Personal AGI is here — and the dream is 80 years old
Garry Tan told a room full of founders to stop watching the sky for AGI. The machine he described already has a name — it got it in 1945.
The heresy
At Startup School 2026, Garry Tan — President and CEO of Y Combinator — gave a talk called “Own Your Intelligence.” It opens, improbably, with Spinoza: the seventeenth-century philosopher who ground lenses by day, wrote the most dangerous book in Europe by night, and was excommunicated for suggesting that God was not a king on a throne but something spread through everything that exists.
The point wasn’t historical. It was that we are making the same category error about intelligence that Spinoza’s contemporaries made about divinity. We keep waiting for AGI as an event — an announcement, a threshold, a god in a data center, a day the sky changes color. Tan’s updated heresy is that “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.
He gave this diffused arrival a name: personal AGI. Not artificial general intelligence for everyone all at once — general intelligence for one person. Your agent, running on your context, doing your work.
And then, tracing the lineage of the idea, he said six words:
“Vannevar Bush called it the Memex.”Garry Tan, “Own Your Intelligence,” Startup School 2026
The dream is older than the transformer
In 1945, Vannevar Bush — the engineer who ran America’s wartime research effort — published “As We May Think” in The Atlantic. In it he imagined a device where an individual could store all their books, records, and correspondence: “an enlarged intimate supplement to his memory,” mechanized so it could be consulted with speed and flexibility. Items would be joined by trails of association, the way a mind actually moves — hypertext, described decades before the web, and something more personal than the web ever became.
Bush’s machine was never about knowing everything. It was about one person’s accumulated reading, thinking, and deciding — made permanent, connected, and instantly available to its owner. Not a library. A memory.
Every generation since has rebuilt a fragment of it — the personal computer, the wiki, the note-taking app — and every fragment stopped short of the same thing: a memory that doesn’t just store, but works.
Renting is not owning
Tan was precise about what personal AGI is not, because the words “personal AI” have already been captured by marketing departments. It is not a chatbot you pay twenty dollars a month for. That is a subscription you rent — a corporate AGI that resets when you close the tab, knows only what everyone else already knows, and when the company behind it pivots, your so-called assistant gets — his phrase — “a lobotomy on someone else’s schedule.”
The distinction is ownership, and ownership has a concrete architecture. If you take the idea seriously, an intelligence you own must meet five requirements:
- It lives on your machine. Not your account — your hardware. Confidential work stays confidential by physics, not by policy.
- It works on your files. Plain, portable formats you can open without it. If the software disappears tomorrow, your memory doesn’t.
- Its memory compounds. Every book, meeting, and decision makes the next answer better. Renting resets; owning accrues.
- Its answers are auditable. Every claim cites the source in your own corpus, so you can check it the way you’d check a colleague.
- The model is swappable. The intelligence layer is a component you choose — local when privacy demands it, frontier when the problem does — never a landlord.
Read that list again. It is not a description of a smarter chatbot. It is a description of Bush’s machine with an engine in it.
Why our machine carries the name
This is the machine we have been building. We called it Memex — the name is a debt, openly acknowledged, to the 1945 blueprint — your sovereign personal AGI that lives on your Mac. It turns your books, articles, PDFs, and decisions into a connected memory it can reason over, act on, and cite. It runs a local model when nothing may leave the building, or the best frontier model when you choose — your call, per question. Underneath is MemexLab, the open-source governed memory layer it’s built on: local-first, citable, auditable.
To be clear about what this essay is not: Tan wasn’t talking about us, and no endorsement is implied. He was describing a category — and the category finally has demand to match its eighty-year-old supply of ideas. When the president of Y Combinator tells every founder in the room that the intelligence they rent will never compound like the intelligence they own, the argument for a machine like this stops being contrarian. It becomes a to-do list.
Conatus
Tan anchored the talk in Spinoza’s word for the engine that kept him grinding lenses: conatus — the striving in every living thing to persist and to increase its power to act. Not your title, not your résumé. The striving itself.
That is the honest job description for personal AGI. Not to replace your thinking — to compound it. You can outsource your thinking. You can’t outsource your understanding.
The sky isn’t going to change color. The machine is already on your desk. The only question left is whether you own it.