What Are OpenAI Dots? Always-On Agents, Explained

What OpenAI announced

TechCrunch reported that OpenAI announced Dots at its DevDay event on September 29, 2026, describing them as remarkably capable, always-on agents built to handle everything. Unlike ChatGPT or Codex sessions, Dots are designed to operate independently of a specific interface, pursuing user-defined goals continuously in the background. They run on OpenAI's GPT-6 Astra model and became available that day in ChatGPT for Pro and Business Premium users in eligible markets.

Users interact with them through ChatGPT and Codex, and through workplace messaging such as Slack and Microsoft Teams, with text message support described as coming soon. TechCrunch noted the agents' cartoonish persona, comparing it with Meta's Muse agent.

How they work, according to reporting

VentureBeat described Dots as persistent agents that keep working after an employee closes the chat window: monitoring projects, using software, responding to changing information, and bringing completed work back for approval. It reported that each dot receives its own cloud computer and browser, can connect through OpenAI's plugin ecosystem to more than 4,000 applications, and gradually learns a person's preferences, standards, and working habits.

OpenAI also introduced ChatGPT Space, which VentureBeat described as a collaborative layer where employees, ChatGPT, Codex, and Dots work against the same shared context, with pages and files at launch and presentations and spreadsheets announced as coming.

Controls and pricing

According to VentureBeat, OpenAI introduced Custom Rules that can permit particular actions, require approval, or prohibit them, along with an Activity View for oversight, and an integration with Microsoft Agent 365 for enterprise governance that was described as still in development. Some sensitive operations, including changing passwords, remain reserved for the human user. The first dot is included at no additional charge for Pro and Business Premium customers; pricing for additional dots and enterprise deployment had not been disclosed at launch.

What an always-on agent needs to work well

A chat assistant fails visibly: you see the bad answer. A background agent fails quietly, acting on stale information or the wrong assumption while nobody is watching. Three things decide whether persistent agents help or create cleanup work.

Context. An agent monitoring customer feedback or a project needs accurate, current knowledge of products, policies, and past decisions. Preferences learned over time help, but organisational knowledge should come from maintained sources it can retrieve, not from what it happens to have picked up.

Cost. Agents that run continuously make many model calls. Repeated reading of the same documents, long contexts, and regenerating answers that already exist add up. Retrieval that supplies only relevant passages, and reuse of stored answers, keep a background agent affordable.

Permissions. An agent with a browser, credentials, and thousands of possible integrations needs narrow access, approval rules for consequential actions, and an audit trail. The Custom Rules and Activity View OpenAI described are the right kind of control; they work only if teams configure them deliberately and review them as the agent's scope grows.

How teams should evaluate them

Start with one bounded, recurring task with a clear output, such as triaging incoming feedback into a weekly summary, rather than an open-ended mandate. Grant only the minimum integrations and credentials it needs. Require approval for anything external or irreversible. Review the activity log for the first weeks, and measure whether the work is correct and actually saves time. Expand scope only when the record supports it, and document what each dot is allowed to do so the whole team knows.