Tan’s reframe is that AGI is not arriving as a single event in a data centre but diffused — “your agent running on your context, doing your work” — and he is precise about what he is not describing. Not a chatbot at $20 a month, not a better autocomplete, not an assistant that knows your calendar and nothing else. Those are “a 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 assistant “gets a lobotomy on someone else’s schedule.” Personal AGI is an agent running on your infrastructure, reading a memory you own, executing procedures you wrote. The asymmetry he draws is about the derivative: the rented one improves when the vendor ships, the owned one improves every day you use it.
His own instance is a knowledge library of roughly 220,000 markdown pages covering 25 years diarised — email, meetings, notes, drafts, the things he got wrong — compiled, curated and searched by agents, so that he never re-asks a question he already answered. The starting point was not the warehouse: it was a folder of a few markdown files about the companies he worked with and the people he kept emailing, and the standing advice is that nobody builds the warehouse first, you build one shelf. He argues most people are already sitting on five or ten years of their own history in an inbox — “that’s your moat just lying there, unindexed.” The claim is a personal-scale version of Context is the product, not the model and the individual counterpart to Closed harnesses behind APIs create memory lock-in by design, with the ownership question answered in the user’s favour rather than the vendor’s; the political stakes are spelled out in Own your skill files, because if you don't, your job becomes a skill file.