Thoughts on knowledge, tools, and AI.
Writing about personal knowledge management, building in public, and the ideas behind Harbor.
The Price of Legibility
Making your notes machine-readable is supposed to make them more useful. The part no one mentions is what gets lost in translation.
The People You Forgot You Know
Your contacts app stores names, not relationships. That's a design choice with consequences that compound over years.
Memory Is the Wrong Word
Why calling AI knowledge storage 'memory' sets up exactly the wrong expectations — and what the right word reveals about how these tools should work.
Every File Is a Bet
The 1086 Domesday Book is still readable. The 1986 digital version wasn't, sixteen years after it was made. What choosing a file format is really asking you to decide.
The Semantic Layer Between Humans and Machines
Not code, not prose. The case for structured Markdown blocks as the grammar for human-AI collaboration.
Notes Apps Are Lying to You About Search
The promise of note-taking is recall. What most apps actually give you is lookup — and that only works when you already remember.
Beliefs Without Error Bars
Notes record what you believed. Almost none of them record how sure you were — and that omission gets worse when an AI is the one reading them.
Building Trust with Software, One Visible Action at a Time
Trust in AI tools isn't declared — it's earned through repeated, observable, reversible actions.
When Should AI Update Your Contacts?
You mention in passing that someone changed jobs. The AI heard it. Should it update your contact record automatically, or ask you first?
Why Self-Hosting Is Having a Moment
Docker made it easier. But the real driver isn't technical enthusiasm — it's something that happened to trust.
Why I Want My AI to Show Its Work
Fluency is not the same as trustworthiness. The AI tools I actually rely on are the ones that show diffs, log what changed, and let me check before they commit.
Knowledge Without Expiry Dates
Your knowledge base is an archive with no expiry dates. Research on misinformation suggests that's more dangerous than it sounds.
The Hidden Cost of Convenience in AI Tools
When AI tools write automatically, learn silently, act without asking — you trade legibility for speed. Some of that friction was doing something.
The Word You Used Then
Your notes are hard to find not because they're disorganized, but because past-you wrote in words that present-you doesn't think in.
The Workspace as a Calm Place
Not every tool needs to be an infinite canvas. The argument for focused software that knows what it isn't.
The Folder as a Permission Boundary
Folders aren't just for finding things. They're one of the oldest security primitives in computing, and the right unit of control for AI access to your knowledge.
Publishing Your Knowledge Base to AI (Safely)
A public MCP endpoint isn't just a developer curiosity. It's a fundamentally different way of sharing knowledge — not with human readers, but with AI systems that query it by permission.
What Gmail Knows About Your Life (and Why Your Notes App Should Too)
Email became the most comprehensive personal knowledge store most people have. The places where you actually write to know almost nothing.
What It Means for a Tool to Respect Your Time
Software that constantly demands your attention is software that disrespects your time. A meditation on calm design, async work, and the tools that know when to wait.
The Brilliant Stranger Problem
Why intelligence without context produces the worst kind of advice — and what it takes to build an AI that actually knows you.
Why Experts Take Bad Notes
Your notes are densest where you were confused. They thin out as expertise deepens. What your AI reads isn't you — it's a portrait of a previous version of you.
Why Your Notes Rot
Information has a half-life. Notes taken in 2021 often feel useless in 2025 — not because you forgot them, but because the world moved.
The Context Window Is Not the Constraint
Everyone assumed bigger context windows would solve the AI memory problem. The research says otherwise. The bottleneck was always retrieval.
The context that compounds
The most useful AI context is exactly the kind that feels too small to bother capturing.