Dot: OpenAI's 24/7 Cloud AI Agent Redefines Continuous Autonomous Work

OpenAI announced Dot on September 30, 2026, marking a shift from chat-based assistants to continuous, cloud-resident AI agents. Unlike traditional assistants that wait for user prompts, Dot operates with an independent cloud computer, browser, and identity, enabling it to work persistently, proactively report progress, and request decisions only at defined checkpoints.

Core Capabilities: Tasks That Outlive the Chat Window

Dot is powered by GPT-6 Astra and connects to over 4,000 applications via a plugin ecosystem. It can be reached through ChatGPT, SMS, Slack, Teams, or voice calls. Its persistent workspace includes:

An isolated cloud computer and browser.

Independent access credentials and permission configuration.

Ability to write, run, and test code.

Proactive progress updates, questions, or decision requests.

User-accessible inspection of its cloud environment at any time.

For developers, a typical workflow delegates bug-fixing: Dot reads customer feedback, locates the code, implements and tests the fix, then submits a pull request. The engineer intervenes only at the review stage, making Dot function more like a remote colleague than a step-by-step chatbot.

From "Write Code for Me" to "Own a Responsibility"

Dot's use cases span multiple roles:

Developer's Dot: monitors feedback, completes development and testing, submits PRs.

Product manager's Dot: updates release materials and creative assets when scope changes.

Scientist's Dot: re-runs analyses and updates paper figures when new data arrives.

Sales lead's Dot: verifies requirements, builds proofs of concept, refreshes proposals.

Content team's Dot: edits clips per style guide, prepares show notes and social posts.

These tasks share a need for cross-application, cross-file, and cross-time execution. To integrate such agents, teams must define tasks with at least four layers:

Goal: what must be delivered
Scope: which resources may be read or modified
Stop conditions: when the agent must pause for human input
Deliverables: PR, test report, screenshots, or change log

Without stop conditions, continuous execution risks uncontrolled resource consumption; without deliverables, autonomous completion cannot fit into team processes.

Proactive Research Is Not Unlimited Autonomy

Dot includes a "proactive research" mode that uses permission-limited tools to access already-connected apps when not directly collaborating. However, it cannot send messages, modify app content, or control the user's browser. This boundary is critical: the agent may gather information, organize options, and draft artifacts, but any action that mutates external systems requires explicit authorization.

Development teams can classify operations into three tiers:

Auto-allow: read repositories, query monitoring, run read-only tests.

Request approval: modify code, create PRs, update internal docs.

Always human: change passwords, alter payment info, grant new permissions.

Dot also supports custom rules that set specific operations to allow, require approval, or deny. This rule-based approach is more production-ready than per-action pop-ups because permissions become durable policies rather than ad-hoc decisions.

WorkDots: Agents as Organizational Members

OpenAI also introduced WorkDots — dedicated Dots that hold specific responsibilities within an organization. They possess independent identities, credentials, and system access, and are already used for procurement, invoice processing, email marketing, customer support, and commercial contract execution.

Microsoft is integrating WorkDots with Agent 365's enterprise governance and security controls. The organizational imperative is not simply buying more bots, but treating agents as system identities that require onboarding, privilege separation, auditing, and offboarding.

The article recommends creating a minimal asset profile for each long-running agent:

Identity name and owner.

Allowed applications and data scope.

Actions permitted to run automatically.

Actions that require approval.

Location of task logs and anomaly alerts.

Decommissioning and credential rotation procedures.

Availability and Domestic Trial Path

Dot is rolling out to eligible Pro, Business Premium, and Enterprise users. Pro and Business Premium plans include one Dot; Dot conversations do not count against ChatGPT usage limits. Additional Dots or higher speeds can be added monthly.

For teams in China wanting to validate task decomposition, approval workflows, and delivery acceptance, the article suggests using Code80 to access compatible model endpoints, focusing first on workflow and permission design rather than full system integration.

Frequently Asked Questions

What is the biggest difference between Dot and a regular chatbot?

Dot has its own cloud computer, browser, and identity, enabling continuous cross-time task execution; regular chatbots typically generate text or suggestions within a single session.

Can Dot directly operate my local computer?

By default, Dot works on an isolated cloud computer. Users may explicitly authorize local device connections, but this should be managed as a high-risk permission change.

Why do agents need independent identities?

Independent identities let teams know who accessed data, modified files, or submitted PRs, and simplify permission setting, log tracing, and access revocation during personnel changes.

How can domestic teams trial continuous-execution agents?

Use Code80 to connect compatible APIs and run a low-risk pilot: read-only collection, draft generation, human confirmation, then controlled execution of changes.