Garry Tan: Own Your Intelligence

Why does it matter? Because your skill files in the company's repo aren't a career—they're an extraction.

Garry Tan frames this as a property rights question dressed up as a productivity talk. The skill files you build on the job—your judgment, encoded in Markdown—can now be stored, versioned, and owned by someone other than you. For the first time in history. Whether you understand this will determine whether you're building a career or donating one.

• If your skill files live in the company's repo, the company keeps running your judgment after you leave—your name won't be in the commit history • The gap between 2x and 100x AI users has nothing to do with which model they use • AGI isn't an announcement you're waiting for; it's already diffused through agents running on personal context • AI-native companies are hitting revenue-per-headcount ratios that have never existed in any industry, including oil and railroads

Your skill files in the company's repo aren't a career asset—they're an extraction

Same 40 skill files. One variable.

Maya in version one keeps them in her own repo. Changes jobs; the files go with her. Day one at the new company, she's operating with years of compounded judgment on tap.

Maya in version two builds the same files inside the company's IT infrastructure. "Maya leaves with nothing. The company keeps running her judgment without her." Forty files executing forever. Her name isn't even in the commit history. "She didn't have a career. She had an extraction."

This isn't an IP clause problem—it's a structural shift. For the first time in history, your judgment can be extracted, stored, versioned, and owned by someone other than you. Craftsmen owned their tools; that's what made them free. The factory broke that. Knowledge workers assumed they were safe because their tools lived in their heads. Skill files end that assumption.

The historical parallel Tan reaches for: every comfortable arrangement where your judgment compounds in someone else's repo is "a thousand guilders a year"—the exact bribe Spinoza was offered to show up at synagogue and stop building. The doctrine is simple: keep every skill file in a personal repo you control before any employer, platform, or acquirer has an opinion about it.

Personal AGI isn't the $20/month chatbot—it's an agent that compounds every day you live your life

The $20/month chatbot resets when you close the tab. It knows what everyone else already knows. When the company behind it pivots, "your so-called assistant gets a lobotomy on someone else's schedule."

Personal AGI is a different animal: an agent running on your own infrastructure, reading from memory you own, executing procedures you wrote. It gets better every single day because every day it knows more of your life. "One of these is a product you consume. The other is an asset you build."

Tan's own implementation: 220,000 Markdown pages spanning 25 years—diarized email, meeting notes, decisions, what he got wrong, compiled mostly by agents. A founder emails him about a crisis; before he finishes reading, the agent has already pulled every prior conversation with that founder, found three portfolio companies that hit the same wall, and surfaced what actually worked. "When my agent does anything, it does knowing everything I know. And that's the difference between an assistant and a colleague."

The underlying arithmetic: human working memory holds seven items. An AI agent holds a million tokens—roughly a thousand pages, three Harry Potter books open simultaneously. Almost everyone on earth is still running their professional life on systems designed for the seven-digit brain.

AGI isn't arriving as an event—it's already in the room, and it looks like a folder of Markdown files

AGI isn't an announcement waiting to happen—it's already present, diffused through agents running on personal context. "It doesn't look like a God. It looks like infrastructure, a terminal wall, a folder of markdown files, a job that finishes while you sleep, spread through everything."

Tan updates Spinoza's heresy for the 21st century: Spinoza said the divine isn't a king on a throne, it's spread through everything. The crowd watching the sky for a singular AGI threshold event is making the same mistake.

What personal AGI looks like in practice: five days before this talk, Tan decided it needed Spinoza. His agent acquired three biographies—Nadler, Goldstein, Stewart, roughly 1,500 pages—read all three, synthesized a dated chronology, flagged every point where the biographers disagreed with each other, and ranked the ten most tellable moments with delivery notes attached.

"AGI isn't arriving as an event. It's arriving diffused as your agent, running on your context, doing your work." Waiting for the announcement means handing years of compounding to people who understood this earlier.

Garry Tan clocked 400x his 2013 output—and even the most punishing discount still yields 8x minimum

He's at 400x. Same brain, same hours, plus a five o'clock kid pickup—Tan compared his 2025 output to his 2013 baseline, when he was shipping roughly 14 useful lines of code per day as a YC partner building Bookface at night, which he notes is dead on median in programmer productivity research. "I did the math on my output and I'm at about 400x what I did in 2013."

He deflates it before anyone else can: assume the agent writes bloated code, assume half is scaffolding, assume he's flattering himself. "Apply the most pathological verbosity penalty you can stomach... It's still 8x at the absolute floor and 10 times that in the middle of the range. The number is large no matter how you torture it."

The multiplier isn't only for engineers. Design, product management, growth—every piece of knowledge work moves by the same factor. A quarter of companies in the Winter 2025 YC batch had codebases that were 95% AI-generated, and those companies now use agents across every function. That batch is tracking to become one of the fastest-growing and most profitable in YC's history.

The caveat is blunt: "If you're not doing it, your competitor is, and they will eat your lunch politely and thank you for it."

The gap between 2x and 100x AI users isn't the model—it's whose context runs it

Two founders. Same Claude. Same weights, same context window, same API. One gets 2x leverage. The other gets 100x. The delta: "The leverage is not in the weights. It's in what context you give it, how relevant it is, and does it happen at the right step."

When every frontier model is a commodity getting cheaper by the quarter, the race shifts entirely to the driver and the map. "The weights are everyone's. The library is yours. At least I hope it is."

The direction is self-reinforcing: a better model makes your context library worth more, not less. A smarter reader extracts more from the same books. Every model release the labs ship is, in Tan's framing, a free upgrade to a workforce you already own—which means obsessive investment in curating your personal context library today is an asset that appreciates with every release, not depreciates.

The failure mode runs exactly the other way. "A brain nobody curates is a garbage dump with great search." Retrieval surfaces stale facts with total confidence. A bad skill file encodes a bad process forever. The primitive isn't just memory—it's memory plus hygiene: provenance on every fact, contradiction checks when new information collides with old, and a librarian whose actual job is pruning.

Markdown is code, and anyone who can write clear instructions in English is now a programmer

One page of English. A smart intern who can read it can follow it. An agent can execute it. That's the abstraction collapse in full.

"Markdown is actually code. If you can write clear instructions in English, you're a programmer. The compiler is a language model."

At YC, media staff, events coordinators, and finance people who have never opened a terminal are building skill files and scheduled jobs. One finance person compiled roughly 100 Excel workbooks into a single app she built with an internal agent. "She's not a programmer. She's a manager of agents now. Everyone is about to be."

A skill file is an employee with one job, written clearly enough that someone new could run it. A resolver is an org chart: a task comes in and routes to the right Markdown file. Before you have a co-founder, a logo, or a deck, you can already be running an organization—one founder plus agents, headcount set by whatever you decide to write down.

The quality bar Tan sets is blunt: "If you have to ask for something twice, you failed." Every exception, every edge case, every "oh, and also" belongs in the file. Then the page is an employee. Run it.

AI-native companies are hitting revenue-per-headcount ratios that have never existed in any industry in history

$15 million in annualized revenue. 15 people. Eight months from public launch to nine figures.

Emergent, from YC's Summer 2024 batch. Retell, Winter 2024, hit $60 million annualized with roughly 40. "That revenue per person did not exist before. Not in software, not in oil, not in railroads. These aren't freaks of nature. They're the first companies built natively on the new physics."

The hiring plan from last year is a legacy document. The headcount you assumed you needed was a workaround for a constraint—one person, seven things in working memory—that no longer holds the way it did. Software doesn't have to be precious anymore. Build the tool you need for an audience of one. Some of those tools-for-one turn out to be entire companies; you'll know because other people start begging to borrow them.

Old advice: scratch your own itch and hope it's the market. The updated version: scratch your own itch because scratching itches is nearly free.

When powerful technology stays private, you get a priesthood. When it gets given away, you get a renaissance.

"Every era has a private technology of leverage. A thing the powerful have and everyone else doesn't. For a long time it was literacy. Then it was capital. Right now, today, it's this."

The harness, the library, the workforce made of Markdown. The people who have it are operating at a different scale, quietly, and the gap is widening every month.

That's why Tan open-sourced G-Brain, the harness, the skills, the entire personal operating system—not as a growth strategy. "When something like that, that powerful stays private, you get a priesthood. When it gets given away, you get a renaissance. I know which one I want to live in."

The pattern to expect when building in the open: Leibniz spent three days with Spinoza in an attic and then spent 40 years publicly lying about it, while his notebooks showed obsessive commentary on Spinoza's ideas. The loudest critics are often just the adoption curve announcing itself. "First they quote-tweet you, then they git clone you."

The context library you're not building today is the lead time you're handing to people who started earlier

The compounding curve on context libraries looks like every other compounding curve: flat, flat, flat—then not. Most people haven't started. Every model improvement makes an existing library worth more, which means early builders collect a free upgrade with each release while late starters are still indexing last year's inbox.

This isn't ultimately a productivity argument. It's a compounding argument. The people who understand it now will look, in a decade, like they had access to something others didn't—and they did. They just started earlier.

Your moat is lying there, unindexed.

---

Topics: artificial intelligence, personal AGI, productivity, startup founding, knowledge work, AI agents, skill files, context windows, ownership, YC, Garry Tan, Spinoza, philosophy, future of work