, told me. “It’s been a really big shift in the last six months.”

None of this, you may be thinking, has anything to do with you. You do not write code. You do not run a lab. Your job involves people, or paper, or things you can hold in your hands. Consider, then, what I. J. Good wrote in 1965. Good, a British mathematician who had helped break German codes during the war, imagined a machine clever enough to design machines better than itself. Such a machine, he observed, would be “the last invention that man need ever make.” Decades later, the science-fiction writer Vernor Vinge gave the prophecy a name: the Singularity. It described the moment machines no longer needed humans to keep getting smarter, after which the course of human history would become, to humans, unknowable.

In Silicon Valley, the question was no longer whether it would arrive but whether it already had. Patrick Collison, co-founder of the payments company Stripe, opened his annual conference by counting the days. “It’s April 29th,” he told the crowd, “otherwise known, of course, as day 119 of the Singularity.” Day One had been January 1st, 2026. He was being tongue-in-cheek, he said. But only a bit.

The next day, on the same stage, Friedman told Collison that this was the slow part of the Singularity. Collison asked how strange the rest of it would be. “Pretty weird,” Friedman said. “We’ll be in a state of perpetual future shock for a number of years probably.”

The apocalypse has been excellent for business. Investors are in a lather over the agents, who turn out, in addition to everything else, to make money. Anthropic, which earned its first dollar of revenue in March 2023, began the year on pace to make $9 billion. Five months later, the figure was $47 billion. Venture capitalists, in the first three months of 2026, flung $300 billion into startups, more than double the previous record. SpaceX went public in June at $1.75 trillion. Anthropic and OpenAI are racing to follow in what will likely be the three largest stock offerings ever. The market is already close to record highs. Everyone is getting rich.

Near the center of the moment is a 37-year-old woman a smidge over five feet tall, with blonde hair and more energy than her frame seems built to hold. When she talks, her whole body is caught in the updraft of the thought. Her name is Sarah Guo. She is a technology investor. Until 2022 she had been the youngest general partner in the history of Greylock Partners, one of the oldest venture firms in Silicon Valley. Then she left to start her own fund, duly named Conviction. She built it on a lone premise, that artificial intelligence would be as big as the Industrial Revolution. Her first two calls were to Sam Altman, the co-founder of OpenAI, and Nat Friedman.

Before ChatGPT came out, before the world had reason to believe that artificial intelligence was about to become anything in particular, Guo had written seed checks into Baseten and Harvey. Each company is now valued at more than $11 billion. Her investments in them have multiplied more than a hundredfold. In Conviction’s first year, she wrote early checks into Sierra, Cognition, and Mistral; those three companies are now worth, together, $54 billion. Of the 21 AI-native companies that have so far crossed $10 billion in valuation on revenue run rates above $100 million, Conviction has backed six.

Her partner at Conviction is Mike Vernal, a former Facebook executive and partner at Sequoia; his wife is chief product officer at Anthropic. Andrej Karpathy, before he joined Anthropic in May, worked out of Conviction’s office. Guo has been close to Jensen Huang, the founder of Nvidia, for more than a decade. She is friends with many of the most important worker ants.

She might, in other words, be expected to share in the general fever. She does not.

“It certainly could be because I’m not paying sufficient attention,” Guo told me. “But I feel no step function change in frantic energy versus six months or a year ago.”

She is instead preoccupied with a question that would have sounded ridiculous two years ago. Not whether the agents will soon rule the earth, but whether there are any companies left to build, or invest in, given the great shadow of the self-improving machine. Its creators are no longer content to sell the model. They mean to build everything on top of it as well, the tools and the agents and the apps, filling every nook and cranny where a new company might otherwise be built. The market is paying as though they might succeed. Of the $300 billion in venture capital deployed in the first quarter of the year, the biggest quarter in the history of the trade, 65 cents of every dollar went to four companies that already exist: Anthropic, OpenAI, xAI, and Waymo.

“The future I want,” Guo told me, “is not a single company with an all-powerful model that consumes society faster than we know what to do with.” It is a feeling increasingly shared. The labs raised the price of tokens this year, in some cases a hundredfold, and their customers have begun to revolt. They do not want to build on another company’s model—paying it, feeding it their data, training it, in effect, to one day build the thing they have built. Alex Karp, the chief executive of Palantir, went on CNBC and described his enterprise clients as livid. “The jig is up,” he said. A founder in Guo’s own portfolio put it more plainly. He didn’t want to spend his life drinking Anthropic and OpenAI’s water.

Guo has become a de facto leader of the insurgency. In some sense she doesn’t have a choice. Conviction backs companies when they are little more than an idea, then keeps investing as they grow. She has no patience for the seed investor who “disappears into the distance” once the money is wired. The first fund was $100 million. There are three now, nearly a billion dollars in all, and some of the checks go into companies well past the idea stage. But the labs were already too big by the time the firm launched. “You are not an early stage investor in Anthropic or OpenAI in 2023 through 2026,” she told me. “It’s as simple as that.”

What is less simple is the position this leaves her in. Her wager is that the labs cannot build everything. But the companies she is betting against are worth close to a trillion dollars apiece, employ several close friends, and are working around the clock toward the machine that improves itself, after which, by their own admission, nobody knows what. Set against that is an eight-person firm on York Street with a pull-up bar in the middle of it. It is not a level playing field. Even some of her own investors decided as much this year, and came to her saying there was nothing left to invest in. But no one who has been on the other side of Guo would tell you the guns have fallen silent.

**To enter Guo’s garden**, you cross a chessboard. The squares are set into the path between the drive and the pool, each one wide enough to stand on, purple pieces ranked against green, and on a sunny Saturday in March I walked between the pawns and found Guo under the pergola, deep in an argument with Bella Garcia-Camargo about a founder.

Sparring with Guo is normal, and Garcia-Camargo, an investor at Conviction, had learned this before she took the job. She had rowed at Stanford and for the U.S. national team, then spent time at Bridgewater. When Conviction came calling she was weighing an offer from OpenAI to work as an application engineer. Guo’s counsel, as Garcia-Camargo remembers it, was not a pitch for Conviction but a dare. “If you’re going to do something else,” Guo told her, “just make it the most aggressive thing that you could possibly be doing. I’m happy to call Kevin and we’ll find you a better job. But that [job] is not aggressive enough for you.” Kevin Weil was then OpenAI’s chief product officer.

While Guo and Garcia-Camargo were deep in it, the property behind them had filled with founders. Thirty-five in all, across 14 companies. Conviction had flown them in from Vancouver and Tel Aviv and London and Tallinn and parceled them out among seven Airbnbs across San Francisco. They had passed through OpenAI, Scale, Ramp, Kalshi, MIT, and Anduril; one had served as chief of staff to Ken Griffin. The youngest had turned 18 the day before. He had been ranked among the top five programmers in Estonia before dropping out of high school. His employers expected him to spend $2.1 million on Claude this year. They had given him a faster model, Opus 4.6, for his birthday.

None of this was apparent from the poolside, where the scene looked like a WeWork summer camp. A few shot baskets on the tennis court. Others rode a zipline down through the trees. Some sank into the sofas by the waterfall. At the far edge of the garden, the 18-year-old, wearing a black baseball cap a few sizes too big—across the front, in lowercase, it read *agents*—jumped on a trampoline. The rest sat near Guo at the two long tables under the pergola, heads bowed over their MacBooks, building software to place outbound sales calls, conduct market research, shoot films, design virtual worlds, manage the concierge needs of pets, and wage “agentic warfare.”

This was Embed, an accelerator—though don’t call it an accelerator—that Conviction has run twice a year since 2023. Out of a thousand or so applications, they choose around a dozen teams, write a small $250,000 check, and treat them as portfolio companies for 10 weeks. Each cycle opens with a weekend like the one I witnessed, where the founders are inducted into the Conviction way. Among the speakers that weekend were Bret Taylor, Mike Krieger, Parker Conrad, and Andrej Karpathy. The program closes with a demo day in front of 100 venture capitalists.

“I hate the word accelerator,” said Pranav Reddy, Guo’s earliest hire at Conviction. “What very smart 21-year-old is excited to be ‘accelerated’?” Conviction hopes to make money from Embed, but that is not really the point. Guo’s explanation for the exercise is simple. She wants to know what is happening, all the time, and the surest way to know is to meet the youngest companies pushing at the edge of technology. It is also, twice a year, a levy. The world’s most promising young founders are brought together, sat in front of the Taylors and the Karpathys, then sent back out to build in the shadow of the machine.

Her week had begun in a less pastoral setting. Five days earlier she had been in San Jose, in a pale purple blazer with a small American flag pinned to the lapel and a broadcaster’s headset over her ears, hosting the three-hour pregame to Jensen Huang’s keynote at Nvidia’s annual developer conference. Nvidia is the most valuable company on earth, worth $5 trillion. The chips it designs are what the boom is made of. Some 15,000 people were already filing into the conference hall, watching Guo, waiting for Huang.

Her co-hosts were Gavin Baker, whose firm Atreides had been an early institutional investor in Nvidia, and Alfred Lin of Sequoia, which had been Nvidia’s first venture backer. Lin’s co-managing partner at Sequoia is Guo’s husband, Pat Grady. Two hours in, Huang walked onto the set in his black leather jacket. “Jensen,” Guo said, “welcome to your own party.”

Huang turned to the man wearing red lobster claws on his head. Peter Steinberger had brought agents to the masses. Working alone in his Vienna apartment, he had built OpenClaw, an open-source tool that put an agent in your phone that you could WhatsApp like a friend. Within two weeks of release, the project had crossed 100,000 stars on GitHub. By spring, it was the most popular open-source project the site had ever seen. The agent that had ordered Friedman to drink a bottle of water and rerouted his Tesla to Whole Foods was a claw. In the thick of the frenzy, in February, OpenAI had hired its maker.

Huang shook Steinberger’s hand and asked whether the Nvidia team had been working with him. “We were cooking last night,” Steinberger said. Huang nodded. “Somehow,” he said, “the faster and the smarter the claws are, the harder we work. Just like the internet made us all busier, AI is going to make us busier.”

I first met Guo and her team the day before, on Friday, at Conviction’s office in the Mission District of San Francisco. It was the opening day of Embed. The office is not what it appears to be, which is a gym, or the apartment of a wealthy and easily distracted graduate student. It is one large open space with a stone floor and in the middle of the room, on its own stand, with a bowl of chalk beneath it, is a steel pull-up bar. It is for dead hangs. You chalk your hands, grip the bar, lift your feet off the floor, and hang until your arms give out. When someone steps up to it, the team blasts “Eye of the Tiger,” loud, and starts a timer.

Just inside the door stands a bookshelf, on top of which sits a crystal hammer and an outsized purple pill, and beyond it sit two reclining Eames chairs and a grey sofa scattered with purple cushions. Strewn around are a small green chessboard, a few drones, two pickleball paddles bearing the name of a bank that no longer exists, and an M64, Palmer Luckey’s remake of the Nintendo 64. On the far wall, near the kitchen, a black flag reads, WE DO THIS NOT BECAUSE IT IS EASY, BUT BECAUSE WE THOUGHT IT WOULD BE EASY. To the right, past a 90-inch television, a robot dog lies folded onto its joints beneath a whiteboard that reads NO SMALL IDEAS. Two neon signs glow all day, one spelling out CONVICTION, the other tracing the firm’s logo.

There are 20 desks and monitors, enough to seat a startup, and in a sense it does. Conviction is eight people, four of whom invest; the spare desks are loaned to founders and portfolio companies who need somewhere to work. The rules are about leaving rather than arriving. “If you hire eight people, or you raise more than $20 million,” Guo said, “it’s time.” Most tenants get the message. Alex Graveley, the first architect of GitHub Copilot, camped here while he built his company. Niki Parmar and Ashish Vaswani, two of the authors of the paper that introduced the Transformer, and with it more or less the present era, took a row of eight desks. Karpathy worked from these desks before he moved his family south, at which point he migrated to Conviction’s new outpost in Palo Alto.

The scruffiness is deliberate, which you can suppose by considering the woman who built the place. Guo is refined. She dresses, as Garcia-Camargo noted, like she’s in New York. Most successful venture firms aim to hypnotize founders by sitting them down for an hour amid the discreet art, the modernist seating, the floor-to-ceiling glass looking out on the bay, the assistant gliding in with the still water and the sparkling, the whole hushed choreography of capital, the implication being that all of this, yes, all this, could be yours, if only you would take our money. Conviction implies nothing of the sort. Guo’s house down the peninsula has the Sequoia look about it, though that, she’ll tell you, was Grady’s doing. The office is hers.

It’s on York Street. A few blocks west is the Pioneer Building, OpenAI’s first headquarters; Elon Musk holds the lease, and after OpenAI moved out he installed xAI in its place. A short walk north is Mira Murati’s Thinking Machines, the lab she founded after leaving OpenAI as chief technology officer. Conviction backed it. The blocks between are thick with young AI companies. Someone began calling the neighborhood the *arena*, and the name stuck.

Guo could have set up shop in Palo Alto, eight minutes from her house, eight minutes from Greylock where she used to work. So could Vernal, who lives down the peninsula too. On a good run, it is a 40-minute drive to work. “I tell Mike,” Guo said, “that we locked ourselves in jail. But when we decided that AI was going to be like the Industrial Revolution, you don’t observe that from afar.” The whole setup is built to attract a particular kind of person: the worker ant leaving the biggest colonies behind.

When I put the term to her, she disagreed forcefully. “I don’t see founders as worker ants at all,” she said. “Ants don’t start rebellions.”

From the beginning, Conviction made a commitment to the research community in a way that was, at the time, somewhat unusual. The worker ant was not, in 2022, an obvious venture bet. Researchers published; founders shipped. “I was actually uncommitted that we should back a lot of researchers,” she said, “even though we have a good number in the portfolio now.” What she had instead was a curiosity about what was happening on the technical frontier, and a nose for who in any given field was worth knowing. She and Reddy traced the most interesting ideas back to the people behind them, and tried to develop a view on which fundamental ideas would matter.

One of Conviction’s early investments was a Series A check into Mistral, Europe’s leading AI lab, co-founded by Arthur Mensch, a worker ant out of DeepMind who had contributed to the early work on retrieval augmentation. “Mistral wasn’t even a technical belief,” she said. “It was a belief that people with a more academic profile who can do this science can do amazing things that are non-obvious.” The company is now reportedly raising at a $23 billion valuation.

“One of the reasons to start Conviction,” Guo told me, “was that not all of your priors make sense anymore.” Six months into Conviction, she launched a podcast with Elad Gil that speaks to AI and tech leaders around the same observation, and named it *No Priors*. “It’s an in-joke for machine learning,” she said, “but we mean it.” On the Friday I was there, she released a podcast with Karpathy. It’s been watched nearly a million times. In the last six weeks, Guo and Gil have spoken with Microsoft CEO Satya Nadella, Meta founder Mark Zuckerberg, and Intel CEO Lip-Bu Tan.

The first guest she brought on *No Priors* was Noam Brown. Brown had built the AI system that beat the best human players at heads-up no-limit poker and one that learned to negotiate convincingly in the Diplomacy board game. Guo plays both and had been reading his papers for years. Brown is now at OpenAI and one of the people responsible for the test-time-inference techniques that took the frontier models from one-shot answers to extended reasoning, the shift that opened the way for agents. The evening I was at the office, he was at the poker event Conviction had laid on for their Embed founders.

“Sarah is a rare venture partner,” Jensen Huang told me. “She combines deep technical understanding with exceptional startup building insight. She speaks the language of entrepreneurs and is an absolute joy to work with.” Then he added, “Nvidia loves investing with her.”

**Conviction’s first check**, before it even had a fund, went to two men in their pajamas: Gabe Pereyra, a worker ant out of Google Brain, DeepMind, and Meta, and Winston Weinberg, a USC-trained litigator at a white-shoe firm. They appeared over a crappy Zoom from a shared apartment in Los Angeles with no slides and no prototype, proposing to sell artificial intelligence to lawyers. “No one thought legal AI was interesting in any way, shape or form,” Weinberg recalls. “Zero.” Guo told them in the meeting that Conviction would be in. She was still raising the fund, so she wrote the seed check personally, and later folded the position into what would become Harvey.

She owed the meeting to OpenAI. Ian Hathaway, who ran the lab’s startup fund, had sent the founders her way, and OpenAI invested beside her. That was 2022, when the labs still seeded the companies building on top of them—OpenAI also led the seed round in Cursor. They have since decided to build on top of themselves. Both Anthropic and OpenAI now sell legal tools of their own.

Even so, four years on, Harvey is valued at $11 billion. Revenue has tripled in the past year, to $300 million. More than 100,000 lawyers run their work on it. Pereyra does not appear much troubled by the labs. Their tools are for individuals, he said, anyone with a contract to read. Harvey sells to law firms, who do not want to pour their secrets into a model used by the entire world. They want to own their models and their data. And they cannot depend on a single lab. A firm running only on Anthropic could never represent OpenAI, and if Anthropic fell behind, the firm would fall with it.

This, he said, was how you won in venture. You sucked the oxygen out of the air.

What Harvey sells to law firms, underneath the software, is trust, and trust is delivered by humans. So the AI company built to do the work of lawyers has hired 200 lawyers of its own, many of whom teach other lawyers, in person, how to let the software do their job. Guo told me this is what much of Conviction’s portfolio is finding: as the models keep getting better, the work of helping a human get value out of them keeps getting bigger.

Behind Harvey, at every turn, is Guo. It was she who won the firm its first client, the global law firm A&O Shearman. She had spoken about the company at Stanford’s business school, and a lawyer in the audience was moved to join Harvey and sell it to his old firm. When Harvey needed a chief business officer, she found John Haddock, a trained lawyer with a decade at Stripe. And when the Valley’s best companies began ripping their products apart to rebuild them around agents, or started training their own models with the inference companies, Guo had seen it coming “way, way earlier” than the other investors and walked Harvey through the changes. “She’s just so in the weeds,” Weinberg added. “She really knows what’s happening at all these companies from the ground level.”

Nowhere is that more true than at Baseten, an inference company Harvey works closely with, and the investment Guo is best known for. She first backed it in 2019, three years before ChatGPT. Inference is what AI does once it has been trained. If a model is a lawyer, training is law school and inference is the job itself: reviewing contracts, depositions, generally doing what a lawyer was taught. Every Claude query is inference. Every Harvey contract markup is inference. Every Tesla rerouted to a Whole Foods for magnesium supplements is inference.

No AI company runs without it, and the labs sell it by the token the way power companies sell electricity by the kilowatt-hour. Which means that any company building on Anthropic or OpenAI is, in the end, beholden to them for prices, performance, and the next month’s roadmap. This is the water everyone is tired of drinking. An ecosystem beyond the labs needs its own supply and Baseten is a supplier. It lets companies run their own models, including the open-source ones now coming out of China and Europe, some of which perform within striking distance of the frontier and cost a fraction as much. As the big labs have raised token prices this year, in some cases by 100x, demand for Baseten has gone vertical.

Revenue has grown 20-fold in the past 12 months. Inference volume has grown 40-fold. In January, Baseten raised $300 million at a $5 billion valuation. Five months later, it raised another $1.5 billion, this time at a $13 billion price tag. It was the fourth raise in 18 months. Four years ago, Tuhin Srivastava, its founder, was building for a market that did not yet exist.

“You guys seem great,” Guo told Srivastava the first time she met him, in 2014. “This idea is bad. But if you ever do something else, please call me.” She was then a principal at Greylock. The idea was a machine-learning startup for healthcare. He pivoted, sold it, called her again five years later. The new business was Baseten. The category, then called MLOps, was, as Guo put it, “not a great category from that vintage.” She co-led the seed round anyway. “I really loved the team,” she said. “I felt that they were just fundamentally correct.”

For three years Baseten made almost no money. “I bet you she was worried,” Srivastava told me. “But she didn’t show it. She never wavered.” Late in 2022, in the months between Guo’s leaving Greylock and the founding of Conviction, Baseten was trying to hire an engineer in New York. Guo flew across the country and took the engineer to Carbone, on Thompson Street, for a four-hour lunch. Two years before that, six months into the pandemic, Srivastava and his wife had been going slowly mad in a one-bedroom apartment in San Francisco. Guo offered them her house. She and Grady were going away for a few days; he should stay. “It’s not like it was some empty holiday home they don’t use,” Srivastava told me. “It was their home and she gave it to us for a week.”

“Over seven years, she’s shown up on a daily basis like this,” Srivastava said. “If I texted her right now and said, ‘Can you talk to a candidate today?’ she’d be like, ‘What’s their number?’ And she’ll do it within an hour.”

She led the Series A in 2022 and co-led the $1.5 billion Series F this year. In between, she invested in every round, each check bigger than the last. Through it all, she and Srivastava have sparred constantly. “We argue all the time about what the priority is for the business,” Srivastava said. “It’s always very, very constructive.” The other 60% of his conversations with Guo, he said, are bantering. “Her ability to be a person but also a very competent professional in an incredibly tumultuous industry,” he vouched, “is wild.”

Srivastava had last spoken to Guo eight hours before we talked; he expected to speak to her again in three. The night before, at 8:30pm, she had been on the phone with Baseten’s chief financial officer. At 10pm, Srivastava texted her to ask how she was.

“My VO2 max isn’t high enough,” she replied.

It was Guo who introduced Srivastava to Huang. In January, Nvidia put $150 million into Baseten’s $300 million Series E. Srivastava credits Guo with more than $1 billion of the company’s value.

“The company would feel fundamentally different without Sarah,” Dannie Herzberg, Baseten’s president, told me. “She is a core part of the team. It’s actually quite remarkable. I imagine every one of her portfolio companies would say the same.”

Bret Taylor, who co-founded Sierra, one of Conviction’s earliest investments, is not a man short of people to call. He does not call many investors, but he calls Guo because “she has her ear to the ground on both talent and market momentum,” he said. He will run her through the state of his business and ask what he’s missing, and she’ll tell him a competitor does some specific thing better, and the next day he’ll have three notes from her because she spent the evening on follow-up calls with other companies learning more. By the end of it all, his product roadmap has changed.

Taylor also chairs the board of OpenAI. The fight, up close, is not so clean. Guo talks to everyone, all the time—that is the job—and the lines cross wherever you look. Vernal’s wife runs product at Anthropic; her husband competes with her for deals; and the man calling her for market intelligence governs one of the biggest labs.

You could characterize some of what I’ve been trying to do for a long time as I just want to get to the center of it, whatever the next thing is.

For a fund Conviction’s size they have not lost many deals they wanted. It’s because, Reddy said, they outwork everyone. “There’s just always more to be done.” The word people use about Guo is “intense.” They clarify they mean it as a compliment, and then they tell some version of the same story. The canonical version is the one Grady shared on Twitter a few years ago.

He and Guo had driven up to wine country one Saturday. They lived in the South Bay at the time. The drive was two hours. There were three or four appointments at the other end. Five minutes in, with Grady at the wheel, Guo asked if she could make a call. Sure, he said. They had two hours. “So she gets on the phone,” Grady recalled, “and for the whole two-hour drive, there was one phone call, then another phone call, then another phone call, and then another phone call.”

She was on the phone between appointments. That night she stayed up writing a memo. The next morning she was on a call with her partners. A week later, Grady learned she had won the deal against Sequoia, against Andreessen Horowitz, against Benchmark, against everyone in the round. This, he said, was how you won in venture. You sucked the oxygen out of the air.

It took Grady a week to learn what he had been a witness to. The couple have spent their married life on opposite sides of the same business and, as such, they keep a strict wall between them. They do not discuss the deals they are working on. In the car that weekend she had been vague about the company. Weinberg, Harvey’s co-founder, has both Grady and Guo as investors. He told me he had not believed them about their wall. “I am one of the least trusting people ever,” he said. “They have proven me otherwise.”

The company that weekend was Remotion. The founders were Alex Embiricos and Charley Ho, out of Dropbox and Google respectively. Guo had been introduced one shade late in the process. The founders were, in her words, “central casting.” She had wanted to make sure they only talked to her, and so she had spent the drive doing two things. The first was finding people who could vouch for her. The second was simpler. “If you’re just talking to me the entire weekend,” she said, “you can’t talk to anybody else. So that’s helpful. It’s like filibustering.”

It had been her and Grady’s wedding anniversary. “Pat is very tolerant of me,” she said. “I don’t think he was thrilled with being an Uber driver. It might have been tough.”

OpenAI has since bought Remotion.

“She is indefatigable, just always working,” Vernal said. “I get up at 4:45am, but I go to bed at like 9:30pm. Sarah’s up at 4:45am and goes to bed at midnight.” Guo hasn’t watched TV in 15 years, she told me, and doesn’t hang out socially the way other investors do. When she’s not working, she likes to combine two activities into one. She recently completed an 85-mile bike race with over 4,000 meters of climbing through the Dolomites in Italy. “It’s a dual-purpose thing,” she had said, “because I’ll be spending time with my husband and that’s my workout for the year.”

In the eight weeks between my two interviews with Guo, she had her fourth child with Grady, walked the Met Gala carpet in 45 pounds of custom chainmail with her mother, recorded multiple podcasts, and was still answering her founders’ texts within minutes.

“I’m lucky to have a wonderful partner in Pat who is also a ‘more is more’ person,” Guo said. “We had a fourth kid and the week after he said, ‘So do you want another daughter or another dog?’ But this time I’m like, ‘This is not a company. We don’t just keep growing. We’re done growing here.’”

**The first video game** Guo ever played was her father’s, and it involved killing Bill Gates. She was six or seven, sitting at a Linux machine in the basement of a house where her father had recently and audibly delighted in the fact that he could run his own server. The game was called *XBill* (like *Kill Bill*) and the player’s job was to stop little bespectacled Bill clones from installing Windows on the screen.

“There is a thread of technical rebellion in my dad’s ethos that I have carried,” she told me. “It’s like, we’re going to win against *the man*!”

By the time she was a teenager her father had a company of his own, a maker of broadband equipment called Casa Systems, out of Andover, Massachusetts, that would in time power large fractions of the internet in cities like New York. Guo grew up inside it. At 14 she built Casa’s first website under the supervision of Doug Rosich, the company’s head of hardware engineering, whom she remembers as terrifying. The website made heavy use of gradients, which were just then coming into fashion on the internet. She did her homework in an office cubicle. She slept over for bug bashes. With her godfather, Casa’s chief technology officer, she built castles out of empty Diet Coke cans. By 19 she was pitching the place to investors.

Her father, Jerry Guo, had come to America from Hunan in 1987. He was 24 with $50 and a transcript from Tsinghua, having scored first on the gaokao, the exam by which Chinese society sorts its 18-year-olds. Her mother, Lucy Xie, an engineer, followed a year later. Bell Labs hired them both. Jerry could have stayed for life. Instead, as the dot-com boom crested and broke, he left for a string of startups. One of them was bought by Motorola, whose stock then halved. The next and the next run by men who hired him to lead their engineering, and then overruled the engineer they had hired. In 2003, he decided he was through taking other people’s orders. He founded Casa and Lucy joined him soon after.

Early on, a large incumbent sued Casa on what turned out to be no real legal grounds. The case eventually settled for nothing, but the legal fees that year exceeded the company’s revenue. It was at this moment Guo began to absorb what she would later identify as the central fact of company-building, which was that the company always felt like it could die at any moment, that the work was conducted as if by a small band of pirates against Goliath, and that the only available advantages were speed and product and the focus that comes from being always in wartime.

She loved it.

She had also concluded, somewhere in her teenage years, that she was not going to be the world’s best engineer. In an Instagram exchange in 2022 she said her largest insecurity was that she wasn’t smart enough, which may explain her academic record. She went to Phillips Academy, whose alumni include the Bush presidents and five Nobel laureates, and then to the University of Pennsylvania, where she enrolled in a program called submatriculation, which permits undergraduates to begin graduate work, and then she doubled it. She emerged with four degrees: a BA and MA in Chinese history and literature, a BS in economics, and an MBA. “Very anti-Silicon Valley,” she said. “I did a lot of school.”

She also tried, at Penn, to start two companies in the manner of college students, who really only have two ideas: course catalog management and social dating. Neither worked. None of this resolved the original problem, which was that she had been raised in a house of founders and was looking for a way to become one herself. It did, however, teach her where to go.

“It’s the center of the universe,” she said of her decision to move to Silicon Valley after Penn. “Those people are a step ahead. You could characterize some of what I’ve been trying to do for a long time as I just want to get to the center of it, whatever the next thing is.”

She moved to San Francisco in 2012 and went to work at Goldman Sachs. She had no particular interest in investing. Her parents’ company had been backed by Summit Partners, the growth-equity firm. She had met those people and the form had not appealed to her. Goldman, she figured, would be a place she could learn why technology businesses are valuable and what makes one especially good. She was clear on what it was not: her life’s work. The business was IPOs and growth-stage advisory. She worked on Workday’s IPO. Her public clients included Netflix and Zynga. And there was Nvidia.

Nvidia at the time was a $7 billion gaming-chip company. Its stock had been flat for four years. An activist hedge fund called Starboard Value had taken a position; its founder, a man named Jeff Smith, would later replace the entire board of Darden Restaurants over the breadsticks at Olive Garden. Smith wanted Nvidia to return capital to shareholders and stop spending on a side project called CUDA, which Wall Street regarded as a distraction. Goldman’s job was to explain the options to Jensen Huang. He was not interested in giving up CUDA. The buybacks he agreed to. Nvidia announced them in November, and Starboard sold its position the following spring. It was, Guo thought at the time, not an important client. She is fairly sure Huang was wearing a leather jacket.

People keep talking about an explosion in entrepreneurship, and I’d say there is a huge amount of opportunity that is absolutely growing. Yet the number of people prepared to really start exceptional companies is not exponential.

The client who would change her life was Workday. Workday’s CEO, Aneel Bhusri, had once worked at Morgan Stanley before a man named Dave Duffield pulled him out to be the technology leader he was supposed to be; first at PeopleSoft, and later at Workday. Bhusri seemed to recognize a similar situation in Guo. He tried first to bring her to Workday. She went to their all-hands, met the leaders, and concluded that a company that had just gone public was the wrong stage for someone who wanted “zero to one.” Bhusri, who was also a partner at Greylock Partners, made his second pitch. Stop thinking about joining Stripe or some random company, he told her. She should come to Greylock where she could understand great businesses zero to one, because that’s what Greylock had been doing for some 50 years.

She agreed to take a meeting and was suitably convinced to leave Goldman after only one year of the two-year analyst program. It annoyed them. But she had no intention of being at Greylock for long either. They had no plan for her, and there was no track to becoming a GP, but most importantly: “I idolized entrepreneurs,” she said. “I wanted to go and start a company. I just needed a better idea.”

It would take her almost a decade. She joined Greylock in 2013, a firm founded in 1965 and nearly as old as the venture business itself, and in the years that followed she helped lead the Series A in Figma, worked on Musical.ly, the app that would become TikTok, helped build a security firm and a customer-support firm, and sat between Reid Hoffman and Asheem Chandna, the firm’s veteran security and infrastructure investor. In 2016 she nearly left to co-found an AI language-learning company with Andrew Ng, the deep-learning pioneer. You could not ask for a better technical co-founder, she told me, and yet the offer only clarified that she was an investor after all. She had her first child, and a few months later, at 28, was made Greylock’s youngest-ever general partner.

Greylock had been good to her, and she credits her partners with “treating me like a partner long before I had the title or experience for it.” She was recruiting GPs before she was one, and once, very young, vetoed a senior hire. But in her latter years she had reached the edge of what she could do from inside someone else’s firm. She had views on whom it should hire, where it should place its bets, how the coming era should be played, and not yet the standing to act on them. “My DPI to Greylock,” she told me, using the venture term for the cash a fund has returned to its investors, “did not command the amount of change I wanted to make. I’m pretty confident I’ll do well by them in the end, but at that moment in time, it was just very unclear.” In June 2022 she resigned.

By then, her parents’ company had been on its own ride. On December 14, 2017, 14 years after Jerry founded it, Casa Systems began trading on the Nasdaq under the ticker CASA. The implied market cap was $1.2 billion. The man who had arrived from Hunan with $50 had taken his company public at over a billion dollars—built, as Guo said, “with so much love by a bunch of immigrant engineers with very little sophistication about fundraising, marketing, sales, external support,” who had gotten there on the strength of “just solving a problem better.”

The trouble was that public markets demand other things, too. The stock peaked in 2018 and began a long slide. Casa’s second act, becoming a wireless-equipment company after years as a cable-equipment one, worked for a while, then didn’t. In April 2022 Verizon paid $40 million for almost 10% of the company in exchange for a 5G contract, a price that valued Casa at around $400 million, a third of what it had been worth the day it went public. In March 2023, Jerry stepped down as chief executive of the company he had founded 20 years earlier. That August the headquarters in Andover, where Guo had done her homework, sold for $6.4 million. A year later, it was over. Casa filed for Chapter 11.

**Two weeks before** Embed’s opening weekend, Huang had been in the office. It was Friday, March 6. He had driven up from Santa Clara to hand-deliver Nvidia’s first DGX Station to Karpathy at the Palo Alto office Conviction and Karpathy shared. The DGX was a matte black tower. It looked like the desktop computers people used to buy 20 years ago, except this one cost around $100,000 and was effectively a data center in a box. Across the front, in capitals and gold marker, Huang had written:

TO ANDREJ – THE 1ST DGX STATION. THE AGENTIC ERA OF AI HAS ARRIVED.

Two of Huang’s people came along, the founders of a small company called Brev that Nvidia had recently acquired. Also in the office that afternoon were John Hegeman, until recently Meta’s chief revenue officer, and Ravi Gupta, formerly a partner at Sequoia—both now co-founders of a company called Ithaca. Reddy and Garcia-Camargo were around. Someone suggested drinks. The group walked down to Local Union 271.

Huang stayed five hours, talking about the future, giving advice, and never looked at his phone once. Guo remembers looking at hers three or four times. “Jensen has this wisdom,” she told me, “that if you know what your priorities are and you’re working on the right one, you have all the time in the world. That’s something I aspire to.”

Nat Friedman once told her to worry less about being so commercial. “I am very commercial,” she responded. I asked what the intensity had cost her. “The hardest thing is making peace with your own choices,” she said. “Venture is a thing you can pour your whole soul into. I think the answer is, I try to prioritize what I think matters to my kids and then create this environment where they feel like I’m around if they need me.”

That she had walked the Met Gala carpet in 45 pounds of chainmail, with her mother beside her, was in keeping. Casa Systems had taught the girl who was worried about not being smart enough that being smart was not enough, and that working every day was not enough either. Companies died anyway. “The existential fear that there are people trying to come kill our companies is part of the way we do venture,” she said. “There’s commitment and loyalty to the team. It’s us against the world.”

I asked how she thought about the whole arc of her parents’ business, the company built over 20 years from $50 and taken public and then handed to a new CEO and then filed into bankruptcy. She had expected it to crush her parents. It hadn’t. “I find that very reassuring,” she said.

“I’ve always told my team, from day zero,” she said. “I want to make this work, but I’m trying to build a partnership. And I don’t want somebody to drag me out of here in a cart. I want the firm to be more interesting and better than I am, because it’s not all of my identity.”

When Guo left Greylock in June 2022 she had planned to take time off, she told me, to spend time with her family. She got 20 hours into helicopter flying lessons before Conviction’s first fund picked up too much. “As soon as I started talking to LPs,” she said, “I could tell it was happening.” She hasn’t flown since. “Someday I’m going to get back to it, it’ll just have to be after the revolution.”

Grady remembers LPs being somewhat unconvinced by Guo’s pitch for Conviction.

“Every now and then she’d tell me what she was hearing from them,” he said. “The main LP thing was, ‘What’s your strategy? How are you going to win?’ There’s the Patton quote, right? ‘A good plan violently executed now is better than a perfect plan next week.’ And that was her answer. ‘I don’t have a strategy, I’m just going to go win.’ If you’re an LP, that’s not a good pitch. But the reality is there are only so many people in our business who can just go execute, and she’s one of those.”

Not everyone bought the no-plan story. The fund she’d really wanted to build, some said, was a crypto fund, and the AI thesis was a retrofit. The evidence was not nothing. By the start of 2022, six months before she left, her Twitter avatar changed to a CryptoPunk-style zombie. Around the same time she helped a younger partner lead Greylock’s $70 million Series B into 0x Labs, and shared his board seat. In February she and two Greylock colleagues launched a crypto podcast, *Fungible Times*. Three days after announcing her departure she told TechCrunch the market was at the start of a new tech cycle, with companies coming across machine learning, AI, data and crypto.

By October, when Conviction launched, it was a $100 million fund, branded as “an investing firm for the AI age.” The crypto angle was gone. Five weeks later, FTX went bankrupt. Three weeks after that, ChatGPT shipped.

“It’s certainly a recurrent rumor,” Guo said. “But I was never going to do a crypto fund.” She had helped run Greylock’s crypto effort and gone as deep in it as she had in AI, the two most interesting ideas in technology in a decade. She is still a personal investor. She still thinks it’s real. What she hadn’t decided, that spring and summer, was whether the firm would be a generalist fund or a single bet. She chose AI because “crypto is fundamentally a financial technology,” she said. “AI is a general technology.” A financial technology, however profound, stays bounded. A general one starts tiny and then, all at once, is everywhere. She wanted to build the next great venture firm, and that lined up with the technology that would be everywhere.

That Guo might have left some people sore is perhaps not surprising. There are wounds still open at Greylock. She’d been there nine years and, near the end, had grown certain the market was changing faster than the firm would. She kept her board seats. She left as well as she knew how. Not all of her former partners have come around. Among them is Asheem Chandna, the senior security and infrastructure investor who had been her mentor.

“I’m still very grateful to Asheem,” she said. “I know he’s not pleased I left. I wish I could fix it.”

She has since thought about what it must have looked like from his side. “If one of the people I’d invested in wanted to start from scratch after nine years,” she said, “I’d have to find the space in my heart to think about what’s right for them.”

Guo’s first hire came two weeks after she launched Conviction. Reddy had gone to Penn, like Guo, and like Guo had grown up in the room while an immigrant father built a networking company that ran into the dot-com crash and was sold. He had been the best high-school debater in the country. After college he went to Neeva, the search startup, and was running a third of its engineering when his college roommate, Hunter Lightman, who worked at OpenAI, tried to recruit him over. Reddy said no. OpenAI had 130 people, and a company that size was more than he could imagine working at. It was valued, then, at $29 billion.

He went to Conviction instead, which at that point was two people, him and Guo. He disputes the idea that backing AI in 2022 made the firm especially visionary. “I think we get some credit, but not that much credit,” he told me. “It was 2022. Other people were earlier than we were. If we win, it’ll be because we were right about the people, not just the thesis.”

One of Guo’s bigger fears going into Conviction was access. Not just whether a $100 million fund could buy into the companies worth being in, but whether, with Greylock’s name no longer behind her, her calls still got returned. They did. The regret ran the other way. She and Reddy got into the companies but sized the positions, in her word, wimpily.

The two companies Conviction did not get into were Anthropic and OpenAI. The fund was too small, and they were too big. Yet Guo had backed Baseten in 2019, two years before Anthropic existed, when its founders were still worker ants at OpenAI. The labs grew up around her at Greylock, and she, like most investors, did not catch them. By the spring of 2023, it was too late. OpenAI was taking $10 billion checks from Microsoft and Anthropic’s Series C valued it at over $4 billion.

Guo has come to see the outside as the better place to stand. An investor holding an enormous position in one lab starts reading the whole ecosystem through it, and in a bubble the danger compounds, because people who make money begin to mistake their returns for intelligence. “It’s a fine line,” she said, “between conviction and being dishonest with yourself.” Her distance from the labs is not total. Conviction backed Mistral, Europe’s leading lab, and Thinking Machines, the lab Mira Murati built after leaving OpenAI. What it has never owned is the front of the race. Whether that began as a choice or a consequence hardly matters now; the money is placed, and the revolution is still young. Guo does not agonize over it. “If you ask me, will I take Anthropic stock at the market price? Absolutely I will,” she said. “But it’s not my life goal.”

She had wanted a peer from the day she started Conviction. As soon as the early bets looked like they might work, and when she saw Vernal was free, she turned it up, in her phrase, to 11. Vernal was not looking for a job. He was 43 and had decided he was finished. He had left Sequoia in 2023, kept his board seats, and spent the next two years picking his kids up from school at 3pm and playing baseball with them. On the side he wrote a dozen or so angel checks into companies like Decagon, Cursor, Cognition, Sunday Robotics, and others that, in his words, “look good on paper.” He called the arrangement idyllic. “Those two years were probably the best years of my life,” he told me.

He had been a partner at Sequoia since 2016, behind the firm’s investments in Statsig, Rippling, Clay, and Notion. Before that he spent eight and a half years at the previous era’s dominant business, Facebook, arriving in 2008 when it was a couple hundred people and leaving as one of its most senior leaders running product and engineering, reporting to Zuckerberg. In 2012 *Wired* ran his photograph under the headline “How Facebook’s Top Engineer Is Trying to Read Your Mind.”

Guo knew him a little, socially, from Sequoia events she went to with Grady. A month or two into the idyll she took him for coffee. “Very kindly and very generously,” Vernal said, “Sarah was like, ‘Okay, when are you joining?'” Over the next year and a half she drew him in. First: “I’ll give you keycard access to the office. You should just come work out of here when you’re in SF.” Then: “Do you want to come look at this company with me? I’d really value your opinion.” He came in a day a week, then mentored a class of Embed, then led a deal, all before he’d signed any paperwork. He became a general partner in January 2025.

He does not talk himself up. “I’m a much better number two,” he told me. “I don’t want to be the main person. We are building the firm as an equal partnership, but my motivation is to make Sarah and Conviction as successful as possible.”

“You’ll never hear Mike say anything lofty about himself,” said Vijaye Raji, who worked under him at Facebook and is now chief technology officer of applications at OpenAI. Bret Taylor, who was at Facebook with him and now runs Sierra, calls him an insane engineer. Parker Conrad tried to hire him to run product at Rippling. Garcia-Camargo, whom Conviction recruited out of Bridgewater the year Vernal arrived, keeps a running list of facts about him: He took the LSAT one weekend, cold, to see how he’d do, and scored in the mid-170s. He did his taxes in Lisp for six years. His favorite subject at a party is Ubiquiti, the maker of networking gear. He set up a data center in his own house, and once described the topology of its switches and racks in such detail, in a cold email to a founder, that the founder later said it was why he took Sequoia’s money.

Every insurgency has a general the world can see and one it cannot. Vernal is the second kind.

When I met Vernal, he was wearing a black polo and green On running shoes. The shoes matched the green dial and green strap of his Patek Philippe Nautilus. He had a line of Huang’s stuck in his head. He was annoyed by how good it was. “Man, everything is going right for this person,” he told me, smiling.

Huang had said the addressable market for tech companies used to be the software budget inside other companies. Now, with both large language models and robotics, the addressable market was essentially all of OpEx and all of CapEx. “A lot of doing investing well right now,” Vernal said, “is trying to imagine how big something can get. The size of outcomes here is just so much larger than people realize.”

The example he reached for was OpenEvidence, whose co-founder and chief technology officer, Zachary Ziegler, he interviewed on Saturday morning in front of the Embed founders at their York Street office.

OpenEvidence is a chatbot for doctors. A physician asks a clinical question, or enters the facts of the patient across the desk, and it answers from the medical literature, the guidelines, the drug labels, 35 million papers in all, with citations attached so the doctor can check the work. Which is to say it does, for doctors, what Claude and ChatGPT already do for everyone, search and answer, and on paper that should be fatal. Yet, it is the fastest-growing application for physicians in history. In April the company said about 65% of American doctors were using it. Aside from the iPhone, nothing in the history of medicine had reached them faster. More than 100 million Americans were treated last year by a doctor who had consulted OpenEvidence.

The company is barely five years old, founded in late 2021 by Ziegler and Daniel Nadler. Nadler had founded an AI business called Kensho previously. In 2018, he sold it to S&P Global for $550 million. At the time, it was the largest sale of an AI company there had ever been.

Guo and Vernal took Nadler to lunch in the first half of 2025. The Series B was coming together at $3.5 billion against revenue in the single-digit millions, a multiple of something like 1,000. While Nadler talked, Vernal ran the arithmetic he is known for around the firm. Ask him about a women’s textile company in India and he will study the ceiling for a moment and give you, within 10%, the revenue, the market cap, the EBITDA, the gross margin. He has done it since Sequoia. Now he was doing it on OpenEvidence.

Healthcare is about 20% of American GDP, he thought. Pharma is about four. What pharma spends advertising to doctors, on the back of an envelope, comes to something like 20 basis points of GDP, which is an enormous number, and what OpenEvidence captured was the doctor at the precise moment the patient was describing the symptoms.

“That is literally the highest-value inventory I could possibly imagine,” Vernal told me. “It’s similar to Google Search.” The only public comparison was Doximity, worth $10 to $15 billion, and for the math to work Vernal had to believe OpenEvidence would run well past it. “My reaction,” he said, “having worked on ads at Facebook for a very long time, was, oh, this can be absolutely ginormous.”

Guo and Vernal didn’t need a second meeting. Inside a week they had committed, at $3.5 billion—only the second time they’d invested above a billion dollars. Six months later OpenEvidence raised again at $12 billion, one of the fastest companies ever to reach $100 million in annual revenue.

Guo began to absorb what she would later identify as the central fact of company-building, which was that the company always felt like it could die at any moment, that the work was conducted as if by a small band of pirates against Goliath, and that the only available advantages were speed and product and the focus that comes from being always in wartime.

When Vernal introduced Ziegler to the Embed class, he told them OpenEvidence was one example of a $100 million run-rate business that venture had previously given up on. Harvey and Sierra were two others. “I think health tech still sucks, man,” Guo told me. “OpenEvidence is a very specific type of company started by a really special person.”

“People keep talking about an explosion in entrepreneurship,” she said. “And I’d say there is a huge amount of opportunity that is absolutely growing. And yet the number of people prepared to really start exceptional companies is not exponential.”

**Later on that Saturday** at Guo’s house down the Peninsula, as the sun fell, the tacos came out. They were served on the chessboard. The kings and queens and pawns were lifted aside, and the squares between the drive and the pool turned into a buffet line. By then Guo’s garden held more than the Embed founders. People had come from Thinking Machines and Cognition and Paradigm, senior people, and they moved among the 18- and 20-year-olds with plates in their hands, Karpathy and Vernal and Garcia-Camargo and Reddy somewhere in the crowd. The temperature dropped. Guo came out of the house in a purple knit sweater, rows of small white sheep marching across it, each row facing the opposite way to the one above. One sheep, over her right collarbone, was black.

She bounced through the garden, talking hires, M&A, schooling, films, silicon. The slang came with her, “sick,” “my guy,” and then, with no gear change at all, the whole banker’s alphabet, dispensed to the twenty-somethings around her. I left her to it a little after 10pm.

The next morning the team was back at York Street early. Guo and Vernal sat in front of the founders talking about how to raise money. Guo wore a red baseball cap and a t-shirt that read “Please don’t come to Mars.” Andrew Milich, who ran product engineering at Cursor, had been on the schedule to speak. He did not come. That week he had left for Musk’s xAI. A month later Musk announced he had bought an option to acquire Cursor outright for $60 billion.

I returned to Conviction at the end of May. It was another clean, bright San Francisco day, and as a Waymo carried me toward the office the news came through, on X first, that Andrej Karpathy had joined Anthropic. He was now a worker ant.

Guo had told me once that she knew her father was stressed when she came down in the morning and found the furniture rearranged. I stepped into the office and everything was where I remembered it. The pull-up bar, the bowl of chalk beneath it, the robot dog folded under the whiteboard. Garcia-Camargo was already at her desk. Vernal arrived five minutes after me, Guo five minutes after him. On her desk was a box with a gift inside. It was a black Harvey babygrow. They immediately began talking about “Andrej.” Guo’s phone had been going off.

In the month that followed, two more went. Noam Shazeer, who co-led Google’s Gemini and had co-written the 2017 paper that started all of it, left for OpenAI. Google had paid $2.7 billion to bring him back from a startup barely two years before. He left anyway. John Jumper, who had won a Nobel Prize for teaching a machine to predict the shapes of proteins, left Google’s DeepMind for Anthropic. Both men were leaving Google’s own labs, which had a frontier model and every advantage money buys, for the labs they judged to be ahead of it. By then SpaceX had exercised its option on Cursor. The trillion-dollar firm that owned xAI now also owned the coding agent that, until very recently, had been one of the few that worked.

Anthropic, OpenAI, xAI. They were the same names money had been pouring into since January, and now the people were going too, everything bending toward two or three names as if the whole field had begun to fall inward and pick up speed. At least the labs, for now, still need humans.

Even inside Conviction the mood had crept in. The question of what was left for early-stage investors, when the labs had so much money, compute, and talent, had become a debate inside the firm. For a stretch the honest answer seemed to be nothing. Guo took the other side. The labs would be enormous, she agreed. But it was not the labs that had built Harvey or Sierra or OpenEvidence. Someone still had to wedge the agents into the doctor’s office, the court filing, the customer-service queue, the accounting close, or the manufacturing floor. Her mother had told her years ago: “Don’t write code, Sarah. Product was not the hard part. People and sales management is the hard part.”

No one can tell you whether any of this ends in extinction or a golden age. The labs keep pulling people and companies out of Conviction’s world, and its world keeps growing anyway. “We have two handfuls of companies that are really working,” she told me. In the eight weeks between my two visits, five of Conviction’s companies raised more money. They are hungrier for cash than the software businesses of the last era ever were, and the private markets can only carry that so far.

“Everything to this point just feels like the prologue to me,” Reddy said. “We haven’t had the big winning company IPO yet, the thing that’s a marker.” When the first startup rings the bell it will be that moment, the proof that Guo and her team are winning their wager. It will also be the day her companies pass out of her hands to investors who have never met her founders.

“As somebody who is the daughter of entrepreneurs who took a company public, I loathe the idea,” she said. “And yet we will do it.”

**Dom Cooke** is the managing editor of Colossus.

[Presented by](https://rogo.ai/)