Write-Only

You know that moment. You open your notes app to find something you wrote six months ago: a decision you made, a name someone gave you, an idea that felt important at the time. You try three searches. Nothing useful comes back. You end up Googling it from scratch.

I blamed myself for years. Bad tagging, lazy organization, inconsistent practice. But I've come to think the problem isn't discipline. The entire note-taking industry spent two decades solving the wrong problem.

The capture obsession

Every major note app improvement of the last decade has been about capture. Faster inbox. Web clipper. Quick-add keyboard shortcut. Automatic sync. The design assumption is that notes fail because capture is too slow or too inconvenient, so the solution is always to make writing easier.

This isn't crazy — it's just incomplete. Getting things in matters. But the moment a note lands, most apps consider their job mostly done. What happens next — finding it, making sense of it, connecting it to what you're actually working on right now — is usually an afterthought.

William Jones, a researcher at the University of Washington, spent years studying this asymmetry. His 2007 book Keeping Found Things Found is probably the most comprehensive account of how people actually manage personal information. One of his central findings: we are much better at getting information into our systems than getting it back out. We find things and then immediately begin losing them. He called it the re-finding problem.

The problem predates digital notes. But apps made it worse in a specific way: they made capture so cheap that it became automatic. You can clip an article in two seconds, forward an email with one tap, dictate a voice note mid-commute. The inbox grows. The retrieval difficulty stays constant or increases. Eventually you have ten thousand notes that all feel useless.

The asymmetry nobody fixed

Here's what the industry missed: the value of a note isn't in the writing. It's in the reading.

A note you can't find when you need it is functionally the same as no note at all. Possibly worse — it creates a false sense of having captured something. You "know" you wrote it down, so you don't hold it in working memory. When you need it, it's gone, and the effort to reconstruct the thought costs more than the original capture ever saved.

Cognitive scientists who study learning have a useful frame here. Retrieval practice — pulling something back out of memory — is where encoding actually deepens. The write moment doesn't strengthen the memory. The read moment does. That's why testing beats rereading, consistently, across a hundred studies. Note-taking tools optimized for one half of this loop and mostly ignored the other.

What changes when AI is doing the retrieving

There's a version of this argument that lands where it always lands: be more disciplined about tagging. Better organization, better folders, better habits. Most people have tried this. It helps a little.

But something genuinely different is happening now. If an AI agent can search your knowledge on your behalf (not keyword search, but something that understands what you're asking), the retrieval problem looks different. The question shifts from "did I organize this well enough that I'll remember the folder path?" to "is the underlying structure rich enough for an agent to find it?"

This reframe matters. You're no longer designing notes for yourself-in-six-months, who has to remember that she filed the thing under "client X / decisions / Q3." You're designing notes for a system that can read them at query time and understand context, relationships, and intent.

That changes what good note-taking looks like. Prose works better than fragments. Context beats brevity. A note that says "Talked to Lars, he's skeptical about the vendor — worth raising before we sign" is much more useful to an AI than "Lars: vendor concerns." The former survives retrieval. The latter might not.

Structure matters too. Knowing that Lars is a person, that there's a vendor relationship, that signing something is a pending decision — these aren't organizational niceties. They're the anchors an agent uses to surface the right thing at the right time. Plain text buried in a folder doesn't give much to work with. Typed relationships do.

The honest complication

I don't want to oversell this. AI-assisted retrieval is better than keyword search, but it's not magic. It still needs something to retrieve. Sparse notes, context-free fragments, vague observations — these are harder to work with whether a human or an agent is reading them.

And you still have to write the note in the first place. A system designed for retrieval still requires capture. The two ends of the pipe both matter.

But the emphasis has shifted. If you're building a knowledge practice, the question worth asking isn't "did I write this down?" It's "would a careful reader be able to find this and understand it six months from now?"

Most note-taking tools were not built to prompt you toward that question. They optimized for the capture moment and left retrieval as your problem. That's a reasonable way to ship a product. It's a strange design for a memory system.

---

Asgeir Albretsen is the founder of Harbor.