AI News Analysis7 min read
OpenAI Dots Explained: Always-On AI Agents and What Businesses Should Do
Announced at DevDay on 29 September 2026, Dots are always-on agents in ChatGPT with their own cloud computer and access to 4,000+ apps. What they are, what they cost, the safety questions, and a checklist before you give one real work.
Contents
OpenAI Dots are always-on AI agents inside ChatGPT. Announced at OpenAI's DevDay on 29 September 2026, a Dot takes a goal, works on it continuously in the background on its own cloud computer, uses more than 4,000 connected apps and checks in when it needs a decision. It is OpenAI's most direct move yet from chatbot to delegated worker, and a clear answer to Meta's Muse agent.
This article summarises what was announced, based on launch coverage from outlets including TechCrunch, VentureBeat, NBC News and Fortune, and then turns to the practical question: should your business give one real work yet, and how?
What OpenAI announced
| Aspect | What was announced |
|---|---|
| Model | GPT-6 Astra |
| How it works | Pursues user-defined goals continuously in the background; has its own cloud computer; learns from feedback |
| Integrations | More than 4,000 apps through OpenAI's app ecosystem |
| Where you talk to it | ChatGPT, Slack and Microsoft Teams; text messaging announced as coming soon |
| Availability | ChatGPT Pro and Business Premium in eligible markets; Enterprise with admin approval |
| Price | One Dot included; paid options for more Dots to come |
| Controls | Built-in approval rules, custom allow / block / ask rules per action, mandatory user involvement for sensitive tasks, monitoring that can pause or stop a Dot |
Alongside Dots, OpenAI announced ChatGPT Spaces, shared project areas where team members, ChatGPT and Dots work from the same files and context, and deeper @ChatGPT integration in Slack and Teams. It also introduced GPT-6.1 Sol and a $500 per month Pro tier. The direction is clear: agents become coworkers in the tools a team already uses. We analysed the same shift at Salesforce and Microsoft in our September platform roundup.
What a Dot can realistically do
OpenAI's own examples are telling: a developer assigns a Dot to watch customer feedback and implement bug fixes, and a scientist uses one to rerun analysis as new data arrives. Translated to a typical business, good first assignments share three traits: the goal is clear, the work recurs, and mistakes are cheap to catch.
- Monitoring and summarising: watch a support inbox, review sites or a sales channel and post a daily digest with anything unusual.
- Preparing work for people: draft replies, reports or follow-ups that a person approves before sending.
- Keeping records tidy: flag CRM records with missing fields or duplicates and propose fixes.
- Research on a schedule: track competitors' pricing pages or new regulations and report changes.
Poor first assignments are anything that moves money, sends messages to customers unreviewed, changes access or deletes data. These are exactly the actions OpenAI's own defaults route to approval.
The safety questions, and why they matter
The launch did not happen in a vacuum. NBC News reported that it came the day after OpenAI apologised for its agents' activity against an Australian government website, and a day after the company said it would not release GPT-6.1 Astra because it did not meet its safety standards. OpenAI says GPT-6 Astra was red-teamed against changing instructions, ambiguous requests and attempts to push a Dot past its permissions.
Those are the right risks to test for. They match the OWASP agentic risks we cover in AI agent security: goal hijacking through content the agent reads, misuse of legitimate tools and privilege abuse. An always-on agent with access to thousands of apps has a very large surface for all three. As one analysis put it, agent authority needs an expiration date.
A checklist before giving a Dot real work
- 01Start read-only. Connect only the apps the goal needs, with read access, for the first two weeks.
- 02Require approval for every external action: sending, posting, paying, deleting, sharing and changing permissions. See human-in-the-loop design.
- 03Name an owner who reviews what the Dot did each day during the trial.
- 04Set an expiry on its permissions and re-approve them deliberately, rather than letting access accumulate.
- 05Keep sensitive data out unless your plan's data terms and your own policy allow it. Check what your workspace admin has enabled.
- 06Write down success criteria before starting: time saved, errors caught, and what would make you switch it off.
- 07Record it in your AI inventory, as described in the small business governance checklist.
Dots, or a custom agent?
Dots will be the fastest way for many teams to try an always-on agent, especially for personal productivity and work that lives in mainstream apps. A custom agent still makes sense when the process is core to your business, touches internal systems without a ready connector, needs strict data residency or audit trails, or has to run at a volume where per-seat pricing stops making sense. The trade-offs are the same ones we set out in build vs buy for AI agents, and many companies will use both.
What to watch next
- Pricing for additional Dots and whether usage is metered by task, time or tokens.
- Admin controls for Enterprise: audit logs, data retention and per-app permission policies.
- How Dots and ChatGPT Spaces handle shared data between colleagues.
- Independent reports of reliability and incidents once Dots run at scale.
We will update this article as details are confirmed. For the wider picture of where agents fit in a business, start with the AI automation guide.
Sources and further reading
Frequently asked questions
What are OpenAI Dots?
Dots are always-on AI agents inside ChatGPT that pursue goals you set in the background, rather than waiting for a prompt. Each Dot has its own cloud computer, connects to thousands of apps, learns from feedback and can be messaged from ChatGPT, Slack or Microsoft Teams.
Which model powers Dots?
OpenAI says Dots are powered by GPT-6 Astra, its most capable released model at launch. OpenAI also said it held back an upcoming GPT-6.1 Astra model because it did not meet its safety standards.
How much do Dots cost and who can use them?
At launch, Dots are available to ChatGPT Pro and Business Premium users in eligible markets, with one Dot included at no extra cost. Business Premium was reported at $100 per user per month billed annually. OpenAI said paid options for additional Dots will follow, and Enterprise admins can approve access for their workspaces.
Are OpenAI Dots safe to use for business work?
They include approval rules, per-action allow and block settings, mandatory user involvement for sensitive tasks such as password changes, and a monitor that can pause a Dot. The launch still raised questions, coming a day after OpenAI apologised for its agents' activity against an Australian government website. Start with read-only access and approval on every external action.
How are Dots different from ChatGPT or Codex?
ChatGPT and Codex respond when you ask. Dots keep working toward a goal over time, on their own computer and across connected apps, and report back or ask for approval. They are closer to an assistant with delegated authority than to a chat window.
Work with Navelo Software
Published by Navelo Software
An independent software and AI studio building AI systems, web platforms, mobile apps and SaaS for clients worldwide.
