// think
Notes from the work.
Long-form thinking on enterprise UX, building with AI, and what it takes to ship solo, written from the work rather than about it.
an accountable AI system/
A series from designing, building, and operating a personal AI agent system, in reading order. More arrive as the build produces them.
01systems storyJul 27, 2026 · 5 min02systems storyJul 27, 2026 · 4 min03systems storyAug 5, 2026 · 5 min04systems storyAug 5, 2026 · 4 min05systems storyAug 6, 2026 · 7 min06systems storyAug 6, 2026 · 8 min
The part of AI I couldn’t design from the outside
Two months’ notice, a boxed Mac Mini, and the gap between designing AI and being on the hook for one.
Four ways my agent lost its provider
Four outages, one provider, four different fixes. What operating an agent teaches that building one doesn’t.
What I’m building on a headless Mac Mini
An always-on agent team on a Mac Mini with no screen: what the box is for, and what I refused to build.
Building with a paper trail
Four living documents run this project, and their first readers are the agents.
Two agents and a channel
A concierge that pushed back on half its own brief, and what a staff of two actually changed.
Every green light was honest
Four status checks said everything was fine. All four were telling the truth.
notes/
01product storySep 20, 2026 · 6 min02process storyJun 28, 2026 · 6 min03product storyJun 16, 2026 · 5 min
530 tabs and I still hadn't picked an animation style
Twenty-one windows, one unfinished decision, and why I upgraded the tool instead of spending two days sorting tabs.
How I build alone now
What “built with AI” actually means when one person ships the whole product.
Why I built a tab manager that can’t phone home
Six hundred tabs, one refusal, and a Chrome extension now live on the Chrome Web Store.