Garry Tan on Own Your Intelligence

1. Key Themes

Personal AGI as Owned Infrastructure, Not Rented Product

Garry draws a sharp distinction between corporate AI subscriptions and a personally owned intelligence layer. The former resets when you close the tab and gets "lobotomized on someone else's schedule." The latter compounds daily.

"The corporate AGI you don't own gets better only when the company ships something. Your personal AGI gets better every single day you use it because every day it knows more of your life. One of these is a product you consume. The other is an asset you build." 00:06:54

The 400x Productivity Multiplier Is Real—Even With Pathological Discounting

Garry does not just assert productivity gains; he stress-tests his own number with maximum skepticism and still lands at a floor of 8x.

"I did the math on my output and I'm at about 400x what I did in 2013. Apply the most pathological verbosity penalty you can stomach... assume I'm flattering myself... 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." 00:08:12

Context Is the New Moat—Not Model Weights

With frontier models becoming commodities, the durable competitive advantage shifts entirely to what unique context you feed the model.

"You're going to have a frontier model, which is rented and a commodity and getting cheaper by the quarter. Plus your context, which is owned by you and unique. And ideally, nobody else on this earth has it... Model quality is rented, but your brain is owned ideally by you." 00:10:58

Skill Files as Externalized Cognition—and a Property Rights Crisis

Garry reframes skill files not as productivity tools but as intellectual property whose ownership determines your economic fate. The same file, in your repo versus your employer's repo, produces radically different futures.

"A skill file is not a document. It's a piece of your cognition. How you do the thing. Extracted from your head, written down and executable. Every skill you teach an agent is you, externalized... Same files. Same Maya. One variable." 00:29:01

Markdown Is Code—The Compiler Is the Language Model

This reframes who can build software and who counts as a programmer, with direct implications for workforce transformation.

"If you can write clear instructions in English, you're a programmer. The compiler is a language model. And that's why it's not just for engineers anymore. At YC, our media people, event staff, finance team, people who never opened a terminal in their lives, are building skill files and scheduled jobs." 00:17:48

Revenue Per Employee Has Broken Historical Precedent

YC portfolio data shows revenue-per-headcount ratios that have no historical analog in any industry—software, oil, or railroads.

"Emergent out of our summer 24 batch went from public launch to nine figures of revenue in eight months. When they crossed $15 million in annualized revenue, they were 15 people. Retool winter 24 hit 60 million annualized with about 40. That revenue per person did not exist before. Not in software, not in oil, not in railroads." 00:22:08

The Compounding Flywheel: Never Do One-Off Work

The discipline that separates compounders from dabblers is converting every task into a reusable skill file, creating permanent institutional memory.

"Never do one off work... At the end of every task, ask the agent to skillify what it did... If you have to ask for something twice, you failed. The person who captures what they learn gets smarter every single day." 00:27:04

The Library Plus the Librarian: Working Memory as the Core Constraint

The architectural insight is that the bottleneck is not model intelligence but context selection—deciding which three books out of the library are open at any given moment.

"Three Harry Potter books versus seven digits. You could argue that's not quite AGI yet, but it is already a different operating regime. And almost everyone on Earth is still running their life on an org chart and a way of doing things designed for the seven digit brain." 00:13:37

Privacy Is Won by Taking Custody, Not by Avoiding Consolidation

Garry reframes the privacy objection entirely: your life is already scattered across ten clouds with misaligned incentives. Consolidating into your own infrastructure is the security model, not the risk.

"My brain runs on my own infra, in my own repo, under my own keys. Compare that to the default, which is not privacy. The default is your life is already scattered across ten clouds owned by companies whose incentives are not yours, searchable by everyone except you. I didn't create the risk by consolidating my context. I took custody of it." 00:34:45

Open Sourcing Tools of Leverage Is a Renaissance vs. Priesthood Decision

Garry frames the choice to open source his entire personal operating system as an ideological stance about who gets to participate in the new economy of intelligence.

"Every era has a private technology of leverage, a thing the powerful have and everyone else doesn't... Right now, today, it's this... When something like that, that powerful stays private, you get a priesthood. When it gets given away, you get a renaissance." 00:36:03

2. Contrarian Perspectives

Better Models Make Your Personal Library Worth More, Not Less

Most people believe rapid model improvement makes personal infrastructure investment obsolete. Garry inverts this: each model release is a free upgrade to a library you already own.

"The better the models get, the more the differentiator moves to context. When everyone's engine is a thousand horsepower, the race is won on the driver and the map... A better model makes your library worth more because a smarter reader extracts more from the same books." 00:33:50

Retrieval Is Trivial—Being Worth Retrieving From Is the Hard Product

When people dismiss personal AGI as "just RAG," they are confusing the primitive with the product. The unsolved problem is curation, enrichment, and arbitration—not retrieval itself.

"Sure, and Postgres is just B-trees. Retrieval is the primitive, not the product. The hard part is everything around it. What gets written down in the first place, how it gets enriched and linked... Retrieval is easy. Being worth retrieving from is the product." 00:34:17

AGI Is Not a Singular Event—It Is Already Here, Diffused

While the industry waits for an AGI threshold announcement, Garry argues the functional equivalent is already operating as distributed personal agents, invisible because it doesn't look like what people expect.

"Everyone is watching the sky. And the thing they're watching for is already in the room. It doesn't look like a God. It looks like infrastructure, a terminal window, a folder of markdown files, a job that finishes while you sleep." 00:05:08

Comfortable Employment Is the New Thousand-Guilder Bribe to Stop Building

Garry provocatively reframes standard employment arrangements—where your skill files vest in someone else's repo—as a historical repeat of the offer Spinoza's community made to suppress him.

"Every comfortable arrangement where your judgment compounds in someone else's repo is a thousand guilders a year. To show up, keep quiet, and stop building your own thing." 00:31:28

The Loudest Critics Are Always the Fastest Adopters

Public dunking on AI-native workflows is not genuine skepticism—it is the leading indicator of adoption, not opposition.

"First they quote tweet you, then they git clone you. The dunks are just the adoption curve announcing itself." 00:38:00

3. Companies Identified

Y Combinator

Accelerator program. Mentioned as the institutional vantage point from which Garry observes portfolio-scale AI adoption, and as the organization already running non-engineers on skill files and agent frameworks.

"At YC, we get to watch this at portfolio scale. A year and a half ago in the winter 25 batch, a quarter of the companies had code bases that were 95% AI generated. Those companies use AI agents for everything now, not just code. And that batch is on track to becoming one of the fastest growing, most profitable batches in the history of YC." 00:09:11

Emergent (YC Summer 2024 batch)

AI-native startup. Cited as a breakout example of the new revenue-per-headcount physics.

"Emergent out of our summer 24 batch went from public launch to nine figures of revenue in eight months. When they crossed $15 million in annualized revenue, they were 15 people." 00:22:08

Retool (YC Winter 2024 batch)

Software company. Cited alongside Emergent as evidence that the new revenue-per-employee ratios are systemic, not anomalous.

"Retool winter 24 hit 60 million annualized with about 40. That revenue per person did not exist before. Not in software, not in oil, not in railroads." 00:22:08

G-Brain / G-Stack

Garry's open-source personal AGI infrastructure. G-Stack has 123,000 GitHub stars, placing it in the top 100 open source projects in GitHub history. G-Brain.io is the hosted free version.

"On top of this library sits my agentic coding framework, G-Stack, 123,000 stars now, which put it in the top 100 open source projects in the history of GitHub." 00:16:25

Circleback

Meeting transcription tool. Named as an integration point in Garry's actual skill file workflow for processing meeting recordings.

"When a meeting recording lands from Circleback, transcribe it with speaker labels. Pull out the commitment made, who made it, and the deadline." 00:16:52

Claude / Anthropic (Claude Code)

AI coding agent. Named as one of the recommended harness options for running personal agents.

"I use OpenClaw and Hermes agent with Gbrain... Codex, Claude Code, whatever. Any of them will do 99% of this." 00:24:24

OpenAI (Codex)

AI platform. Named as an alternative harness option alongside Claude Code.

"Codex, Claude Code, whatever. Any of them will do 99% of this." 00:24:24

4. People Identified

Baruch Spinoza

17th-century philosopher. Used as the structural metaphor of the entire talk—a person who built precision tools by day and world-changing intellectual work by night, with no permission, no institution, and owned everything he produced.

"His response was to build precision tools by day and write the most dangerous book in Europe by night alone, with no permission from anybody." 00:04:07

Vannevar Bush

American engineer and science administrator. Cited as an early visionary of the personal AGI concept through his idea of the memex.

"This was a dream of a great many people. Vannevar Bush called it the memex, a machine that would be an extension of yourself and your brain." 00:06:03

Marshall McLuhan

Media theorist. Cited to frame the concept of technology as cognitive extension.

"Marshall McLuhan said that technology is an extension of man." 00:11:25

Steve Jobs

Apple co-founder. Cited for the "bicycle for the mind" framing of personal computing, updated by Garry to "self-driving rocket."

"Steve Jobs called a computer a bicycle for the mind. And if you have what I'm describing here, then you have a self-driving rocket." 00:11:54

Paul Graham

YC co-founder. Cited for the two foundational startup principles that still govern everything, now amplified by agents.

"Paul Graham taught every founder in this building two things. Make something people want and do things that don't scale. Both still govern everything. What's new is the multiplier on the second one." 00:11:54

Andrej Karpathy

AI researcher (formerly Tesla, OpenAI). Referenced approvingly for his "knowledge wiki" approach, which Garry used as inspiration for G-Brain's architecture.

"My personal open claw has a Karpathy style knowledge wiki with about 220,000 markdown pages." 00:14:32

Gottfried Leibniz

17th-century mathematician and philosopher. Cited as an illustration of the pattern where the loudest public critic is privately the most obsessive adopter.

"He spends the next 40 years lying about it. Publicly, the visit was a few hours in passing. Privately, his notes are crammed with obsessive commentary on Spinoza." 00:37:32

Albert Einstein

Physicist. Cited at the opening to establish Spinoza's intellectual stature and the longevity of his influence.

"The most famous scientist alive, asked the biggest question there is, pointed at Spinoza." 00:01:01

5. Operating Insights

The "Skillify" Primitive Eliminates Amnesia Tax

The single most operationally powerful habit Garry describes is converting every completed task into a reusable skill file immediately after completion—never discarding context. This is not a workflow suggestion; it is a compounding mechanism that means every hour of work produces both the output and the procedure for reproducing it.

"At the end of every task, ask the agent to skillify what it did. Skillify is a special skill you can find in Gbrain. You can point it at that repo and say, extract skillify. Learn how to do it. Turn it into a markdown file you can use and reuse forever." 00:27:04

Latent Space vs. Deterministic Space: Knowing Which Computation Goes Where

Every agent failure Garry has witnessed traces to this single misdiagnosis. Taste and judgment belong in the model; arithmetic, SQL, and constraint-solving belong in code called by the markdown file. Mixing them produces confident errors no one can trace.

"Some computation belongs in latent space. Taste, judgment, reading what a human actually wants from a vague request. That lives in the model and you steer it with a markdown file. And then some computation belongs in deterministic space. The arithmetic, the SQL query... Being smart about this goes a long way." 00:18:19

The Morning Briefing Architecture: Process Inbox, Don't Sort It

Garry's agent does not filter email—it processes it with full context from his personal library, producing a briefing and a per-meeting prep doc. The operational standard is waking up to finished work, not to a pile of inputs.

"While I slept last night, my agent processed my inbox. Not sorted it, processed it... I wake up to a briefing, not a pile of emails. Before every meeting, a prep doc. Who I'm meeting, what we said last time, what changed since, and what I should ask." 00:15:28

Provenance and Contradiction Checks Are Non-Negotiable Infrastructure

A brain without hygiene is worse than no brain, because it surfaces stale facts with total confidence. The operational primitive is provenance on every fact and an active contradiction-flagging layer.

"A brain nobody curates is a garbage dump with great search. Retrieval will surface a stale fact with total confidence. A bad skill file encodes a bad process forever. Provenance on every fact. Contradiction checks when new information collides with old. And a librarian whose actual job is pruning." 00:23:28

The Smart Intern Test for Skill File Quality

Before deploying any skill file, apply this single evaluation: could a smart person on their first day follow these instructions exactly? If yes, the agent can run it. If not, the file is not yet a skill.

"It's a page of English. A smart intern, anyone really, who could read could follow it. And that's the test actually. If a smart intern could follow it, an agent can run it." 00:17:19

6. Overlooked Insights

A Father Built a Rare-Epilepsy Expert System in a Laptop—This Is the Template for Vertical Personal AGI in Healthcare

Garry mentions this in a single paragraph, but it is arguably the most consequential signal in the talk. A non-expert individual with no institutional affiliation built an 80,000-file expert system for one rare pediatric condition—enabling real-time evaluation of novel specialist suggestions against the complete known literature and full patient history. This is not a curiosity; it is the proof-of-concept for an entirely new category of ultra-personalized medical intelligence that no hospital, pharma company, or health AI startup has built, because they all optimize for populations, not individuals.

"He built a repo of 80,000 markdown files. A brain for one small boy. And pushed himself to the absolute edge of what humanity knows about his son's exact condition. Every specialist visit, every paper, every seizure log, every drug interaction. Indexed and cross-linked and ready. So that when a new doctor has an idea, he knows in minutes whether it's already been tried." 00:38:30

The investment implication: the first company to productize this pattern—personal AGI tuned to a single patient's complete longitudinal record, pointed at rare or complex disease—addresses a market where no good solution exists and willingness to pay is effectively unlimited.

The Non-Engineer Finance Employee Who Replaced 100 Excel Workbooks Is the Real Enterprise AI Story

Garry mentions almost in passing that a YC finance team member—with no coding background—used an internal agent to consolidate roughly 100 Excel workbooks into a single application. This is not a footnote. It signals that the real productivity unlock from agentic AI in enterprises is happening not in engineering teams (which everyone is watching) but in operations, finance, and administrative functions where process knowledge has never been formalized and where the ROI from consolidation is immediate and enormous.

"One of our finance folks compiled about 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." 00:17:48

The investment implication: the companies building tools that let non-technical operators create and manage skill files in enterprise back-office functions—not developer tools, not copilots for engineers—may capture a larger and faster TAM than the current developer-focused AI tooling market suggests.