Your AI is not thick. It is starving. You just never fed it anything about you.
People keep hunting for the magic prompt, the secret phrasing that turns a chatbot into a genius. It does not exist. The folk whose AI feels like sorcery did one unglamorous thing: they gave it a database of their life, so it answers with context instead of guessing.
Angus Sewell (Angus the Nontechnical) lays out the exact build, one table at a time, with Claude doing the technical lifting so you do not have to. It is the most useful framing of “getting good at AI” I have read in a while, because it moves the effort from clever wording to plain groundwork. And the tooling has caught up. Claude’s Projects hold around 200,000 tokens of your own material as standing context, and by 2026 ChatGPT’s memory reportedly juggles over 10,000 separate facts per user. Under the bonnet this is usually retrieval-augmented generation: the model pulls the relevant scraps from your store before it answers, so it works from your reality rather than the internet’s average.
One distinction worth nailing, because vendors blur it. A big context window is not the same as memory. The window holds a lot in one sitting then forgets everything when the chat closes. Real memory persists across sessions. Angus’s database is the second kind, which is why it compounds instead of resetting every morning.
Now the part the excited build guides gloss over. A single store describing your whole life (finances, health, clients, relationships) is a magnificent thing to hand an assistant and a magnificent thing for someone else to steal. Centralise everything about yourself inside one vendor’s cloud and you have built a very tidy target. So before you pour your life into a text box, ask where it lives, who can read it, and whether a local-first option (there are open-source agents that keep memory on your own machine now) would do the job. A database that runs your life should not sit one breach away from running someone else’s con.
Feed the machine, absolutely. Just keep a copy of the spoon.
File this away: a large context window resets the moment you close the chat, while true persistent memory survives across sessions, and confusing the two is why so many people swear their AI has “forgotten” them.