A learning system that knows where you are, and keeps getting sharper the more you use it.
Schooling ends. An education isn't supposed to.
For a long time, knowing things was the differentiator.
Anything lookup-shaped is now free. An assistant hands it over in seconds.
What's left is judgment: knowing which question to ask, noticing when an answer is wrong, and having the thing in your head when it counts.
The version of it that started this:
“Tens of thousands of strong engineers are on the market. If I were too — what makes me distinguishable?”
Staying distinguishable is the job now, whatever your field. That means choosing what you learn, instead of letting circumstance choose for you.
You only learn what work or circumstance puts in front of you. Everything else stays a blind spot until it matters.
Tutorials are written for everyone, so they fit no one. Happy-path demos, and a good share of them teach what the field has already moved past.
Read the article, feel smart, retain almost nothing. Fluent prose feels like learning. It isn't.
A search engine, a chatbot, and a folder of bookmarks don't fix any of this. They make progress easy to feel and hard to accumulate.
In The Diamond Age, the Primer is a book that notices who's reading it and reshapes itself to fit them. It teaches through story and back-and-forth, always one step past what the reader can already do.
“The magic was never the nanotech. It's that the Primer is a relationship, not a syllabus — patient, genuinely curious about its reader, and built around exactly one person.”
Stephenson had to invent the nanotech. The rest of it — a tutor that knows where you are, adjusts to how you think, and is still there next year — turns out to be buildable without it.
Prompted by one field that keeps moving. Driven by everything else: how mountains get made, what counterpoint is, why the cat does that.
A Claude Code skill. You type /primer <topic>; it runs a calibrated lesson and writes down what it learned about you. The rules it never breaks:
Talks to you like a peer, at whatever directness you ask for. No cheerleading. Disagrees when you're wrong, and holds that position under pushback.
Makes you predict or derive before it explains. You build the idea in your own head. That's the part that sticks.
Every technical claim tagged verified via docs or from-training. Never teaches a take the field has moved past.
Calibrated to you, and sharper the more you use it.
The split is what lets your data be honest about what you know and what you care about without any of it touching a public repo. The engine never goes digging through your own files.
It opens by asking what you already believe, then probes with a real failure scenario and waits. No auto-complete. It teaches only at your edge: skip what you know, slow down where you're fuzzy.
It closes by handing you the pen. You put the takeaway in your own words, which is where the encoding happens, and the session emits retrieval prompts for spaced review later.
Every topic carries a depth marker with confidence and evidence, never a certain-looking guess:
async event loop confidence: low
↳ intake probe: reasoned well, but the
blocking-freezes-the-loop invariant isn't locked
outbox pattern confidence: high
↳ derived it live, sketched the schema
Confidence moves both ways, and the loop doesn't get to grade itself:
So the picture of you gets truer with use. It is allowed to get worse.
Ask a chatbot to “explain X” and you get a clean paragraph. You nod. You feel sharper. You retain almost none of it, because nothing was built in your head.
Primer is built against that:
The goal isn't a good answer in the moment. It's getting good at the thing, and better at learning in general.
Every session writes a lesson file: TL;DR invariants, worked examples, diagrams, an honest tradeoff table, and the exact questions you asked.
Atomic Q/A pairs from every lesson, built to test understanding. Review them in Primer or in your own SRS.
Markers, evidence, open threads, next-lesson suggestions. The map sharpens, so the next lesson starts where you are.
It's all yours: plain markdown in your own private repo. Nothing locked inside a product.
No quiz, no generic curriculum. One real scenario shows it the gap between what you said and what you can do.
Then it builds the lesson on that edge, with sources already verified against current docs, ready when you are.
(A real exchange from setting this up. The subject happens to be Python; it works the same on plate tectonics.)
Where you are, what you're reaching for, and the way you like to be talked to. The picture gets truer every time you show up.
$ git clone … && ./tools/install.sh
› /primer init # one-time intake, then go
› /primer <topic>
Open core, private data, any subject: geology, astronomy, a new language, game design, music theory, distributed systems.