WritingJourneyletter
Journeyletter
Field notes on operating AI agents for real work, and letters I draft by hand.
01Two lanes
Two lanes
Two lanes, same masthead. Notes start from the exhaust of real work and get edited. Letters start as my own draft and get structured. Every piece ends with a colophon naming the stack, the agents, and what I did. How it works.
- NoteThe fix was done. The record said it wasn't.A short field note on the quiet failure: real work a system finished but never wrote down.
- LetterEfficiency belongs to the house, not the box.A letter on what two years of thermostat data and a stack of utility bills taught me about the machines in my basement, and why efficiency only makes sense for the whole system.
- LetterApproval was never one thing.A letter on building my own Steward to slowly earn the trust to use it, and what Intercom's real PR-approval experiment taught me about what 'approval' actually means.
- LetterSomeone has to be the squeaky wheel.A letter on why I'm building a dedicated skeptic into my agent workflow — and what happens when nobody's assigned to doubt the green checkmark.
- NoteA button is a claim.Field notes on agent legibility — or, what I saw the first time my own honest interface asked me for seven decisions.
- NoteI couldn't answer four questions about my own button.Field notes on agent legibility — or, the day I tapped a control I built and had no idea what I was executing.
- NoteAll of my agents write memos. None of them act.Field notes from handing the visionary role to an AI — and what it saw in my system that I didn't.