The Patient Reader

For eighty years, the math was never quite right.

In 1945, Vannevar Bush published "As We May Think" in The Atlantic — an essay imagining a personal information device he called the memex. It would compress and store all your books, records, and communications on microfilm, retrievable "with exceeding speed and flexibility." Bush had watched scientists drown in their own literature and thought technology could help. The memex would let them organize knowledge by association, the way the mind actually works, rather than by arbitrary alphabetical or numerical schemes.

The hardware he imagined never materialized. But the deeper problem he identified — that the economics of organizing personal knowledge were broken — has persisted right through the Obsidian era.

The broken math

In 2008, German researchers Max Völkel and Andreas Abecker published a formal cost-benefit analysis of personal knowledge management systems. Their model was straightforward: the benefit of a PKM system comes from efficiently retrieving task-specific knowledge; the cost comes from the effort of externalizing and structuring that knowledge in the first place. The conclusion was that the math only worked if your retrieval benefit was large enough to justify your structuring cost.

It rarely was.

The structuring cost is borne immediately, by you, right now. The retrieval benefit is deferred, uncertain, and also borne by you — a future you who may or may not remember the note even exists. Most knowledge workers internalized this tradeoff correctly, even if unconsciously. They took minimal notes. They organized loosely. They rebuilt context from scratch every time they needed it. This wasn't laziness. It was rational. The expected return on a well-tagged, carefully structured note was too low to justify the work.

What Bush actually imagined

The part of the memex people talk about least is the trails.

Bush wasn't just imagining a filing system. He was imagining a system where someone could pick up your associative trail and continue from it — where a future researcher would follow the exact path your mind took through a problem, branching where you branched, noting where your path diverged from theirs. The reader in Bush's vision wasn't just a future you. It was a future collaborator with patience and purpose, someone who wanted to understand the full shape of your thinking.

That never happened, for a boring reason: there was no such reader. Human collaborators have limited attention. Successors have their own contexts. Even a motivated colleague reading your notes would miss connections, skip structure, and give up when the path grew cold. The trails went nowhere, so nobody laid them carefully.

The reader, finally

What's changed is not the writer. It's the reader.

An AI agent connected to your knowledge base reads everything you've written in seconds. It doesn't skim. It doesn't miss the note from eighteen months ago. It finds the connection between the person record you filed in January and the project decision you made in March, which links to the preference you added in August. It does this every time you ask a question, not just when you happen to remember to look. It's available at 3am and it has read everything.

Andrej Karpathy, writing about what he calls "LLM knowledge bases," made an observation that cuts to the core of this: the right mental model isn't a search engine over your notes. It's closer to a research assistant who has read your entire library before you sit down to talk. The value isn't in the moment of retrieval. It's in the comprehensiveness of the prior read.

This changes the math completely. The structuring cost stays the same — you still have to write the note, tag the person, record the decision. But the retrieval benefit has multiplied by an order of magnitude. The reader is now infinitely patient.

What earns returns now

I used to think of careful structuring as a nice-to-have. I don't anymore.

A typed person record — name, relationship context, last contact, relevant history — used to be optional overhead. It took five minutes you didn't have and paid off irregularly when you happened to need it. Now it pays off every time you have a conversation about that person with an AI that has already read the record. The five-minute investment becomes a standing dividend.

A structured decision note — the context, the alternatives considered, the choice, the reasoning — used to mean writing for your past self's benefit at the cost of your present self's time. Now an agent can surface the relevant constraints when a similar decision comes up and notice when you're about to contradict a principle you already documented. The structure earns compound interest.

None of this required new technology on the writer's side. You write the same notes you've always written. The change is entirely on the reader side: for the first time, there's a reader who will actually use what you wrote, reliably, every time.

Bush imagined this in 1945. He just didn't have the reader to complete the circuit.

---

Asgeir Albretsen is the founder of Harbor.