Save As

Iconic 'whiz kids' in tech

Think about clicking save.

You’re picturing a floppy disk, right? Except that is an image of the 3½ inch diskette, which was rigid plastic and a major improvement over the older 5¼ inch disk that actually flopped.

Not like it matters. You click the picture of the little plastic square to keep your work because in computing, icons exist to teach you how to interact with the mechanism underneath. They are heuristics. They are proxies.

Here is another one. Close your eyes for a sec and picture a “Founder.”

I know what I see. A guy. He’s young. In a hoodie. You know who I’m talking about: it’s Mark Zuckerberg.

Correction: it’s Jesse Eisenberg playing Aaron Sorkin’s interpretation of Ben Mezrich’s version of Mark Zuckerberg.

It sticks because he’s the archetype. The nerdy, perfect-SAT-score, probably sneaky good at chess, whiz-kid founder that the tech industry adores. The geeky dropout who leaves an elite school to build a category-defining tech company.

Last week I saw an a16z partner post on LinkedIn: “AI will make teenage billionaires a thing.” Software will be built by the time-rich, for the time-rich. The future belongs to young people, and billions will be made as a result. In the comments, he mentioned this was the thinking behind a new grant program his firm runs.

It seems like this archetype is hotter than ever. The Thiel Fellowship just doubled its grant to $200,000. Palantir is paying high schoolers $5,400 a month to skip college. The NYT ran a piece about a “youthquake” in San Francisco. The precocious whiz kid exists for a reason. It also came from somewhere. And where it came from might tell us something about where it’s going.

But I keep thinking about the save icon. The floppy disk meant something once. Then the thing it pointed at changed, and changed again, and the icon stayed the same because the icon still worked. Nobody had any reason to update it. Facebook was founded twenty years ago. What if the founder in the hoodie is the same kind of icon?

When a proxy works, nobody notices it is a proxy. Proxy decay is not an instant process. It happens slowly. When it finally breaks, the people closest to it are the last to see.

Five Gates

Bill Gates left Harvard in 1975 to go build Microsoft. That was the original version of this story. Thirty years later, Zuckerberg ran it back from the same campus.

The ‘Zuck’ archetype was not about the hoodie or the dropout transcript. It was a proxy for something highly specific: this person can ship software.

Given that many of us are so steeped in the “I can ship anything because of AI” moment, let’s take a sec to place ourselves in history. Thefacebook launched exactly sixteen days before the iPod Mini hit store shelves.

Knowing how to build and ship software was exceedingly rare compared to today. It was hard. The computer science departments at Stanford, Harvard, and MIT were literally the only places where people had the tools, the T1 connection, and the knowledge to teach others.

‘Zuck’ represented a replicable ability to do something most people could not. Youth was not the variable. Youth was a byproduct of four gates. Five including Bill.

The first gate was compute. Bill Gates needed Harvard because a DEC PDP-10 cost $1.7 million. One in a million had comparable access. The machine room was the gate.

The second gate was the network. Zuckerberg had an address, not an idea. Harvard’s directory was a verified map of a closed population. His dorm had a T1 line. His parents’ house had dial-up. The campus connection was thirty times faster. Enrollment was the gate.

The third gate was capital. Sequoia, Kleiner Perkins, and Accel within a ten-minute drive of each other on Sand Hill Road. No AngelList, no cold email, no DMs. The warm introduction was the entire pipeline, and it started at elite campuses. Stanford was three miles away.

The fourth gate was the building itself. The thing these founders were making could only be made from inside the institution. Leaving meant the thing had outgrown the room. It was not a signal. It was a consequence. The dropout meant something because it cost something, and it cost something because the alternative was real. Gates didn’t leave Harvard to signal conviction. He left because he was building the thing and staying was physically incompatible with that. Zuckerberg didn’t drop out to impress VCs. Thefacebook was doubling and he had to go run it.

Four access gates. All transient.

The runway outlasted the airstrip

Each gate opened in sequence. The proxy didn’t update.

Compute became a URL. AWS, 2006. The machine room stopped being a place. What used to require a university building became something you could rent with a credit card.

The network got mapped. The first time someone maps a real-world social graph into a digital network, it’s discovery. Every time after, it’s a copy. Yik Yak peaks and sells for nothing. The vacuum doesn’t reappear. It gets replaced by a new one nobody predicted.

Capital decentralized. YC moved the first check to a batch application in 2005. AngelList put deal flow online in 2010. A thousand accelerators, rolling funds, and scout programs followed. Sand Hill Road is still prestigious. No longer the only road. No longer even the fastest road.

Leaving stopped meaning anything. The dropout was never a choice. It was an inevitability: the thing outgrew the room and you had to go run it. When the tools are available at the same URL from your dorm room, your parents’ house, or a coffee shop in Omaha, dropping out doesn’t prove you have something. It proves you left. Dropping out of Stanford in 2004 was a consequence. Dropping out in 2026 is a LinkedIn post.

The industry watched all four gates open and kept encoding the proxy. Youth. The hoodie. The dorm room.

Clinkle: “He sells... a 20-year-old, white male, Stanford CS major. He appears to be the next Mark Zuckerberg.” $30 million. Never shipped a product.

Palantir’s Meritocracy Fellowship pays high schoolers $5,400 a month to skip college. A program that rejects elite institutions, using elite institution entrance metrics as its filter. The application requires a 1460 SAT.

Like a cargo cult: the islanders on Tanna built bamboo control towers and carved coconut headsets because the runway correlated with the cargo. The pattern was real. The error was mistaking the correlate for the cause.

She bought the turtleneck first

By this point the proxy had become legible enough to reverse-engineer. You could list the ingredients and assemble them. That’s what Holmes and SBF did.

Holmes tracked down the exact $270 Issey Miyake turtleneck from Steve Jobs’s keynotes. Apparently deepened her voice to a baritone. An employee reported she forgot to put on the baritone and slipped into her actual voice. Which seems like a lot of effort for a proxy. Kissinger, Shultz, and Mattis on the board. The machine demonstrated to Safeway made noise and did nothing. Safeway invested $350 million retrofitting 800 stores. $9 billion.

SBF refused to cut his hair. “I think it’s important for people to think I look crazy.” Beanbag in a $40 million penthouse. Ethics were “mostly a front.” Sequoia published a 13,000-word profile. Invested $214 million. Forty-eight days later, marked it to zero.

The proxy was performing so well that nobody needed to look underneath.

Fountains of youth

The proxy pointed at “this person can ship software.” Now everyone can ship software.

AI didn’t recreate closed social graphs. It didn’t recreate gated capital networks. It made the substance available to everyone, at the same URL, for the same price. The last gate didn’t just open. It dissolved.

Last year a single product designer shipped a scheduling app in a weekend that handled waitlists, payment processing, and automated reminders. No backend engineer. No DevOps hire. No sprint planning meeting. She described the process as “talking to the computer until it worked.” In 2019, that was a three-person team and four months of runway. The gate didn’t crack open. It evaporated.

The Thiel Fellowship doubled its grant to $200,000. YC’s youngest applicants surged 110% in two years. Palantir is paying high schoolers $5,400 a month to skip college and requires a 1460 SAT to apply. Andreessen partners are writing essays about why this time is different. The NYT is calling it a “youthquake.” The thing that youth was proxying for is now free, and the proxy is louder than it has ever been.

In fast-moving technical frontiers, deep fluency can genuinely skew young. A 24-year-old with five years of AI exposure sometimes has more relevant context than a veteran in a legacy domain. That’s real. But it’s not what the proxy is selecting for. The proxy is selecting for the costume.

Some of these founders will build real companies. Cursor shipped from MIT. Mercor has actual revenue.

Building got easy. Knowing what to build did not.

In 2004, leaving meant the thing had outgrown the room. In 2026, leaving means you clicked the save icon. There is no disk inside. It is all in the cloud.

Selling picks & shovels

In 1848, Samuel Brannan heard about gold at Sutter’s Mill and made a decision that sounds obvious in hindsight: he didn’t go looking for metal, he went shopping. He bought up pans, picks, shovels, boots, denim. Literally everything a prospector would need to turn hope into labor. Then walked the streets holding a vial of gold dust, shouting about gold from the American River. He never panned a day in his life. He owned the supply store.

The modern supply store is compute, model serving, inference, and developer tools. Andreessen Horowitz runs Speedrun grants. Founders Inc. hosts summer programs investing $100,000 to $250,000 in high schoolers. The founder raises a seed round and spends a nontrivial portion on inference, tooling, and distribution that routes through infrastructure the same fund’s portfolio companies already own.

Every funded AI startup becomes a customer of that stack. When a GP at a $90 billion fund writes in lowercase about teenage billionaires, he is not predicting the rush. He is selling pans.

Ship it

Holmes needed the machine to seem like it worked. SBF needed the balance sheet to seem like it added up. Both reverse-engineered the proxy, and both got caught, because both were still making a claim about reality. The turtleneck said “I built something revolutionary.” The beanbag said “I’m too busy building to care about furniture.” Fake claims, but claims nonetheless. The performance only worked as long as the audience believed something real was underneath. When it turned out nothing was, they went to prison.

Roy Lee understood something Holmes and SBF didn’t. You don’t need the underneath.

He got into Harvard. Harvard rescinded his acceptance. He never set foot on campus. He enrolled at Columbia. Columbia suspended him for building an AI tool that cheats on coding interviews. (The thing it helps you cheat at is itself becoming irrelevant because of AI. Layers of obsolescence.) He co-founded Cluely in 2025. Harper’s sent a reporter to do a profile. Someone had left copies of Chaucer and Boccaccio on a shelf in the office. They were unread. The reporter suggested Roy might find something valuable in the books. “Well,” said Roy, “I don’t care about the Decanterbury Tales.”

The missteps continued. The reporter tried using Cluely during the interview. It got stuck in a recursive loop, suggesting he narrate the process of using the product he was already using. He told Roy this didn’t seem useful. Roy, genuinely confused: “I mean, what would you have wanted it to say?” Then Cluely crashed. Roy stormed downstairs. “Cluely’s not working!” His team spent fifteen minutes getting it back online. They resumed. It crashed again.

I enjoyed Sam’s piece. The art of the true long-form profile feels like it’s dying. But I think he might have missed the point. The product was working perfectly.

Close only counts in horseshoes

I thought the pattern was a line. Gates, Zuckerberg, Holmes, SBF, Roy Lee. Substance to signal to costume to air. A clean story about decay.

It’s not a line. It’s a horseshoe.

Roy Lee took the proxy so far past the original that it curved back into something that works. Not works like “the code compiles.” Works like “the name gets said again, in rooms he was never in, by people who will never open the app.” Persona is the product. Controversy is distribution. The software crashes on camera and the coverage does more than hold. It actively improves. Every malfunction is a clip.

Every clip gets shared. The more it breaks, the more it spreads, and the spread is the entire point. He does not need the machine to work. He needs the machine to be seen. A machine failing on camera is more engaging than a machine working quietly in someone’s pocket. Brannan didn’t need gold in the river. He needed people to show up at the store.

The horseshoe closes when the distance between proxy and substance gets wide enough that the audience stops checking underneath. Not because they can’t. Because checking was never the transaction. The transaction is the story. A narrative that moves fast enough doesn’t get verified. It gets forwarded.

Roy Lee almost certainly knows the word is Canterbury. He probably knows those are two different authors. That information is dead weight.

‘Decanterbury Tales’ is load-bearing. It’s the pull quote. The screenshot. The line you read aloud to someone across the room. He didn’t need the Harper’s reporter to be on his side. He needed the reporter to be good at his job. A skilled long-form profiler finds the most revealing moment and puts it in the piece.

The reporter did exactly that. Long-form profiles like that one are a dying art. The clip is the life support. Roy understood which side of that equation he was standing on.

This is the part worth sitting with: nobody was tricked. Everyone in the chain did their job with real craft. The reporter wrote a careful, detailed profile. The editors ran it. Readers pulled the best lines and shared them. The proxy propagated. The system doesn’t produce this outcome in spite of everyone doing good work. It produces this outcome because everyone does good work. The incentives are aligned. The results are as designed. You just might not love what they’re designing.

The save icon on your toolbar has never touched a floppy disk. It points at something that left the building decades ago, and it held its shape because the shape still works. Nobody has any reason to replace it, because the people who click it every day are not asking what it once referred to.

They just need something to click.