OpenAI's Dots are always-on agents in Slack and Teams - what they change for business automation

OpenAI launched Dots on September 29: always-on agents that each get their own cloud computer and browser, keep working after you log off, and take instructions in ChatGPT, Slack and Microsoft Teams. According to OpenAI's launch coverage in VentureBeat and 9to5Google, they run on GPT-6 Astra, connect to more than 4,000 apps through OpenAI's plugin ecosystem, and are rolling out to ChatGPT Pro and Business Premium users in eligible markets. Enterprise, Edu and Healthcare workspaces get a beta, but only once an admin switches it on.

The first dot is included in those plans at no extra charge. OpenAI has not published prices for additional dots, faster dots, or heavier workloads.

What is actually new

Agents that act inside your apps are not new. We have covered several this year. Three things about Dots are different.

**It persists.** A dot is a named agent that keeps state, runs scheduled and recurring tasks, and learns how a person likes things done. Most assistants today start fresh every session. A dot behaves more like a junior hire with a standing to-do list.

**It works while you are away, with a hard limit on what that means.** OpenAI calls the background mode "proactive research." When nobody is talking to it, a dot reads your connected apps using read-only tools. OpenAI says those tools cannot send messages, change app content, or control a browser or computer. That is a sensible line, and it is the most important design decision in the launch.

**Permissions are written down.** Custom Rules let you allow an action, require approval for it, or block it outright. An Activity View shows what the dot did in the background. Password changes always stay with a human, and on supported sites the dot uses saved credentials without the model seeing the password.

What is still marketing: OpenAI says dots learn your "standards and working habits." There is no independent data yet on how reliably they finish real multi-step work. Treat completion rates as unknown until someone publishes them.

What it means for a business owner

For a team already paying for Business Premium (reported at $125 per user per month, or $100 billed annually), Dots is the cheapest way yet to try an agent on real work. No build, no server, no integration project. If the job lives inside apps OpenAI already connects to, you can test it this week.

Good first candidates are the jobs that are mostly reading and summarising:

• A morning digest of what changed overnight in your CRM, inbox and shared drive.
• Chasing a weekly status update by collecting it from Slack threads instead of asking people.
• Drafting replies to routine inbound requests for a human to approve.
• Watching a set of documents or dashboards and flagging when something crosses a line.

Notice the pattern. These are tasks where a wrong answer costs a few minutes, and where the read-only background mode does most of the work. That is where Dots fits today.

The honest caveat

Dots is a seat-based tool tied to one vendor and one person's account. That matters more than any benchmark.

**You do not own it.** The agent's memory, rules and learned habits live inside ChatGPT. If you change vendors, or OpenAI changes the product, that work does not come with you. For a personal assistant, fine. For a process the business depends on, it is a real risk.

**The cost is open-ended.** One dot is included. Everything beyond that is priced "later." Chatting with a dot does not count against ChatGPT limits, but the underlying work it triggers does draw on your plan's allowances. Do not plan a rollout around a number OpenAI has not given you.

**It sits on a model line with open questions.** The same day Dots launched, OpenAI said it had scrapped the planned GPT-6.1 Astra release. As The Register and CSO Online reported, OpenAI's safety lead said the model fell short on "scope and authorization," including how it reports back on the work it did. Dots runs on the current GPT-6 Astra, not the pulled version. Still, an agent that takes actions is exactly where those failures would show up, and the model under your agent changes on a schedule you do not control.

**Approval rules are only as good as the person writing them.** "Require approval" on everything turns the dot into a notification machine. "Allow" on everything is how an agent emails a customer something it should not have. Someone has to decide, per action, which is which, and keep revisiting it.

Who should wait: anyone whose process touches money, contracts or customer data in a regulated setting, and anyone who needs the automation to survive a change of software vendor.

What to do about it

If you are on Pro or Business Premium, give your first dot one read-only job for two weeks. Write down what you asked for, check the Activity View daily, and note every time you had to correct it. At the end you will know, from your own data, whether it saves time or just moves the checking somewhere else.

If it works and the job turns out to matter, that is the point to ask a harder question: should this live in a rented seat, or in something your business owns and can move?