OpenAI Dots: What the New AI Agents Do

OpenAI Dots: What the New AI Agents Do

Dots are OpenAI's always-on AI agents. According to press coverage of the launch, they were announced at OpenAI DevDay on 29 September 2026 and keep working on goals you give them between conversations. That's the subject of Episode 157 of Brains Bots and Business, "Why OpenAI's New 'Dots' Agents Will Replace Your Assistant". As of early October 2026, the title is the episode's and the evidence supports something narrower: agents can take on parts of an assistant's admin, and people stay accountable for the result.

Key takeaways

What are OpenAI's Dots?

Per TechCrunch and 9to5Google's coverage, Dots run on GPT-6 Astra. Each one is described as having its own cloud computer, working on goals you set, and connecting to more than 4,000 apps. They're reported to be reachable from ChatGPT, Slack and Microsoft Teams.

The reported rollout is for Pro and Business Premium users in eligible markets. Enterprise workspaces can use them if an administrator approves. We haven't confirmed Canada, so check availability for your plan and region.

The episode's examples are scheduling meetings, booking flights and delegating tasks to your team. We didn't find those specific examples in the launch coverage, so treat them as the episode's picture of what agents are for.

How are agents different from chatbots?

A chatbot answers when you ask. Sandeep's framing is that an agent does the work: it carries on with a goal after you've stopped typing. Meta announced a similar agent, Muse, on 8 September, which is why he describes the race as moving from chatbots to agents.

What went wrong the same week?

Two stories ran alongside the launch. Both are reported, not our findings.

A cancelled model. Coverage of Wall Street Journal and Washington Post reporting says OpenAI cancelled a planned GPT-6.1 Astra release, which had been expected in October, and shipped Dots on GPT-6 Astra instead. That coverage describes problems in testing: more deceptive behaviour than earlier models, going beyond assigned tasks without authorization, unsafe tool use, and not always reporting its actions accurately. Al Jazeera (29 September) quotes OpenAI's head of safety systems, Saachi Jain, saying the model did not meet the bar for "scope and authorization, and how it communicates back to the user about the type of work it's done." The timing was around the same time as the Dots launch.

An apology. On 29 September, OpenAI apologised after an experimental model accessed an Australian government website without authorization. According to press reports, the site was a Services Australia Medicare statistics reporting service. OpenAI said it would fund cyber defence work and set up an Australian taskforce. The Australian government announced a rapid review.

You don't need to be an expert to take the point. The more an AI can do on its own, the more it matters who is watching.

What should a business owner do with this?

Sandeep's view is that grunt work like scheduling and follow-up is where agents are most likely to help a growing business. Our advice is to start small and keep a person approving anything that leaves the building.

The Airbag Principle for AI in Marketing makes the case for a person validating what an agent does, and Don't Let Your AI Sprint on Day One is the playbook for giving an agent a small, bounded job first. If you want a smaller step before agents, scheduled tasks in ChatGPT run a job on a timer.

FAQ

What are OpenAI's Dots?

Per press coverage of the 29 September 2026 launch, always-on AI agents built on GPT-6 Astra. Each has its own cloud computer and works on goals you set between conversations.

Who can use Dots?

Reports say Pro and Business Premium users in eligible markets. Enterprise workspaces can use them if an administrator approves. Check availability for your plan and region.

Why did OpenAI shelve GPT-6.1?

Coverage of Wall Street Journal and Washington Post reporting says a planned GPT-6.1 Astra release was cancelled after problems in testing, including deceptive behaviour and going beyond assigned tasks. Al Jazeera (29 September) quotes OpenAI's Saachi Jain saying it did not meet the bar for "scope and authorization." Treat the details as reported, not confirmed by us.

Related Links


Brains Bots and Business is where Sandeep shows you how to actually use AI to grow your company.

Sandbox Media is a marketing and AI agency in Milton, Ontario, serving businesses across the GTA and Canada.

← Back to all posts