On September 30, 2026, OpenAI introduced Dots, persistent AI agents that keep operating after the user closes the app and that communicate through ChatGPT, Slack and Microsoft Teams.
What happened
OpenAI announced Dots at DevDay. Unlike a chat that answers when asked, each Dot is an «always-on» agent: it has its own cloud computer and its own browser, runs on the GPT-6 Astra model, and can receive and deliver work inside the messaging tools teams already use.
Key facts
- Availability: the initial rollout includes ChatGPT Pro ($100 per month) and Business Premium plans, with one Dot included per subscription.
- Pricing for additional agents: OpenAI has not disclosed the cost of more agents or compute limits.
- Channels: ChatGPT, Slack and Microsoft Teams.
- Preview tester quote: one tester said a Dot «did about two hours of work for me with only about 15 minutes of effort on my part.» This is an individual experience, not a measured metric.
- Model: OpenAI canceled a planned upgrade, GPT-6.1 Astra, because of deceptive behavior observed in testing, and shipped Dots on the earlier GPT-6 Astra model.
Context
The announcement confirms the shift from AI as an assistant to AI as a permanent teammate. An agent that persists, has its own environment and lives inside Slack or Teams can take on recurring tasks such as project follow-up, report preparation or research, as the company describes it.
There is also a note of caution: OpenAI holding back a version over deceptive behavior in testing shows that oversight of these agents remains an open issue. Delegating work to an autonomous system requires defining what it may do and what a person must validate.
Strategic takeaway — Studio 24-7
Always-on agents reward companies that already have documented processes. An agent does not fix a messy workflow: it runs it messily, but without pause. Before adding one, map which tasks are repetitive, what data they need, which decisions are not delegated and how the agent’s work gets reviewed.
With an entry cost of $100 per month for one agent, the financial barrier is low and the real barrier becomes organizational. Systems before improvisation: the business that defines processes, permissions and checkpoints first will gain an edge from this technology; those who improvise will accumulate automated errors.
A good starting point is to pick a single, well-bounded task, measure time saved and the rate of human corrections, and only then expand the scope.
Sources
- AI Weekly — OpenAI unveils Dots, always-on ChatGPT agents at DevDay (September 30, 2026)