Why Your Notes Rot

I have a note from 2022 titled "career directions I'm considering." It has four bullet points. Three of them describe paths I'm no longer remotely interested in. One was the thing I ended up doing. But when I open the note now, there's no marker telling me any of this. It just sits there, confident in its outdatedness, like a horoscope that didn't age well.

This is the quiet problem with personal knowledge management. Not that you don't write things down — most people who care about this stuff write plenty. The problem is that notes age. They degrade. And almost no note-taking system is designed to handle that.

Knowledge has a half-life

The concept of a "knowledge half-life" — the time it takes for half of what you know about something to become wrong or obsolete — was formalized by Fritz Machlup in 1962. His original observation was about engineering: a degree that had a useful shelf life of 35 years in 1930 had already shrunk to 10 years by 1960. Modern estimates put it somewhere between 2.5 and 5 years.

That's for technical knowledge, where "wrong" has a reasonably clean definition. For personal knowledge — notes about decisions, people, preferences, projects — the decay is harder to measure but just as real. A preference you recorded in 2021 may be genuinely obsolete. A note about someone you worked with might be missing three job changes, two major decisions, and whatever happened in between. The words are still there. The accuracy isn't.

Nobody talks about this as a design problem, but it is one.

The decay is in the context, not the words

Notes don't usually become "wrong" in the way that a fact becomes wrong when new evidence arrives. They become disconnected. The context that made them meaningful shifts — you change your mind, your situation changes, your understanding develops — and the note just sits there, frozen at the moment of capture, increasingly orphaned from the world it described.

Niklas Luhmann, whose Zettelkasten produced an extraordinary body of work before his death in 1998, had a partial solution: he actively revisited notes, linked new ideas back to old ones, and treated his slip-box as a living system of evolving thought rather than a static archive. His daily practice wasn't just capture. It was active curation — reading what he'd written, updating connections, letting old ideas be transformed by new ones.

Most people don't do this. Most people have a capture practice, not a curation practice. And capture without curation produces an archive that grows in volume and declines in relevance at roughly the same rate.

What makes a note go stale

There's a distinction that most tools don't help you make: between notes that were always meant to expire and notes that were meant to last.

A meeting note from a project kickoff is a point-in-time artifact. It captures what was said, who was there, what was decided. That note should gradually matter less as the project moves forward and the decisions get revised or executed. A vault full of meeting notes isn't really a knowledge base — it's a log. Logs are useful, but they're not the same thing as knowledge.

A real "knowledge note" is a claim about something meant to stay true over time: how someone prefers to communicate, what a project is actually trying to accomplish, a decision you made and why. These notes need maintenance. Not because the words change, but because they accumulate context — amendments, corrections, newer observations that belong alongside the old ones.

Most tools don't help you tell these apart. They definitely don't help you identify which notes have quietly gone stale while you were busy adding new ones.

Where AI fits in

This is the part I keep thinking about, and part of why I built Harbor the way I did.

The maintenance problem is fundamentally a pattern-recognition problem. Which of your notes are probably out of date? Which contacts haven't been updated in two years, even though you've talked to those people weekly? Which preferences are from a role you no longer have? Which decisions have downstream notes that now contradict the original record?

A human doing this manually is exhausting. But it's not a hard problem for an AI with structured access to a knowledge base — if the notes are organized in a way that exposes time, relationships, and data types.

That's why Harbor stores notes not just as text, but as typed entities with timestamps and explicit links between them. When an AI reads your knowledge base through Harbor, it's not scanning a pile of Markdown files. It sees people, tasks, projects, preferences, and decisions as distinct objects with known fields. It can ask: when was this last updated? Does this still match what I know about this person? Is this preference consistent with what you've told me more recently?

The goal isn't an AI that automatically rewrites your notes. That would be worse. The goal is an AI that surfaces what might have rotted, so you can decide what to do about it. A diff proposal. A flag. A question: "You wrote that you preferred async communication in 2022. Is that still true?"

That's a different kind of tool than a note-taking app. It's a knowledge maintenance system with an AI layer that works against entropy, rather than just accumulating files alongside it.

The honest answer

Right now, most of us are losing this battle quietly. Our vaults grow. Our notes age. Our context refills with stale beliefs we'd update if someone asked.

Nobody set out to build a system that degrades over time. But capture without curation is what almost every tool encourages. Write it down, tag it, link it, and forget it — right up until the moment you need it and it's three years out of date.

The fix isn't more notes. It's a different relationship to the ones you already have.

---

Asgeir Albretsen is the founder of Harbor.