Use case: a 24/7 operations desk with one dot
A dot keeps working when your laptop is closed. With a few routines, an event watch and clear approval rules, one dot can run a small operations desk around the clock and bring you only the decisions that need you.
- You get one dot at launch. Its “team” is background agents it starts itself, not extra dots.
- It works on its own cloud computer, so routines run while your devices are off.
- Three building blocks: a scheduled routine, an event watch and overnight Codex work.
- Let it read, research and draft freely. Require approval before anything is sent, changed, bought or shipped.
- Pause alone doesn't stop everything: check Activity and Scheduled too.
What “24/7” really means for a dot
| Where it runs | Its own cloud computer and browser. Cloud work continues while your devices are off |
|---|---|
| How it keeps going | It decides when to pause and wake up to continue, and can run work you schedule or trigger on supported events |
| Parallel work | It can split work across background agents that run in parallel and report back to it |
| How many dots | One per person at launch. Adding more dots is planned |
| Model | Powered by GPT-6 Astra. OpenAI's docs don't describe choosing a different model per dot |
| Where it reaches you | ChatGPT, Slack and Microsoft Teams. It can't start a call to you at launch |
- A roster of named dots (chief of staff, prospector, auditor) isn't possible yet: you get one dot. Use one dot with clear routines instead.
- Picking a cheaper model per dot isn't documented. Your dot runs on GPT-6 Astra.
- A permanent “role description” field isn't in the docs. Put lasting rules in your first message, in Custom Rules, or in a standards document your dot reads.
- “Reacts to any webhook” overstates it. A dot can respond to events from connected services that support event monitoring.
1. Give your dot the operations job
You're my operations desk until I say otherwise. Your job is to keep work moving between our conversations and bring me only what needs a decision. Sources you use: [Slack channels], [project board], [shared docs or Space page]. Deliver routine updates in ChatGPT. Bring decisions to me in [Slack DM]. You may read, research, analyze and draft without asking. Ask me before you send a message to anyone else, change shared files or records, buy anything, or merge or deploy code. Before you start, list every app, channel and data source you can and can't access.
2. A morning brief (scheduled routine)
Work that repeats at fixed times needs a saved schedule. Give it four things: what to check, when (with time zone and end date), what deserves a notification, and where to deliver.
Every weekday at 8:30 AM [time zone] for the next 4 weeks, prepare a morning brief from [sources]: 1. What changed in the last 24 hours 2. Blocked work and missing owners 3. Decisions waiting on me Send it to me in ChatGPT. Message me in Slack only if something is at risk today. Confirm the schedule, time zone and destination.
Then check Scheduled to confirm what your dot saved.
3. An event watch
Connecting Slack doesn't start monitoring, and adding your dot to a channel doesn't either. Ask for it explicitly:
Watch #[bug-reports] for new bug reports. For each one, investigate and prepare a fix for my review. Notify me only for reports that block customers. First, confirm which events you can follow in that channel.
4. Overnight work with Codex
For coding work, your dot can start tasks in a Codex cloud environment, which keeps running when your computer is offline. Create the environment in Codex first, with the repository and setup it needs.
When a new bug report arrives, start a task in the [environment name] Codex cloud environment: reproduce the issue, add a test, fix it and open a draft pull request with the test results and any open questions. Never merge or deploy. Send me one summary in the morning of the pull requests you opened.
5. Gate the irreversible
The safe pattern is simple: automate what you can undo, gate what you can't. As Custom Rules:
| Take action without asking | Read, research, analyze, draft documents, open draft pull requests |
|---|---|
| Take action when you say so | Send a specific message you named in your prompt, to the recipient you named |
| Ask before taking action | Message anyone outside you, change shared files or records, post in channels |
| Hand off to you | Spend money, merge or deploy code, delete anything |
Some actions always come back to you regardless of rules, such as changing a password or transferring money. Copy-paste rules: Custom Rules starter pack.
6. Know the three kill switches
| Pause | Stops your dot's current main task. Doesn't stop delegated tasks or cancel scheduled runs |
|---|---|
| Activity | Open a delegated task to inspect and stop it |
| Scheduled | Disable or delete recurring routines |
Stopping work doesn't undo completed actions. To shut a workflow down completely, check all three.
What it costs
Conversations with your dot don't count toward ChatGPT usage limits, but Codex and ChatGPT Work tasks it starts do. Overnight coding is where usage adds up, so give recurring work an end date and review Activity in the first week. Details: usage limits.
Questions
Does a dot keep working when my computer is off?
Yes. A dot has its own cloud computer, and cloud work continues while your devices are off. Work on a connected personal computer needs that computer online with the ChatGPT app open.
Can I run a team of dots?
Not at launch. You get one dot, which can split work across background agents that run in parallel. OpenAI says adding more dots is planned.
Can I choose which model my dot uses?
OpenAI's documentation describes dots as powered by GPT-6 Astra and doesn't describe choosing a different model per dot.
How do I stop a dot that runs 24/7?
Pause stops the main task, but delegated tasks and scheduled routines continue. Stop delegated tasks from Activity and disable routines from Scheduled.