How I split an AI assistant into 6 seats so nothing sends without a human yes

Описание и применение

Автор разделил AI‑ассистента на шесть специализированных «сидений», каждое из которых выполняет одну задачу, имеет список запрещённых действий и требует явного одобрения человека перед любыми внешними действиями.

Задача

A single general AI assistant can cause scope creep, unclear authority, and accidental actions such as sending emails or making payments.

Результат

The six-seat structure confines each AI agent to a specific role, enforces explicit human approval for any outbound action, and reduces scope creep and accidental commitments.

Рабочий процесс

• Create six agents named after the seats (Chief, Inbox, Research, Builder, Money, Tuner).
• Paste each seat’s description into the agent’s profile field.
• Interact with Chief first to run an onboarding checklist with a “do not contact anyone” instruction.
• Approve one job and let the owner seat deliver the result inside the chat.
• Review any slow or unclear steps and adjust the seat text; later use Tuner to refine.

Доказательства

I use six standing seats, each with one job, a written "never" list, and a human gate on anything that leaves the chat.

Источник

The rule I like most is the third. It is easy for an assistant to fill a gap with a plausible number. Telling it to stop when the source is missing makes the gap visible, which is what you want.

Источник

Create six agents (or reuse empty ones) and name them after the seats.

Источник