A 10-step setup for running a seven-role team inside one OpenAI Dot
The author turned the architecture for running seven roles (Triage, Calendar, Money, Scout, Builder, Scribe, Steward) inside a single Dot into a 12-page PDF, stressing permission tiers, stop conditions and queue design.
OpenAI just dropped their features for Dots I collected the architecture for running a seven-role team inside that one Dot into a 12-page PDF here are the 10 steps: step 1 → it is not a chat window with a cron job. own machine, own memory, 4,000 connectors step 2 → learn the four rules: act freely, act if pre-approved, ask, hand off step 3 → "pre-approved" means you asked for that exact action. not that you once said yes step 4 → one band never moves: a password change stays with you, whatever you configure step 5 → write seven jobs before one prompt: Triage, Calendar, Money, Scout, Builder, Scribe, Steward step 6 → every rule in one shape: job, never, always, hand off, stop step 7 → the stop condition is the only line a helpful agent cannot drift past step 8 → connect read before write, by a week. inbox send comes last, months later step 9 → design the queue, not the output. more than a handful a day means the rules are wrong step 10 → name one owner, then diary a date 90 days out: is this still doing real work the result: not an agent that does more, but one you can leave running because you know what it may do without you Send this 12-page PDF to your Dot, then read article below on how to make your Dots team better
This post quotes another post: x.com/vartekxx/status/2105729944047284420