How I Take Notes: Capture, Distil, Resurface

In my final year of university, studying mostly alone through lockdowns, I realised I needed a better way to learn. My bet was simple: if I understood something properly the first time, I wouldn’t need to cram it later. I adopted a Zettelkasten-style method, and it carried me through a dissertation that meant reading hundreds of papers.

Five years on, the principles are the same. The machinery is very different.

The original method (still true)

1. Consume slowly. Read or listen with intent. At each step, ask what it means in your own words and how it connects to something you already know.
2. Draft in your own words. Copying and pasting teaches nothing. Paraphrasing forces understanding.
3. Write to teach. Rewrite the draft as if explaining it to someone else — or to yourself in three years. Add diagrams and examples, and link it to related ideas.
4. Organise into a web. File the note among your others and link it, so knowledge grows as a network, not a pile. I used Obsidian then and I use it now.

What I got right: own words, small notes, links. What I got wrong: I treated organising as my job. Over the years my vault filled up with half-processed notes I never looked at again. Capture was easy; distilling and revisiting weren’t, so they didn’t happen.

The loop I use now

These days the system runs on three verbs, borrowed from a “log, review, discover” approach and adapted:

Capture. Everything goes into today’s daily note under one heading: a thought on the train, a lesson from a code review, a training note, a photo of a notebook page. No filing, no tags, no templates. The only rule is that capture must be faster than forgetting.

Distil. At the end of the day an AI assistant (Claude Code, running against the vault) reads the day’s log and triages each entry as signal or noise. Signal gets promoted: a learning becomes or extends an atomic note, a decision goes into its project, an idea goes into an ideas list as a seed, a task stays a task. Each promoted note links back to the day it came from. Noise stays where it was — and crucially, my raw words are never rewritten. Cleaned-up versions live in the destination notes; the original log stays as written.

Resurface. Every distilled note carries a review date on a spaced schedule — a few days, then longer, then longer again. When a note comes due, it shows up in my daily plan. I either confirm it still holds (the interval grows), improve it, or let it go. Spaced repetition for ideas rather than flashcards.

The harness matters more than the model

The part that makes this trustworthy isn’t clever prompting. It’s a few boring constraints:

• One constitution file at the root of the vault that says what counts as signal, where each kind of note lives, the frontmatter conventions, and the rules (never rewrite the log, never delete — archive instead).
• Small, named commands for each recurring job: start the day, close the day, weekly review, process scanned notes.
• Git as the safety net. Every run ends in a commit, so any change can be inspected or undone.
• A publishing gate. Some distilled notes become pages on this site, but only after a privacy check: ideas, patterns and lessons get published; personal details never do. When in doubt, it stays private.

Without those, an agent in your notes is a very confident intern with write access. With them, it’s a librarian.

What hasn’t changed

The AI doesn’t do the learning for me. Writing in my own words is still where understanding happens, and the daily log is still mine. What’s automated is the filing, linking and remembering-to-revisit — the parts I never did consistently.

The other constant: notes are never finished. The status badges on this site (seed, growing, evergreen) and the “last tended” dates come straight from that idea. A note is a plant, not a publication.

Takeaways

• Own words, small notes, many links — that part of Zettelkasten is timeless.
• Make capture frictionless and separate it from organising.
• Automate the parts you reliably skip (distilling, linking, resurfacing), and keep a human in charge of meaning.
• Put guardrails around any agent that touches your notes: clear rules, small commands, version control.

Related: Learning When Everything Wants Your Attention covers the attention side of the same problem.

---

Originally written in 2021 → How I write notes: Zettelkasten Technique