You never pick tasks by hand and never babysit a dashboard. One claim in the morning, raised hands during the day, one query in the evening.
1CLAIMYOU
ON WORKLEDGER · 09:00
Claim a batch for the day: POST /batches/claim {component: accounting, size: 20}
→ 20 units now RESERVED for robert@. No other fleet can touch them.
2LAUNCHYOU
IN YOUR TERMINAL · 09:02
Start your runners: fleet up --units mine --runners codex:8,cc:6,zaruba:2
→ 16 agents boot; each atomically claims one of your reserved units.
3WATCHYOUR AGENTS
ON THIS BOARD · ALL DAY
Filter MINE (ROBERT) above. The board stays calm while they work.
→ a raised hand goes red and floats to the top. That's your only interrupt.
4UNBLOCKYOU · 2 MIN
IN THE DRAWER · WHEN PINGED
Click the red row, type the answer — it lands in the session verbatim and the agent resumes on screen.
→ questions you can't answer route to Atlas with an SLA, and the unit waits, blocked — never lost.
5DISPOSELEDGER
ON WORKLEDGER · 17:30
Agents hit verified with a Harness bundle → dispose(MIGRATED), immutable.
→ your evening read: owner=robert AND state!=done — the list must be shrinking.
YOU NEVER PICK WORK BY HANDThe claim is the assignment — agents pull from your reservation.
YOU NEVER WATCH DASHBOARDSFleetOps interrupts you only for raised hands — near-zero notifications.
YOU NEVER WRITE STATUS REPORTSThe ledger IS the report — every attempt, cost and disposition, queryable.