Thoughts on knowledge, tools, and AI.
Writing about personal knowledge management, building in public, and the ideas behind Harbor.
The Undo Problem
Warren Teitelman invented the undo command in 1971. Fifty-four years later, most AI tools that write to your data still don't have one.
The Unit of Knowledge
Linnaeus figured it out in 1767. Luhmann refined it in the 1950s. Most note-taking apps still haven't caught up.
When AI Should Ask Permission (and When It Shouldn't)
Asking for approval isn't automatically safe. The design question is which actions earn a prompt, and what happens when you get that wrong in both directions.
MCP and the New Shape of AI Tools
Model Context Protocol is being called 'USB-C for AI,' which is accurate but undersells it. A closer look at what it actually changes — and what it doesn't.
The More You Tell the AI, the Worse It Gets
Context rot is real: every major frontier model degrades as input grows. What that means for how you build a knowledge base for AI.
Two Kinds of Memory, One Kind of Notes App
Tulving's 1972 distinction between episodic and semantic memory reveals a structural flaw most note-taking tools have never acknowledged.
The Briefing Problem
Notes are written for the writer, not the reader. That becomes obvious the moment you need to hand your context to someone else.
The Fifty-Year Question
In 1972, researchers built the first natural language database. It worked. Then everything went wrong for fifty years — and what broke it tells you something important about personal knowledge.
Bring Your Own Keys and What It Signals
BYOK is framed as a pricing feature. It's actually a statement about whether a product wants to be in your data flow at all.
What Librarians Figured Out That Note Apps Forgot
Before 1876, libraries organized books by arrival order. Melvil Dewey fixed that. Most note apps haven't caught up.
What Your Notes Can't Tell You
Every note looks equally authoritative when you read it later. Most of them weren't.
The Problem with AI Memory Living Inside the AI
AI memory you can't read, edit, or verify isn't really yours. Opacity is the core design flaw in how most AI systems remember you.
Why Your Notes Never Surprise You
We build knowledge systems as archives. Luhmann and Darwin built them as incubation chambers. The difference turns out to matter.
Good Enough for Who?
A note that works perfectly for its author can fail completely for every other reader — including the AI you've connected to your knowledge base.
What the Empty Field Asks You
Typed entities in a knowledge base are not just better storage. The schema is a question, and it changes what you notice.
Why Agents Need Memory Files
A 2025 survey found that the gap between agents with memory and without is larger than the gap between different models. Nobody talks about this.
The Moments You're Too Busy to Write Down
The most important decisions in your working life are systematically absent from your knowledge base. That's not a discipline problem. It's structural.
The Note Has No Schema
A hospital in 1967, a software engineer in 2011, and the note-taking app on your computer all face the same problem: the note is not a good data structure.
The Person Record as a Primitive
Contacts apps store who people are. Your notes store what you talked about. Neither is where you look when you actually need to remember.
What Lab Notebooks Got Right
Scientists spent centuries figuring out what makes a record trustworthy. Personal knowledge tools mostly ignored the answer.
When Taking Notes Makes You Forget
Linda Henkel's 2014 museum study found that photographing objects made people remember them worse. The same mechanism operates in note-taking, with implications most PKM tools ignore.
Where the Context Goes
The 23-minute interruption stat is well-known. But nobody talks about the cost of returning to a project you left three weeks ago.
The Approval Problem in AI Tooling
Binary approval models are broken. The interesting design space is in the middle: policy-based, context-aware, and granular.
The question you wouldn't have asked
Rich personal context doesn't just improve AI answers. It changes which questions feel worth asking.