Designing a single Dots agent as a 24/7 work system
The author uses one GPT-6 Astra dot to coordinate specialized tasks, building the architecture in five layers: responsibility, permissions, triggers, verification, and delivery.
OpenAI just dropped dots, and I mapped one into a 24/7 operating system for real work instead of pretending five dots are already a company, I designed one primary dot to coordinate specialized tasks, keep context between conversations, and bring only the decisions that actually need a human back to you here's how the architecture works: 1 → Responsibility defines the outcome it owns, then Sources limit which apps, folders, channels, and systems it can trust 2 → Permissions separate research and drafting from actions that change data, contact people, publish content, spend money, or delete anything 3 → Triggers start the right work on a schedule or supported event, while focused Work and Codex tasks handle research, analysis, browser work, and implementation 4 → Verification forces every specialist to return evidence, test results, unresolved questions, and the exact actions still waiting for approval 5 → Delivery sends you one clean decision packet instead of five chats, twenty notifications, and another pile of AI output to manually untangle that’s one GPT-6 Astra dot coordinating the work of an agent team without pretending native multi-dot companies already exist full architecture, job contracts, approval rules, and rollout checklist in the article below ↓
This post quotes another post: x.com/leiroops/status/2107182232737575194