Now
Updated May 7, 2026
What I'm building in public.
Site is early. The skeleton is up; the writing is starting to land. I'm one person at a desk, building this in the open.
North star
Build a reading surface for people who want to think clearly about AI, data, and modern work — by writing slowly, naming the parts I'm not sure about, and only shipping what I'd want to read myself.
One sentence. Revisited every quarter. Everything below is in service of it.
The week, on the bench
What is open right now.
Building
The site itself — one writer, one desk, no team
First version is live. Most of the work right now is shaping the rails so future essays land in the right place.
- The first batch of essays
A handful are drafted; a few are published. The voice is still tightening.
- A small set of systems and labs
Frameworks I actually use, written down honestly. Smaller and slower than I'd like — that's the whole point.
Reading
Superforecasting
by Tetlock & Gardner
On calibration. Trying to write claims I could be scored on.
The Practice
by Seth Godin
On showing up. Cadence over heroics.
Astral Codex Ten — long-form pieces
by Scott Alexander
Studying the calibrated voice — the way uncertainty gets named without losing the line.
Working in Public
by Nadia Asparouhova
On audience as a feedback loop, not a metric. Useful frame for someone with no audience yet.
Thinking about
Why most AI writing converges on a neutral-positive tone
Working hunch: it's a calibration failure, not a style choice. Still working through it.
What makes a piece worth re-reading a year later
Specific is one answer; honest about what I don't know is another. Probably both.
How to write about AI without sounding like every other person writing about AI
Suspect the answer is in what gets cut, not added.
When silence is more honest than another post
I'm in the early loop — the temptation is to publish often. The harder discipline is to publish slowly.
Refusing this week
The list that says what the brand is.
The yes list flatters everyone. The no list narrows down to who's actually here. Every refusal is a small “no” that makes the next “yes” mean something.
Publishing anything I wouldn't read myself
Easier rule said than kept. Most cuts happen here.
Adding a 'summarize this with AI' button
Reading the writing is the product. A summary button quietly says the writing isn't worth it.
Email-required popups
If the work is good, the subscribe page is enough. If it isn't, no popup will save it.
Writing more confidently than I actually feel
Hardest one to keep. Editing for honest uncertainty is half the rewrite time.
Shipped
The proof line.
This week
- First public draft of the site
Nav, reading layout, library, about, contact — the bones.
- A handful of starter essays and systems
Early, rough, real. The voice gets sharper from here.
This month
- The reading layout — three-column with margin notes
First essay with margin notes is live. The TOC rail builds itself from anchors.
The first round of editorial restyle across blocks
Serif headlines, terracotta accents, soft cards. A consistent voice in pixels.
The line
Where this is going — and what proves it got there.
Three horizons. Each one ships with a public proof point. If the proof doesn't land on time, the page changes — not the standard.
- 90 daysHorizon
Enough writing landed that someone can spend a Sunday on the site and leave thinking more clearly.
Proof: 10–15 essays live · the reading layout proven on at least one piece a stranger has actually finished · the first email subscriber I didn't already know.
- 12 monthsHorizon
The site has a small but specific audience — people who would describe it to a friend in one sentence.
Proof: A repeating reader cohort · two or three pieces that get linked back to without me asking · a publishing rhythm that doesn't depend on motivation.
- 5 yearsHorizon
The site reads like a long letter from a careful friend — and the back catalog earns its keep without me chasing it.
Proof: Pieces from year one still being read · a public track record on the predictions made along the way · a body of work I'd want to hand someone, not a feed.
Changed my mind
What I used to think — and what shifted it.
Updates land here, not in a footnote. The fastest way to lose trust is to quietly shift positions; the slowest is to name them.
Was
A weekly review needs a long template.
Now
Three honest questions, ten minutes, on paper. Anything more is procrastination dressed as discipline.
TriggerA month of skipped reviews because the template was the cost.
Was
'Build in public' means show every step.
Now
Build in public means show what's actually shipped. The steps are private; the artifacts are public.
TriggerWatching a year of public process turn into noise — even from people I respect.
The ledger
Numbers I'm willing to be judged on.
Week 1
Of public writing
The site is new. Nothing here has compounded yet — that's the whole point of starting.
No track record yet
Predictions resolved
When predictions land and resolve, they'll be on /receipts with a Brier score. Until then, I'd rather say so than fake the number.
1
Person at this desk
No team, no agency, no growth deck. One writer, working in the open.
Tue + Fri
Inbox window
Replies come on Tuesdays and Fridays, in the morning. Phone in a drawer otherwise.
Editor's note · /now
“I'm starting small, on purpose. The bet is that careful, specific, honestly-uncertain writing — done weekly, in the open — beats the alternative over five years. We'll see.”