If your team opens ChatGPT before it opens email, you’ve probably wondered the same thing I did when OpenAI took the stage at DevDay: what if the assistant you already use just… kept going after you closed the tab?
That’s the promise of OpenAI Dots.
Most people use ChatGPT one question at a time. Ask, read, copy, paste, repeat. The work only moves when you’re sitting there. Dots are OpenAI’s answer to that bottleneck: agents that hold onto a goal and keep pursuing it in the background.
Here’s what they are as of October 2026, how they work, who should look closely, and where they sit next to the agents I’ve picked for my own life and business.
What OpenAI Dots are
OpenAI announced Dots at DevDay on September 29, 2026. In OpenAI’s words, they’re always-on personal agents, powered by its GPT-6 Astra model.
The core idea in plain English:
- You set the goal. Not one prompt — an ongoing objective.
- A Dot pursues it continuously in the background. It doesn’t stop when you stop typing.
- You start with one primary Dot, and you can name it.
- OpenAI describes specialist Dots and teams of Dots that work together on bigger jobs.
If you’re new to the idea of an agent, here’s the short version: a chatbot answers your question and waits. An agent takes a job and works it. I unpack that difference on AI agents explained, and the spectrum from prompt to workflow to agent on agentic AI.
Where Dots live
This is the part that matters most for owners. As of October 2026:
- Inside ChatGPT, on certain paid plans in eligible markets. Check OpenAI’s current plan page to see if yours qualifies.
- Launchable from ChatGPT or Codex, OpenAI’s coding tool.
- Reachable through Slack, Microsoft Teams and other workplace platforms. You can message a Dot where your team already talks.
- Text messaging has been announced as coming.
That last group is the quiet headline. An agent that shows up in the channel your team already uses gets adopted. An agent that lives in a separate app nobody remembers to open gets ignored.
How Dots work: primary, specialist and teams
Think of it like staffing.
Your primary Dot
This is your first hire. You name it, give it a goal, and it works that goal in the background. Everything starts here, and I’d keep everything here until it’s reliable.
Specialist Dots
OpenAI describes specialist Dots focused on narrower jobs. In a business, that’s the same pattern I use with my own agents: one generalist coordinating, specialists doing focused work. Individual Dots can be given their own identities, credentials and tools, which means you can scope each one to only what its job needs.
Teams of Dots
OpenAI also describes Dots working together as teams. If that sounds familiar, it’s the multi-agent model — an orchestrator plus specialists. I break down how to organize that safely on multi-agent systems.
Security work in progress
OpenAI has said it’s working with Microsoft on Agent 365 security controls. That’s worth watching if your company runs on Microsoft tools — and worth checking against your own IT requirements before you connect anything sensitive.
Who OpenAI Dots fit best
I’m going to be fair here, because Dots aren’t one of my three picks and that’s easy to misread.
My picks, as of October 2026, are Grok Business (with Grok Bot) for business, Muse for personal life, and Claude for building your own agent. Those are picks for my situation, reasoned from published features. They’re not a verdict that anything else is wrong.
Dots may be the most natural choice for you if:
- Your team already works in ChatGPT every day. No new habit to build.
- Your company talks in Slack or Microsoft Teams. You can reach Dots right in those channels.
- You use Codex and want agents launched from the same place.
- You want an agent that shows up in the conversation, not in yet another dashboard.
Where I’d look elsewhere:
- If you need a full team AI workspace with seats, roles, retention settings and a permission-aware Google Drive connector, look at my reasoning for Grok for business.
- If the job is your personal life — errands, travel, bills — and you live in WhatsApp, see Muse.
- If your process is unusual and nothing off the shelf matches it, build your own on Claude.
| OpenAI Dots | Where it shines | |
|---|---|---|
| Starts from | ChatGPT or Codex | Teams already using OpenAI tools |
| Talk to it through | Slack, Microsoft Teams, other workplace platforms; texting announced | Companies that run on chat |
| Structure | Primary Dot, specialist Dots, teams of Dots | Growing from one agent to a small agent team |
| Control | Individual identities, credentials and tools per Dot | Scoping each agent to its own job |
| Availability | Certain paid plans, eligible markets | Check your plan page |
How I’d evaluate Dots on your own work
Don’t decide from a keynote. Run a short, honest test:
- Confirm your ChatGPT plan and market include Dots.
- Write one goal you’d normally carry yourself for a week — a research list, a follow-up queue, a weekly summary.
- Give your primary Dot that goal and only the tools it needs.
- Check in once a day: what did it do, was it right, how much did you fix?
- Only add a specialist Dot after the first one is reliable.
- Run the same brief through any other agent you’re considering and compare.
The full first-week plan is on how to use ChatGPT Dots. If you’re rolling Dots out to a team through Slack or Teams, read Dots for teams first.
Guardrails that apply to Dots (and every agent)
Same rules I run in my own business:
- Least access. A Dot gets the credentials its job needs, nothing more. Giving a Dot its own identity makes this easier — use that.
- Human approval on anything irreversible. Sending to customers, spending money, deleting, signing.
- Don’t paste sensitive or client data into any tool until your settings and policy say it’s approved. My AI policy guide gives you an outline.
- Disclose AI where people would expect to know they’re dealing with one.
- Outreach rules still apply. Calling and texting customers with AI falls under TCPA consent rules, A2P 10DLC registration for business texting, and Do Not Call. Get counsel.
- Verify the output. Trust the process, not the output.
How Dots compare to building your own
My AI chief of staff runs on Claude Code on a dedicated machine. I built it because I wanted an agent shaped exactly around my written processes. The trade-off: I own every instruction, skill and guardrail.
Dots trade some of that control for convenience. They start inside a product your team may already use and meet you where you already talk. For a lot of teams, that convenience is the whole point. Neither path is wrong — they suit different owners.
Your next step
If your team already lives in ChatGPT, Slack or Teams, set up one primary Dot this week with one goal and a daily check-in. If you’re still choosing between agents, compare them on the best AI agents hub, and quick answers live in the OpenAI Dots FAQ. If you’d rather have an agent set up around your real processes, see how Chief works. And I test these tools on real work every week on YouTube.
Questions people ask
What are OpenAI Dots?
Dots are always-on personal agents from OpenAI, announced at DevDay on September 29, 2026 and powered by GPT-6 Astra. You set a goal and a Dot keeps working on it in the background. You start with a primary Dot you can name, and OpenAI describes specialist Dots and teams of Dots working together.
Where do I find Dots?
As of October 2026, Dots are available in ChatGPT on certain paid plans in eligible markets, and you can launch them from ChatGPT or Codex. Check OpenAI’s current plan page for your account.
Are Dots better than Grok or Muse?
They’re a different fit, not a better or worse one. Dots make the most sense for teams already living in ChatGPT, Slack or Microsoft Teams. My personal picks are Grok Business for business and Muse for personal life, and the reasoning is on the best AI agents hub.