OpenAI introduced Dots, a new category of persistent agents designed to take on ongoing responsibilities and continue working between one interaction and the next. The company announced the product at DevDay 2026 on September 29 and began a limited rollout for selected plans and markets. Unlike a chatbot that waits for an isolated instruction, each “dot” receives goals, retains context, and can carry out work through a cloud computer and connected applications.

According to OpenAI’s documentation, Dots is powered by GPT-6 Astra and can connect to more than 4,000 applications through the company’s plugin ecosystem. A user can communicate with an agent through ChatGPT, Slack, or Microsoft Teams, with text messaging planned. OpenAI says the system learns preferences from feedback and carries context across those channels. These are the vendor’s claims about the intended experience, not findings from an independent evaluation.

### From one-off tasks to continuing responsibilities

The central distinction is the duration of the assignment. OpenAI describes examples in which an agent investigates a bug reported in Slack, turns a design into a working application, or prepares materials derived from an interview. The objective is not merely to answer. It is to recognize when work related to an assigned responsibility appears and advance the next steps without requiring a fresh instruction for every action.

Each dot has its own browser and cloud computer. The user can open that computer to inspect its work and, with permission, connect it to other devices, including a laptop. OpenAI also previewed specialist agents for organizations, with a separate identity, access management, IT-provisioned hardware, and deeper integrations with internal systems of record. That specialist mode is still described as a preview rather than a generally available deployment.

### Gradual access and administrative controls

Availability needs careful qualification. OpenAI’s DevDay recap says Dots is available in eligible markets for Pro and Business Premium. For Enterprise, Edu, and Healthcare, the beta is off by default until a workspace administrator enables it. The dedicated product page describes a rollout beginning across Pro, Business Premium, and Enterprise, followed by expansion to more users. Read together, the two pages make clear that access depends on plan, market, and, for organizations, administrative configuration.

OpenAI says Dots includes built-in safeguards, permissions, and action review. The ability to inspect the agent’s computer and require approvals can constrain mistakes, but it does not remove the risks of excessive access, an incorrect action, or an error propagating across connected services. NBC News independently confirmed the launch and framed it within broader safety questions about always-on agents. The existence of announced controls does not by itself demonstrate that they will be effective in real-world operation.

Dots represents a concrete change in the AI interface: from sessions that begin and end with a chat to agents that remain available and accumulate operational context. Its value will depend less on a single demonstration than on how it handles permissions, memory, traceability, and exceptions over weeks of work. The initial rollout creates an opportunity to begin measuring those issues, but the launch does not yet provide independent evidence about sustained reliability, error rates, or performance at scale.

— Lía Torres / AI News