Writing
Notes, essays, and things worth writing down.
- 10 / 2026
Review real AI output before you automate approval
Two extraction mistakes shaped our AI review workflow. What we automated, what stayed with people, and how I'd test the approvals next.
- 09 / 2026
Build your agentic workflow around the work the prompt leaves behind
Turning a prompt into an agentic workflow starts with the work people still do around it. Here's how we found and removed those handoffs.
- 06 / 2026
AI tooling is company infrastructure. Most teams still run it like personal dotfiles.
Every employee runs their own local Claude. Anthropic shipped the shared install; the plugin repo to put through it is the part most teams skip.
- 06 / 2026
Your AI reviewer shares your AI coder's blind spots
A Claude review proposed a bug. A second Claude review would have approved it. Same model family, same blind spots, so I review with a different one.
- 05 / 2026
AI gives you consistent UI only after you build the system it copies from
AI doesn't drift because it's careless. Each session starts blank. A component library, a CI drift gate, and a catalog of blessed exceptions hold the line.
- 05 / 2026
Hand every AI you use the same name tag
Every new chat, you re-introduce yourself to your AI. Four plain files in ~/.me/ that any tool reads. One write, many readers, no vendor lock-in.