OpenAI launches Dots, personal agents that run in the background unsupervised

At a Dev Day event two days ago OpenAI unveiled Dots, agentic assistants powered by GPT-6 Astra and built to operate continuously in the background without tying themselves to any single interface or piece of hardware. Unlike Codex or ChatGPT, Dots take a goal from the user and run with it independently, requiring minimal oversight. The company calls them "agents with exceptional capability, always available, developed to handle everything."

Access opens Tuesday for Pro and Business Premium subscribers in eligible markets. You can spin up a Dot straight from Codex or ChatGPT. Communication happens through Slack, Teams and other enterprise platforms; text-message support is coming soon. Every Dot gets its own identity and can be outfitted with permissions and tools via existing systems.

OpenAI demos two main tracks. A developer points a dedicated Dot at customer feedback — monitoring, fixing bugs, shipping features automatically. A researcher uses one to rerun analyses and flag anomalous results the moment new data lands. Beyond that the company talks about "specialist Dots" that own defined domains, each with its own identity, permissions and toolset.

Integration with Microsoft is already underway, weaving Dots into Agent 365 security controls. The idea is to let organizations run autonomous agents inside a managed environment without surrendering existing permission and compliance policies. It's the first concrete step that plants Dots inside enterprise infrastructure instead of leaving them as a developer plaything.

Most of this functionality was already possible through Codex and similar agentic frameworks. Dots simply packages the pieces into a single bundle that emphasizes sustained autonomous action. OpenAI has not published benchmarks comparing Dots to existing solutions, so the promise of "exceptional capability" remains a statement of intent for now.

The branding leans cartoonish and light: Dots appear as small floating dots, a persona that echoes Meta's recently launched Muse. Turning a language model into a friendly character is part of a broader trend — companies are trying to make AI feel more accessible and "human" at the interface layer, even when the engine under the hood stays the same.