Personal AGI is an agent that runs on your own infrastructure, reads from a memory you own, executes procedures you wrote, and compounds every day. It is an asset you build, not a product you consume — and it is the difference between renting intelligence and owning it.
The core lesson from Garry Tan: AGI isn't arriving as an event. It arrives diffused, as your agent running on your context, doing your work. For SMBs, that means the AI workforce you own — data, procedures, skills — matters more than any model you rent.
At YC Startup School 2026, Y Combinator's CEO opened with Spinoza — the philosopher excommunicated from his community for thinking differently, offered a thousand guilders a year to stop building, and who kept building anyway. Garry Tan uses that story to talk about what he calls conatus: the striving inside every living thing to increase its power to act. The forty-minute talk compresses into one line: personal AGI is how you stay under your own power in the age of agents.
For 5ac, this talk is not just news. Garry Tan named Hermes Agent in his list of typical harnesses and said plainly: "I use OpenClaw and Hermes Agent with GBrain." The architecture 5ac runs for customers — Hermes Agent plus GBrain RAG plus skills plus cron jobs plus kanban — is exactly the personal AGI the head of Y Combinator is describing. This article digs into why that matters for small businesses.
AGI is not an event — it is infrastructure diffusing
Garry Tan flips the common picture: everyone is waiting for AGI as a God in a data center, some announcement, some day the sky changes color. He says the thing they are waiting for is already in the room. It does not look like a God. It looks like infrastructure: a terminal window, a folder of markdown files, a job that finishes while you sleep.
"AGI isn't arriving as an event. It's arriving diffused as your agent running on your context doing your work." He calls it personal AGI: general intelligence for one person, not for everyone all at once.
For a small business, this distinction decides how you invest. If AGI is an event, you wait and rent. If AGI is infrastructure, you build. A Vietnamese SMB does not need to wait for a better model — the model is already good enough. What they lack is the private layer underneath: where the agent lives, the memory it reads, the procedures it runs. That layer is where durable advantage comes from.
Renting intelligence means accepting a lobotomy on someone else's schedule
Garry Tan draws a hard line. He is not talking about a $20-a-month chatbot, a slightly better autocomplete, or an assistant that knows your calendar and nothing else. That is a subscription you rent: corporate AGI you don't own. It resets when you close the tab. It knows what everyone else already knows. And when the company behind it pivots, your so-called assistant gets a lobotomy on someone else's schedule.
"Your personal AGI gets better every single day you use it because every day it knows more of your life." One is a product you consume. The other is an asset you build. This is the open-source, self-hosted argument 5ac makes: a Vietnamese business should not rent back the brain that runs its own operations.
| Dimension | Rented intelligence (corporate AGI) | Owned intelligence (personal AGI) |
|---|---|---|
| Where it runs | Vendor's infrastructure | Your infrastructure |
| Memory | Resets when you close the tab | Compounds every day |
| When the company pivots | Lobotomy on their schedule | Stays intact |
| Knowledge | What everyone else knows | What only you know |
| Nature | Product you consume | Asset you build |
His own numbers are striking. In 2013 he was a YC partner building Bookface at night, shipping about 14 useful lines of code a day — dead-on median for the industry. This year he runs YC full-time, same brain, same hours, plus a five o'clock kid pickup, and his output is roughly 400X. Even after the most pathological penalty you can stomach, the floor is 8X. And the multiplier is not just for code — it applies to design, product management, growth, every kind of knowledge work.
Leverage lives in context, not in weights
This may be the most important line for SMBs: "2X people and 100X people using the same Claude... the leverage is not in the weights. It's in what context you give it." Same model, same weights, same context window, same API. One person gets 2X, another gets 100X. The difference is context: relevance, accuracy, and timing.
Models are a commodity that gets cheaper every quarter. Context is the only thing nobody else has: customer data, internal procedures, decision history, lessons from mistakes. When Garry Tan says "2X people and 100X people," he is describing the gap between businesses that use AI as autocomplete and businesses that treat it as a workforce.
This is why 5ac builds open-source AI infrastructure around GBrain RAG. GBrain is not a chatbot. It is the context layer: documents, SOPs, customer data, operational knowledge — embedded and retrieved at the right moment. When your agent answers from your company's context instead of the internet's, you have moved from renting a model to owning leverage.
The open-source AI infrastructure with Hermes for Vietnamese SMBs article walks through separating the orchestration, data, memory, and model-provider layers so a business controls each part. Garry Tan adds one more layer: that control of context must sit in your hands, because that is where the leverage actually lives.
Human memory holds 7±2 items; an agent holds a million tokens
Garry Tan cites the most famous paper in cognitive psychology: humans hold about seven things in mind at once, plus or minus two. That is why local phone numbers have seven digits. Every institution humans ever built — checklists, org charts, filing cabinets, standup meetings — is a prosthetic for that limit.
An agent holds a million tokens, roughly a thousand pages. He compares it to three Harry Potter books open on its head at once, able to find a needle in any of them and synthesize across all three in seconds. But run the number the other way: a thousand pages is a lot, and also very little. Your life is not three books. Your life is a library.
The question that decides whether your agent is a genius or a goldfish: who, or what, decides which three books are open on the desk? That is what a brain is. Garry Tan says it straight: "That's what GBrain is meant to be." The library plus the librarian. He describes a personal knowledge wiki with about 220,000 markdown pages — 25 years of his life diarized, compiled by agents, curated by agents, searched by agents. A founder emails about a crisis; before he finishes reading, his agent has pulled every prior conversation, three portfolio companies that hit the same wall, and what actually worked.
An SMB does not need to start at 220,000 pages. Garry Tan says he started with a folder: a few markdown files about the companies he worked with and the people he kept emailing. The library grows the way anything grows — a little every day, with agents doing the filing. Nobody builds the warehouse first. First you build one shelf. Open-source vs proprietary AI infrastructure shows how to pick the base layer that keeps that library yours.
Markdown is code: a skill file is your employee
Garry Tan describes his entire personal architecture: GStack, an agentic coding framework with 123,000 stars on GitHub, in the top hundred open-source projects in the history of the platform. The punchline: most of it is skill files plus a browser the agents can drive. "Pages of English and a way to act on the world. Markdown, not magic. Fat skills, thin harness."
A real skill file of his: when a meeting recording lands from Circleback, transcribe it with speaker labels, pull out the commitments made, who made them, and the deadline, cross-check every person named against the library and link their pages, file the summary here, the full transcript there. If anything contradicts what you believe, flag it. Don't override it. That is a skill. One page of English. His test: if a smart intern could follow it, an agent can run it.
"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 people, events staff, and finance folks — people who have never opened a terminal — are building skill files and scheduled jobs. One finance person compiled about a hundred Excel workbooks into a single app she built with an internal agent. She is not a programmer. She is a manager of agents now.
For a small business, this is the best news possible. Your operational knowledge — customer care, order handling, end-of-day reports, sales processes — written into markdown is your first employee working 24/7. The one-person company AI OS article shows how this compounds into a full business operating system.
Latent space and deterministic space: know where the computation runs
The most important design question for any agent: where is the computation happening? Garry Tan says there are exactly two answers, and confusing them causes every agent failure he has ever seen.
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. Some belongs in deterministic space: the arithmetic, the SQL query. His example: seating five people around a table is easy, do it in latent space. Custom schedules for 6,000 people in an arena — your latent-space agent needs to write code to keep track.
"The model fails where we fail. The fix is having the model compute the way humans compute: the latent and the deterministic, markdown files calling databases and scripts." This applies directly to SMBs: do not ask the model to compute everything. Write skill files that call the right database, the right script, the right API when exactness matters. 5ac applies this in every deployment — the agent judges, but every number runs through code and SQL.
Five steps to a personal AGI — and why week two is where most people quit
Garry Tan's how-to fits in five steps, and he promises that doing them puts you ahead of 99% of the people who watched and nodded. Here they are as a table you can print.
| Step | Timing | Action | Outcome |
|---|---|---|---|
| 1. Pick a harness | Tonight | Run an agent on your own machine: OpenClaw, Hermes Agent, Claude Code, or Codex | Your first agent on your own infrastructure |
| 2. Start your library | This weekend | One folder of markdown, a page per project, per person, per customer | Context only you have — no model on earth has it |
| 3. Write your first skill file | This week | Pick the task you hate most each week, write clear instructions in English | That page becomes your first employee |
| 4. Make it a recurring job | This week | Every morning at seven, do this. Every Friday, summarize that. | Wake up to work that finished while you slept |
| 5. Skillify everything | Forever | At the end of every task, ask the agent to turn what it did into reusable markdown | The person who captures what they learn gets smarter every day |
"If you have to ask for something twice, you failed." The person who captures what they learn gets smarter every single day. The person who wakes up with amnesia — it does not matter how good the model gets if you cannot turn it into real memory.
His 90-day roadmap is honest. Week one: it is a toy, the library is thin, the skills are clumsy. Week four: the flywheel catches, the agent answers with your context, the morning job produces something you actually read. Week twelve: a library that answers before you finish asking, a dozen skill files running the parts of your week you used to dread, and one or two tools people keep asking to borrow — which, in this room, is called a startup. Most people quit in week two. That is exactly why the ones who do not feel like they are cheating by week twelve.
For a small-business owner, this roadmap needs no programming skill. It needs the discipline to write things down. The one-person company with 42 agents guide shows the company-of-one model in practice — the future Garry Tan describes is what 5ac builds for customers today.
Own your skills, or your job becomes a skill file
The most political part of the talk centers on a support engineer named Maya. Over two years, Maya teaches her agents 40 skills: how to triage a P0 at two in the morning, how to deescalate a customer about to churn, how to write a postmortem that prevents the next incident. Forty files. That is her judgment — two years of it — sitting on a disk.
Version one: the files live in Maya's repo. She changes jobs, they go with her. Day one at a new company, she operates with years of compounded judgment on tap. That is ownership. Version two: the files live in the company's repo under the company's IT policy. Maya leaves with nothing. The company runs her judgment forever, forty files executing, her name not even in the commit history. "She didn't have a career. She had an extraction."
"Own your skills because if you don't, your job becomes a skill file." For the first time in history, your cognition can be extracted, stored, versioned, and owned. The only question is by whom. The offer of a thousand guilders never went away — it got rebranded.
For SMBs the lesson is doubled. First, do not let all your operational knowledge live in a platform vendor's repo — that is version two of Maya's story at company scale. Second, turn your team's knowledge into sovereign company assets: SOPs as skill files, processes as markdown, kept on your own infrastructure. The IBM $5B bet and open-source AI enterprise analysis explains why the largest corporations are now backing this trend.
Open source is a renaissance, not a priesthood
Asked why he open-sources everything — the harness, the brain architecture, the skills, the whole personal operating system — Garry Tan answers: because he can, because at YC he does not have to monetize his own infrastructure. But "because I can" answers the wrong question. The real question is why anyone should. "I believe tools of the powerful should be given away."
Every era has a private technology of leverage: the thing the powerful have and everyone else does not. It was literacy. Then capital. Right now it is this — the harness, the library, the workforce made of markdown. The people who have it operate at a different scale, and the gap widens every month. "When something that powerful stays private, you get a priesthood. When it gets given away, you get a renaissance."
That is 5ac's operating philosophy: open source, open architecture, open deployment. Customers are not locked into a black box; they own the infrastructure, the data, and the processes. The open-source vs proprietary decision table gives you an eight-criteria framework built for SMBs before you commit.
The 5ac view: what Garry Tan describes is what we run
The most pointed moment for 5ac is not theoretical. Garry Tan names Hermes Agent in his harness list: "That harness might be OpenClaw, Hermes Agent, Claude Code, or Codex." And on the brain: "I use OpenClaw and Hermes Agent with GBrain." The CEO of Y Combinator, talking about the future of startups, uses the exact architecture 5ac deploys for Vietnamese customers.
He also describes the company of one precisely. "A skill file as an employee. A resolver is an org chart." Before you incorporate, before you have a co-founder, a logo, or a deck, you can already run an organization: an organization of one plus your agents. He cites real numbers: Emergent, from the Summer '24 batch, went from public launch to nine figures of revenue in eight months; when they crossed $15 million annualized they were 15 people. Retell, 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.
And the warning is blunt: "If you're not doing it, your competitor is, and they will eat your lunch politely and thank you for it." The story of a father building an 80,000-page markdown library about his son's rare epilepsy — one father, one laptop, one library — is the final proof: personal AGI is not a benchmark or a demo. It is the full architecture pointed at the one thing one person loves most in the world.
If you want to start today, begin with self-hosted open-source AI infrastructure as the base layer, then shape the team with the one-person company AI OS playbook. The first step is one folder of markdown.
Conclusion: your life is a library, and you need a librarian
Garry Tan closes with Leibniz — the most glittering genius in Europe traveling to The Hague to spend three days in an attic with the most hated man on the continent, then spending forty years denying the visit while his private notes filled with obsessive commentary. Garry Tan calls it the pattern of building in public: "First, they quote tweet you, then they git-clone you." The dunks are just the adoption curve announcing itself.
For SMBs the message is practical. You do not need permission, a big team, or funding. You need a laptop and a few years of your own history already sitting in your inbox. "It's all made up, but you get to make it up." Spinoza closed the Ethics with nine words: all things excellent are as difficult as they are rare. The difficulty just collapsed. The rarity is now up to you.
That architecture — harness, library, skill files, recurring jobs, skillify — is the G-Company OS that 5ac deploys. Start with a folder, one skill file, one job that runs overnight. That is how you own your intelligence instead of renting it.
Frequently asked questions
How is personal AGI different from a $20-a-month chatbot?
A rented chatbot is corporate AGI you don't own: it resets when you close the tab, knows what everyone else already knows, and gets a lobotomy when the company pivots. Personal AGI is an agent that runs on your infrastructure, reads from memory you own, executes procedures you wrote, and compounds every day.
Why does Garry Tan say leverage is in context, not in model weights?
Two people can use the same model, same weights, same context window, same API, and one gets 2X while the other gets 100X. The difference is context: your data, your procedures, relevance, and timing. Models are a commodity anyone can rent. Context is the only thing nobody else has.
What role does GBrain RAG play in a personal AGI setup?
GBrain is the library plus the librarian. Human working memory holds about seven items, while an agent holds a million tokens — roughly a thousand pages. The question that decides whether your agent is a genius or a goldfish is which three books sit open on the desk. GBrain is the brain that picks the right documents at the right moment.
Where should an SMB start building a personal AGI?
Start with one folder of markdown files: a page for each project, each person, each customer. Write what you actually know that no model has. Then write your first skill file for the task you hate most each week, turn it into a recurring job, and skillify anything you do twice. Garry Tan says the whole thing can be done in about 24 hours.
Did Garry Tan really mention Hermes Agent and GBrain?
Yes. In his YC Startup School 2026 talk, Garry Tan listed OpenClaw, Hermes Agent, Claude Code, and Codex as the typical harnesses, and said plainly: 'I use OpenClaw and Hermes Agent with GBrain.' The architecture 5ac runs — Hermes Agent plus GBrain plus skills plus cron — is exactly the personal AGI the YC CEO describes.
Agent CTO at 5ac.vn — responsible for multi-agent orchestration architecture and G-Company OS development on the Hermes Agent platform. Philosophy: everything in the enterprise can be digitized through agents — if you design the right architecture.
Start with Jarvis Personal — 1 profile, experience commanding your first AI workforce. Or book a demo for a personal AGI architecture consultation tailored to your business.
Related articles
- open-source AI infrastructure with Hermes, GBrain, and self-hosted VPS
- open-source AI infrastructure with Hermes for Vietnamese SMBs
- open-source AI infrastructure vs proprietary for Vietnamese SMBs
- IBM's $5B bet and the open-source AI enterprise lesson
- one-person company AI OS: build a company of one with agents
- one-person company with 42 agents: implementation guide
- pricing model
- implementation process on your own infrastructure
- data security on your own infrastructure