Why Your Notes Never Surprise You

Darwin wrote the phrase "I think" on a now-famous page of his transmutation notebook in 1837, beneath a rough sketch of a branching tree. He'd read Malthus fifteen months earlier — the population pressure argument that would crystallize into natural selection. The puzzle is what happened in those fifteen months. No experiments. No fieldwork. No apparent breakthrough. Just Darwin continuing to fill notebooks.

Steven Johnson, in Where Good Ideas Come From (2010), looked closely at Darwin's notebooks from this period and found something curious: the full theory of natural selection was already nearly present months before Darwin's supposed epiphany. The notebooks weren't recording ideas that had happened somewhere else, in Darwin's head. They were the place the ideas were happening. The notebook, Johnson argues, was a platform for cultivating slow hunches — not storage for finished thoughts, but a medium in which half-formed intuitions could survive long enough to collide with new material.

That distinction between archive and incubation chamber is the thing most knowledge tools miss entirely.

The Luhmann problem

Niklas Luhmann produced 70 books and over 400 papers in a 40-year academic career. Asked how, he pointed to his Zettelkasten: 90,000 index cards, cross-linked and densely annotated. Not as a storage system. As a communication partner. His 1981 paper on the method makes this explicit: it is an obligatory condition for communication, Luhmann wrote, that both partners can mutually surprise each other. A slip-box works when it reaches sufficient density that adding a new card requires you to browse for related cards to link it to, and in that browsing you encounter things you'd forgotten you'd thought. The collision is the mechanism. The surprise is the product.

The key phrase is "both partners." Luhmann wasn't describing search. He was describing a system that could initiate, not just respond.

Most note-taking apps are pure responders. They wait to be queried. They have no state that changes when you add a new note. They don't browse. The closest thing in mainstream PKM tools to proactive surfacing is a backlinks panel, which shows you what already links to a note, not what might relevantly connect to the note you just wrote. Backlinks are a map of connections you already made. Luhmann's insight was about connections you haven't made yet.

The incubation isn't magic

Psychologists have studied the incubation effect for decades. One strong explanation is "opportunistic assimilation": ideas held loosely in an active state get connected to information you encounter serendipitously, in contexts where you weren't looking for them. The collision happens because the hunch is still live enough to recognize a relevant signal when it appears.

This suggests a design implication. For incubation to happen, the half-formed thought has to remain findable to something — a browsing mind, a scanning process, a background system. Darwin's notebooks sat open on his desk. Luhmann's cards were browsable by accident, during the physical act of filing a new card. The mechanism that enabled the collision wasn't sophisticated retrieval. It was cheap serendipitous access.

Digital notes fail this at the level of interface. Capture is one-directional. You write a note, it disappears into the archive, and nothing happens until you go looking for it. The interface has no concept of a hunch in active incubation versus a thought that is finished and filed. Both look the same. Both wait silently.

What structure makes possible

Here is where an AI working with structured knowledge could do something genuinely different.

When you add a new document to a knowledge base — a decision record, a person note, a project brief — an AI that understands the shape of that knowledge can ask: what in the existing base is relevant to this? Not "what does the user want to retrieve?" but "what old material might meaningfully collide with this new material?" The proactive surface is the point.

Full-text search can't do this well. Keyword matching finds documents that contain similar words; it doesn't find documents that bear on the same question even if they use different vocabulary. And unstructured prose is hard to reason about semantically at useful granularity. If your person record for a colleague is a paragraph of freetext notes, an AI can tell you it mentions the word "Berlin" — but it can't tell you that this person's stated preferences overlap with the constraint in the project you just opened.

Typed entities — person, task, decision, preference — make this tractable. A decision record has a context field and a constraints field. A preference record has a category and a value. A new task document has assignees and dependencies. An AI can traverse these relationships and surface specific, concrete overlaps: this decision you made eight months ago has a constraint that bears on the thing you're working on today.

That's the proactive incubation that Darwin's notebook provided, reconstructed as software.

The archive you built vs. the one you need

There's a pattern I've noticed in how people start using a structured knowledge base. They treat it as a better archive — a nicer place to keep things they've written. The typed entity layer feels like extra work at first. Why specify that a person note is a person rather than just a document? Why use a decision record format rather than a regular note?

The answer shows up a few weeks in, when you're writing a new project brief and something surfaces from six months ago — a person record for someone who'd expressed interest in exactly this kind of work. You didn't search for that. The system found it because the person record and the project brief had overlapping structured fields.

Luhmann called this a surprise. Darwin spent fifteen months cultivating the conditions for one. The difference between an archive and a system that thinks with you isn't the quality of the search. It's whether the system can initiate.

---

Asgeir Albretsen is the founder of Harbor.