Guide
Atomic notes are great until they aren't : a practical critique
6 min · 2026-06-07
Atomic notes — "one idea per note" — became gospel in PKM communities after Building a Second Brain popularised them. The promise: small composable units you can recombine into new thoughts. The reality, for most users: 200 fragments that no longer form a coherent thought. This guide is a practical critique of atomic notes — what they're good for, where they fail, and what the calmer alternative looks like.
When atomic notes work : research, deliberate composition
Atomic notes work for one specific workflow: a researcher or writer who is deliberately constructing arguments and needs to recombine elements. If you're writing a thesis, a long essay, a non-fiction book — having atomic notes you can shuffle into different sequences is genuinely useful.
When atomic notes fail : life capture, fast memory, mixed contexts
For everyone not writing a thesis, atomic notes invent a problem to solve. You capture a voice memo about a client call. It contains: a decision, two open questions, three people, one date, one idea for next time. Atomic-notes orthodoxy says: split into six notes, link them. You won't. You'll either dump everything raw or skip the capture entirely. Both are losses.
The capture-then-split workflow assumes you have time to do the splitting. Real users don't. They capture in 4 seconds between two meetings. The atomic-note discipline is incompatible with how memory actually gets created — in messy chunks, mid-life, fast.
The calmer alternative : keep chunks, let AI split semantically when relevant
What actually serves human memory: keep messy captures as they came in. Trust the AI memory layer to find the right concept inside them when you query. When you ask "what did Adam say about pricing," the system finds the relevant 12 seconds inside a 2-minute voice memo. You didn't have to atomise. The retrieval does the slicing for you, when needed.
Atomic notes are a research-writing technique, not a life-memory technique. Forcing them on everyday capture creates fragmentation without composability. The calmer pattern: capture chunks the way they arrive, let the AI memory find what's relevant inside them when context calls for it. That's the wamid model — and it leaves you free to live instead of curate.