L04 · Writing

How I work with agents, measured.

Long reads on the system I built: what it does, how I keep it honest, and the habits that let one person steer twenty agents. Real numbers from my own work, written for anyone who uses AI.

  1. 01ExperimentThe 206× experimentI went through 35 days of my own logs, every second I spoke, every word I typed and every click I gave my AI agents. Then three blind AI estimators priced the work they shipped. One hour of my instructions produced about 206 hours of skilled work.10 min
  2. 02SystemHelm: the task system my agents runEvery project I take on becomes a card with an end condition. Slack, email and my own words feed it, one AI agent owns each active card, and a single page puts what needs me on top.8 min
  3. 03SkillThe art of context switchingOn a typical day I steer 27 separate jobs across three businesses and move between them about once a minute. Here's how I get my bearings in nine words.8 min
  4. 04ToolsThe signed-in browser: how my agents work behind the loginMost of what my agents need sits behind a login. So they work in my browser, with my sessions and a password vault they can use but never read, and they take requests from my phone.10 min
  5. 05MethodTeach your AI onceI don't correct my agents twice. Every lesson goes into a rulebook my agents read before my first word, or a playbook they load when the job comes up. A lesson usually becomes a rule in under a minute, and the mistake stops.7 min