On September 29, 2026, OpenAI introduced Dots, a new kind of AI agent that stays running in the background instead of waiting for you to open a chat. The core distinction matters: a dot is not a ChatGPT conversation. It is an always-on agent with its own cloud computer, connected to the apps and accounts you choose, that can act on its own between the moments you talk to it.
If you're deciding whether to try one at work, here is what Dots actually are, what they can do, who can currently get one, and the caveats worth knowing before you connect anything to your real accounts.
A Dot Is a Cloud Computer, Not a Chat Window
A standard ChatGPT conversation is generally driven by your prompts: you ask for something, ChatGPT responds, and you initiate the next step. A dot is different. It is an always-on agent designed to take on ongoing responsibility and keep making progress between conversations. According to OpenAI's announcement, each dot runs on its own persistent cloud computer, a dedicated environment that keeps running even when you're not looking at it. OpenAI says this environment is powered by its newest model, GPT-6 Astra, and it is designed to keep context, keep working, and learn about how you operate over days and weeks rather than a single session.
That persistence is the whole point. A normal chat interaction generally depends on you to restart the work and supply context for the next step. A dot is designed to keep context, check back in, and make progress without being re-prompted every time.
What a Dot Can Actually Do
Per OpenAI's announcement, a dot's capabilities center on four pieces:
- Connected apps. You link specific accounts and services, such as email, calendars, or project tools, and the dot can read from and act within them.
- Memory. The dot is designed to learn your preferences over time, so repeated tasks need less explanation with each round.
- Scheduled and proactive work. A dot can run reminders and recurring tasks you have scheduled. It can also proactively review information from connected apps and suggest ways to help without a new prompt. OpenAI says this background “proactive research” uses read-only tools that cannot send messages, modify app content, or control your browser or computer.
- Optional local-computer access. Beyond the cloud environment, OpenAI's help documentation says you can optionally grant a dot access to your own local machine, expanding what it can touch beyond cloud-connected services alone.
OpenAI's announcement also says dots work across ChatGPT, Slack, and Teams, so the same agent can show up in the tools your team already uses rather than living only inside a separate app.
Who Can Actually Get One Right Now
This is the part most newsjacking coverage glosses over, and it's where the specifics matter most if you're deciding whether to try this at work. According to OpenAI's help documentation, Dots are rolling out gradually, not launching to everyone at once:
- Pro subscribers are eligible in markets excluding the EEA, Switzerland, and the UK at launch.
- Business Premium users are eligible across all supported ChatGPT regions.
- Enterprise users, including Edu and Healthcare, can try Dots in beta when their workspace admin enables the feature; it is initially off by default.
Dots are rolling out gradually, and access may take several days to reach an eligible account. For Pro and Business Premium users, the first dot is included at no additional cost, although OpenAI applies usage allowances for deeper work. If you don't see Dots in your account, that may be your plan, region, or rollout timing rather than a bug.
The Controls: Custom Rules, Activity Review, Pause, and Reset
Because a dot acts without constant supervision, OpenAI's help documentation describes several controls meant to keep it inside defined boundaries:
- Custom Rules let you set explicit instructions about what the dot can do on its own versus what needs your approval first.
- Activity review gives you a log of what the dot has actually done, so you can audit its actions rather than guessing.
- Pause stops the dot from taking further action without erasing anything, useful if you want to freeze things temporarily.
- Reset is permanent and broad: OpenAI's help documentation states that resetting a dot deletes the dot itself along with its conversations, saved memories, and scheduled tasks. There's no partial undo once you reset.
One caveat is easy to miss. Disconnecting a connected app stops the dot from pulling new information from it going forward, but it does not delete information the dot already obtained while the app was connected. If a dot read sensitive data before you revoked access, that data isn't automatically wiped by the disconnect itself. A full reset is the step that clears stored information.
What to Watch Out For
OpenAI's own materials are direct about limits. Agents like this can make mistakes, and OpenAI recommends reviewing any consequential output, meaning anything with real financial, legal, or reputational weight, before treating it as final. Autonomy is the feature, but it's also the reason to check the work.
OpenAI's announcement previews specialist dots, narrower agents tuned for specific roles or tasks. OpenAI calls them a preview and says it is starting with focused enterprise pilots. They are not a broadly available product, so treat any specialist dot you encounter as early and limited rather than a finished offering.
The Takeaway
Dots represent a real shift from conversational AI to persistent AI: an agent with its own cloud computer, ongoing memory, and the ability to act between your prompts. That's a meaningfully different trust relationship than a chat window, which is exactly why access is rolling out gradually and why OpenAI ships explicit controls alongside it. Treat this as an early-stage capability worth testing deliberately, not a tool to hand broad access to on day one.
This week's checklist
- Start with narrow permissions. Grant the minimum access a dot needs for one task, not your whole app ecosystem.
- Review every connected app before linking it, and understand what the dot can read and act on through it.
- Make approval rules explicit using Custom Rules rather than assuming the dot will ask before acting.
- Check the activity view regularly so you know what the dot has actually done, not just what you asked it to do.
- Have a reset and recovery plan in place before you need it, since reset deletes the dot, its conversations, memories, and scheduled tasks for good.
If Dots have you thinking about how to adopt AI agents without giving up oversight, the AI Governance course is a practical next step for learning how to set clear accountability, controls, and responsible-use practices around AI.*