OpenAI Dots: Always-On Enterprise Agents Are Live
The way enterprises interact with AI changed on September 29, 2026. At DevDay in San Francisco, OpenAI launched Dots: a class of always-on AI agents that do not wait for a prompt. Each Dot runs continuously on its own cloud computer, holds memory across sessions, and connects to thousands of applications. The shift from reactive assistant to proactive agent is no longer a roadmap item. It is a live product.
Dots are powered by GPT-6 Astra and available immediately for ChatGPT Pro and Business Premium subscribers. Enterprise workspaces get beta access after an administrator enables the feature. The first Dot is included with no additional charge.
What Dots Are, Technically
Every Dot gets dedicated infrastructure: its own cloud computer and its own browser, fully isolated from the user’s machine. This separation is deliberate. Tasks continue whether or not you are logged in. When you return, the Dot’s Activity View lets you watch what happened, redirect ongoing work, or hand back anything that needs a human decision.
Dots maintain context across ChatGPT, Slack, and Microsoft Teams. They can be reached by message, voice call, or via iMessage and Android RCS messaging after setup for Pro users. A user starts with one primary Dot. OpenAI says multiple Dots will be available later.
Connectivity runs through ChatGPT’s plugin ecosystem: 4,000+ apps at launch. A Dot can coordinate tasks across Codex, ChatGPT Work, and connected business tools from a single persistent session.
Inside OpenAI, Dots are already in production. Engineers are using them to fix dozens of bugs per day. Other teams apply Dots to procurement, invoice processing, email marketing, customer support, and commercial contracting.
ChatGPT Space: The Collaborative Layer
Alongside Dots, OpenAI introduced ChatGPT Space: a shared workspace where people and agents work together. Space functions as a team channel. Employees can bring Dots into group chats, assign tasks, and tag agents the way they would tag a human teammate. Multiple people can see what an agent is working on, add work, and redirect it.
Space is available for Business Premium and Enterprise users. For enterprise teams already running ChatGPT Work as part of their daily stack, Space adds coordination without requiring a new platform. The agent layer and the collaboration layer share the same infrastructure.
Specialist Dots for Enterprise
For organizations, OpenAI previewed Specialist Dots: role-specific agents with their own identities, dedicated credentials, and direct integration with business systems. OpenAI announced four initial categories at launch: accounting, email marketing, legal analysis, and ticket resolution.
Microsoft is building the first enterprise management layer for Specialist Dots through its Agent 365 platform. That integration puts Specialist Dots under the same identity and compliance controls enterprises already apply to human employees through Microsoft Entra and Intune.
The Specialist Dots model reflects how OpenAI positions Dots in the enterprise: not one general assistant that does everything, but a roster of purpose-built agents with scoped access, auditable activity, and defined handoff rules. It is a closer analog to a contractor network than to a single employee account.
Enterprise Control Boundaries
The most important design decision in Dots is not the cloud computer or the app integrations. It is the distinction between proactive research and action work.
When a Dot is working in the background without the user present, it operates in proactive research mode. In this mode, connected app access is restricted to read-only. The Dot cannot send messages, change content, delete records, or control a browser or computer. It can observe, gather information, and prepare work for review.
Action work operates on a separate permission path. Users define Custom Rules that allow, require approval for, or block specific action types. An auto-review system checks actions that could affect accounts or share information before the Dot proceeds. Some actions including changing a password or account recovery settings are permanently reserved for the human user regardless of Custom Rules.
Credentials for connected applications are stored separately from the model. Saved passwords are used without being exposed to GPT-6 Astra directly. Each Dot’s cloud computer is isolated from the user’s physical device unless the user deliberately connects them.
| Mode | Access Level | What the Dot can do |
|---|---|---|
| Proactive research (user away) | Read-only | Observe, gather context, prepare work for review |
| Action work (explicit rules) | Controlled writes | Send, change, execute within Custom Rules |
| Reserved actions | Human only | Password changes, account recovery, sensitive mutations |
Private Intelligence: The Enterprise Data Tier
Enterprise adoption of always-on agents brings a specific concern: sensitive data flowing through persistent agents creates a new exposure surface. OpenAI’s response is Private Intelligence, a set of privacy controls for enterprise customers built in partnership with Databricks and Snowflake.
Private Intelligence has two components. The first is Zero Data Retention with Private Safety Processing, which delivers enhanced safety without storing customer content. The second is Private Inference, currently in preview, which uses confidential computing so processing happens with verifiable controls while data is in use.
Business, Enterprise, and Edu workspace content is not used to train OpenAI models by default. Private Inference extends that commitment to the compute layer, addressing the concern that even a no-storage promise still requires trusting the underlying infrastructure.
What This Means for Enterprise AI Leaders
The Dots launch completes a pattern that has been building across the enterprise AI stack. AI agents are no longer a deployment question: the capability is there. The question now is what governance, integration, and workflow design look like when agents work continuously rather than on demand.
Several implications are worth tracking closely.
The unit of enterprise work is shifting from sessions to goals. A session-based AI assistant completes one task per conversation. A persistent agent can hold a goal for days, accumulating context and making progress across multiple touchpoints. That is a fundamentally different planning model for operations and knowledge teams.
Specialist Dots set a new bar for vertical AI deployment. Role-specific agents with scoped access and defined handoff rules are closer to how organizations deploy human contractors than to how they have historically used software. The accounting Dot and legal analysis Dot models suggest a future where AI agent deployment is organized by function and governed by role, not by model version.
Always-on agents require always-on governance. The proactive research and action work distinction is a reasonable first design but it is a starting point, not a complete governance architecture. Organizations adopting Dots at scale need audit trails, scope reviews, and human handoff protocols before agents with production access run unattended for extended periods.
The Microsoft Agent 365 integration is the enterprise distribution story. Most large organizations manage identity and device compliance through Microsoft. Connecting Specialist Dots to Agent 365 means enterprise IT teams can govern AI agents through the same tools they already use for human employees. That removes the primary procurement objection for regulated industries.
OpenAI says Dots are already changing what its own employees feel capable of taking on. For enterprise teams ready to structure an evaluation of always-on agents within their workflows, the Enera team can help with that architecture.
Sources: WIRED: OpenAI Dots launch | OpenTools: control boundaries analysis | OfficeChai: DevDay liveblog | AI Daily Post: technical infrastructure | TechStartups: Dots use cases