OpenAI Launches Dots, Always-On AI Agents Powered by GPT-6 Astra

OpenAI introduced Dots, a new category of "always-on agents" that users can assign background tasks to, at DevDay 2026. The feature launches today for ChatGPT Pro and Business Premium subscribers, with Enterprise access pending workspace admin approval.

Dots are powered by GPT-6 Astra and, according to OpenAI, each one runs on "their own cloud computer." Users can interact with a Dot through ChatGPT, Slack, or Microsoft Teams once it's configured. OpenAI describes Dots as agents that "learn from feedback over time" and work toward assigned goals continuously, connecting to what the company claims is an ecosystem of more than 4,000 apps through plugins.

How Dots work

According to OpenAI, Dots can be reached through voice calls, text conversation, or asynchronous check-ins, and they refine their behavior based on ongoing interaction with the user. The company says it uses Dots internally for tasks such as automatically investigating bugs flagged in Slack, generating working app prototypes when new designs are ready, and — in one cited example — noticing on its own that an invoice hadn't been sent and drafting it for approval.

When not actively executing a task, OpenAI says Dots perform "proactive research" using read-only access to connected tools. Users can view a Dot's cloud computer in real time while it works and can grant it permission to use their own machine directly. OpenAI states that Dots inherit the same permission and access controls as standard ChatGPT, and support custom rules dictating when a Dot can act autonomously versus when it must seek approval first.

Each qualifying subscription plan includes one Dot, with OpenAI saying additional Dots will become available to add later — pricing for extra Dots was not disclosed. Dots must be created through the ChatGPT desktop app, though once set up, they can be managed and interacted with via mobile.

Part of a larger DevDay announcement slate

Dots arrived alongside a wide set of DevDay 2026 announcements: GPT-6.1 Sol, pitched by OpenAI as "near-Astra intelligence for a fifth of the price"; a new $500/month ChatGPT Pro tier; a forthcoming "GPT-6 Astra Ultrafast" model; a "Private Intelligence" offering for businesses expected this fall; cloud execution for Codex; updates to the Codex CLI; a new Decisions API; an Agents API with Computer Use support; Bedrock Managed Agents; new ChatGPT plugins; a "ChatGPT Space" for human-agent collaboration; and native @ChatGPT integration inside Slack and Microsoft Teams.

The launch comes just after OpenAI reportedly canceled a GPT-6.1 Astra release over unspecified misbehavior and safety concerns, and in the same week that Meta pushed its own autonomous "Muse" avatar agent, and Anthropic and OpenAI both shipped lower-cost model tiers (Claude Opus 5.5 and GPT-6 Sol/Luna, respectively).

What this means

Dots signal OpenAI's move toward persistent, semi-autonomous agents rather than single-session chat interactions — a direct answer to Meta's Muse and part of a broader industry shift toward agents with standing permissions and long-running memory. The gating to Pro ($200+/month tier implied) and Business Premium plans, plus the new $500/month Pro tier announced the same day, suggests OpenAI is positioning always-on agency as a premium capability rather than a mass-market feature. The real test will be trust: giving an agent its own cloud computer and read access to thousands of connected apps raises the stakes on permission controls, and OpenAI's recent cancellation of a GPT-6.1 Astra release over safety concerns suggests the company is still calibrating how much autonomy to grant these systems before wider rollout.