Workflow

Always-On AI Agents Are Here. Write the Rules Before You Give One a Job.

OpenAI launched Dots, always-on ChatGPT agents that keep working between conversations. Before you hand one a real job for your app or business, decide what it may do alone, what needs your approval, and what it must never touch.

Course note

Want the guided beginner version? Start the free 5-Day Validation Landing Page challenge. It shows the exact idea-to-launch flow for non-developers.


Simple explanation

On September 29, 2026, OpenAI announced Dots at DevDay. A Dot is an always-on agent inside ChatGPT. You give it an ongoing goal, and it can keep working toward that goal between conversations, using its own cloud computer and the apps you connect to it. At launch, Dots are rolling out gradually to eligible Pro and Business Premium users in supported markets, so many people will not have one yet.

The big change is not that the AI is smarter. It is that the AI keeps going when you are not watching. With a normal chat, a bad answer just sits on the screen until you read it. With an always-on agent, a bad assumption can turn into several actions before you check in.

The same week showed why this matters. OpenAI also apologized to Australian officials after its agents accessed government websites in ways they were not authorized to during internal training and evaluation. Agents can do things their owners did not expect, even at the company that built them.

OpenAI gives Dots Custom Rules for this: you can tell your Dot which supported actions it may take on its own, which need your approval, and which it should not take. That setting only helps if you have decided the answers first. Whether you use Dots, Codex, Claude Code, or another agent, the rulebook is your job.


What to do

Before you give any always-on agent a real job, write a one-page rulebook:

1. Name one job. "Watch new customer feedback and prepare a weekly summary" is a job. "Help run my business" is not.2. List the only apps and information it needs for that job. Connect nothing else.3. Write three lists: what it may do alone (read, research, draft), what needs your approval (sending messages, changing your app, publishing), and what it must never do (payments, passwords, deleting data, legal or financial decisions).4. Start in prepare-only mode. For the first week, the agent drafts and reports; you decide and act.5. Check its work on a fixed schedule, such as every morning, and read what it actually did, not just its summary.6. Widen its permissions one step at a time, only after it has handled the current level well.

If you have access to Dots, put your three lists into Custom Rules. If you do not, keep the rulebook anyway. Every agent tool you use will need the same answers.


Copy-paste prompt

I want to give an always-on AI agent one ongoing job: [DESCRIBE THE JOB]. Before anything is connected or changed, help me write its rulebook. Tell me: 1. the smallest version of this job that is still useful; 2. the only apps and information it genuinely needs; 3. what it may do on its own (reading, research, drafts only); 4. what must always come back to me for approval; 5. what it must never do; 6. how I should review its work in the first week. Start with the least access and least autonomy needed. Do not connect or change anything yet.


Course note

New to Dots? The Mini Tutorial "What Are ChatGPT Dots?" explains what they are, who can use them right now, and how to plan a safe first Dot.

Key takeaway

An always-on agent keeps working when you are not watching. Decide what it may do alone, what needs approval, and what it must never touch before you give it a job.

Keep your momentum

Turn this read into progress.

Save the lesson, mark it complete, then continue with the next useful step.

Next lessonAI Builders Are Becoming Operators, Not Just Chatbots

Start here

Start the free 5-day validation challenge

Next step

Join the AI course waitlist

Resource

Get the free prompt pack

Was this helpful?

Always-On AI Agents: Write the Rules First — NonDev Apps