What happened
At its DevDay conference on September 29, 2026, OpenAI announced dots, which it describes as always-on agents that work on your behalf around the clock. Each dot runs on its own cloud computer with its own browser, learns from feedback over time, and can connect to more than 4,000 apps through OpenAI's plugins. You can reach yours in ChatGPT, Slack, and Teams, and you can open its computer at any time to inspect what it is doing. Dots are powered by GPT-6 Astra. OpenAI says they are rolling out to Pro and Business Premium plans in eligible markets, and that Enterprise workspaces can try a beta once an administrator enables it. Your first dot is included in the plan, and conversations with it do not count toward your usage limits. Tasks it starts in Codex or ChatGPT Work do count. OpenAI also previewed specialist dots for organizations: agents with their own identity, credentials, and access to a company's systems, aimed at defined jobs such as procurement, invoice processing, customer support, and contracting. It says it is starting with focused enterprise pilots and working with Microsoft to bring them into Agent 365.
Why it matters for business owners
A chatbot answers when you ask. A dot keeps going. When you are not actively working with it, OpenAI says it looks for ways to help in the background, which it calls proactive research, using the apps you have connected through tools restricted to read-only. The rules for what it may do on its own, what needs your approval, and what always stays with you are set through built-in rules and Custom Rules. That changes the ownership question inside a company. Until now, an AI tool sat behind a person: someone typed, someone read the answer. An agent with a login, a browser, and a connection to Slack or Teams behaves more like a new team member with credentials. Someone has to decide it may exist, and someone has to be responsible for what it does. It also changes cost. OpenAI says your plan includes an allowance for deeper work, with extended limits for the first month after launch, and that in future you will be able to add more dots or raise how much work each one takes on. OpenAI has not published what background work will consume, so a budget built on today's terms is a guess.
What owners should not misunderstand
Do not read this as a finished replacement for a person. OpenAI's own page says dots can still make mistakes and that consequential work should always be reviewed. The examples it gives end with a person approving: a drafted invoice sent after approval, pull requests for a developer to review, social posts drafted for approval. Do not assume the safeguards are your controls. Auto-review, read-only background access, and pause-on-concern monitoring are OpenAI's settings on OpenAI's platform. They are useful, and they are also the vendor's design. They do not know your customer list, your contracts, or which of your folders should never be shared. Only your own rules can do that. Do not assume one person's setup stays one person's problem. A dot connected in a shared channel or to a shared mailbox touches other people's information, not just its owner's. And a plan that includes a dot by default can put one in front of your team before you have decided anything about it. Finally, this is one launch from one vendor. Meta and Google have launched their own always-on assistants this year, and the terms, controls, and data handling differ. Judging the category by one product's demo is how owners end up with a tool they never chose deliberately.
The operational lesson
Treat an always-on agent the way you would treat a new hire who starts with a laptop, a badge, and access to shared accounts. You would not hand over everything on day one. You would decide what the job is, what systems it needs, who supervises it, and how you would know if something went wrong. The practical version is four questions for every agent, from any vendor. What is it for, in one sentence? What can it read, and what can it change? Who approves the actions that matter? And how do you turn it off, and see what it did, if you need to? If you cannot answer those in writing, the agent is not ready to be connected. That is a normal result at this stage, not a failure. Most companies have not yet written down which tasks they would even hand to an agent.
What a serious business should do next
First, find out whether anyone already has access. Check whether your ChatGPT plan includes dots, whether an administrator has enabled the Enterprise beta, and whether employees are using personal Pro accounts for work. The last case is the one most owners miss. Second, set a short interim rule. For example: no agent connects to customer data, financial systems, or shared mailboxes until it has been approved for a specific task. A rule this simple is easier to follow than a long policy. Third, pick one narrow, low-risk task to test, such as drafting internal summaries from a single read-only source. Keep a written record of what it was allowed to touch and check its output against the source. Fourth, hold off on committing budget. Usage terms are still changing, and the useful question is which task is worth handing over, not which plan to buy.
The Atlacis view
Atlacis helps owners slow down before an AI decision, map the workflow, decide what a tool should and should not be allowed to touch, and choose based on the business instead of the announcement. Always-on agents are a good example. The technology may be useful. The value depends on choosing the right task, setting the right boundaries, and knowing who reviews the result. If you are trying to decide whether agents belong in your business yet, and where they would sit in your workflow, that is a good conversation to have before anyone connects one to company systems.
The short version
- On September 29, 2026, OpenAI launched dots, always-on agents with their own cloud computer that connect to more than 4,000 apps and can be reached in ChatGPT, Slack, and Teams.
- Dots are rolling out to Pro and Business Premium plans in eligible markets, with an Enterprise beta that a workspace administrator has to enable.
- OpenAI says dots can still make mistakes and that consequential work should be reviewed. Its own examples end with a person approving.
- The safeguards, including auto-review and Custom Rules, are the vendor's settings. They do not replace your own decision about what an agent may read and change.
- Before anyone connects an agent to company tools, decide its purpose, its access, who approves its actions, and how to turn it off.
Where ATLACIS can help
Sources
- OpenAI: Introducing dots (September 29, 2026)
- Axios: OpenAI's new agents put safety promises to the test (Ina Fried, September 30, 2026)
- 9to5Google: OpenAI launches Dots, new always-on agents you can assign tasks to (Ben Schoon, September 29, 2026)
- WIRED: OpenAI's Dots Are Always-On AI Agents, and Its Answer to Meta's Muse (Reece Rogers, September 29, 2026)