Building an agent team.
Being the loop.

Every agent gets three short documents, like a new hire.

Who it is. What finished work looks like. What it must never do.

it’s onboarding, not code. if you’ve managed one person, you already know how.

Week ten must be smarter than week one.

A brain file read before every run. A memory file updated after. One rule: what proves true again and again moves from memory into the brain.

an agent without memory is a permanent intern.

A task returns questions. Ownership returns work.

This is where I lost a month: tasks, instructions, approvals — and everything came back to me. It started working the day each agent owned one corner and I got out of the way.

the test: if you’re still everyone’s bottleneck, you handed out tasks.

1. Start from what only you know.
   one page before any prompt: scars, rooms, years. feed in nothing — nothing is what multiplies.
2. Give ownership, not tasks.
   who you are, what the standard is, what you never do — then get out of the way.
3. Keep the decisions, not the keystrokes.
   you’re not a station in the middle. you’re the start and the end of every run.
4. Never accept “fine.”
   your taste is the asset. send it back, and write down what your bar turned out to be.
5. Close the loop every day.
   what worked goes back into the files. a team that learns from you multiplies; one that doesn’t just automates.
6. Measure multiplication, not output.
   not “how many tasks closed” — “how far did what only I know travel today?”

none of the six mentions a tool or a model. tools rotate. the loop stays.