OpenAI has introduced Dots, a new category of AI product it describes as “always-on agents” that work continuously on a user’s behalf rather than waiting for prompts. Unlike a chatbot you query one question at a time, a Dot is meant to run in the background, learn a person’s preferences over time, and deliver finished work such as pull requests, invoices, or revised launch documents before the user even asks. The launch matters because it signals a shift in how OpenAI positions its AI products: from tools you actively operate to agents that operate on your behalf continuously.

What Dots actually are

According to OpenAI, Dots are powered by GPT-6 Astra and each has its own cloud computer, meaning it can execute tasks such as browsing, running code, or manipulating files without relying on the user’s own machine. Dots can also be granted permission to access a user’s laptop directly for closer collaboration. OpenAI says Dots connect to an ecosystem of more than 4,000 apps through plugins, allowing them to take action inside tools people already use, such as Slack, Teams, and ChatGPT itself.

The defining feature is persistence. A Dot doesn’t reset after each conversation; it “learns from feedback over time,” according to OpenAI, gradually adapting to how a specific person thinks, writes, and makes decisions. Users can name their Dot and interact with it through text, voice calls, or by simply assigning it tasks and letting it work independently in parallel with other tasks it’s already handling.

How it works in practice

OpenAI’s examples describe Dots handling multi-step, semi-autonomous workflows rather than single commands. A developer’s Dot can monitor customer feedback, identify recurring bug reports, build and test fixes, and hand back complete pull requests with demo videos attached — while the developer focuses on a bigger feature. A product lead’s Dot can absorb a change in scope and independently revise launch materials, drafts, and assets to match. A researcher’s Dot can rerun analyses as new data comes in and update figures and explanations in a paper, flagging only what needs human review.

One example cited by OpenAI describes an early tester’s Dot noticing an unsent invoice, preparing it, and sending it after the user’s approval — illustrating the kind of proactive, low-oversight task the product is designed for. OpenAI also notes that internally, Dots have started investigating bugs as soon as they appear in Slack and have turned new designs into working apps while human teams stayed focused on customer conversations.

Importantly, OpenAI states that users can open a Dot’s cloud computer at any time to inspect exactly what it is doing, which suggests a degree of transparency is built in rather than leaving the agent to operate as a black box.

How Dots differ from existing AI assistants

Most current AI tools, including standard ChatGPT sessions and code assistants, are reactive: they respond to a prompt and stop. Dots are pitched as proactive and continuous — working “24/7” toward a user’s goals without needing every step directed. The combination of a dedicated cloud computer, persistent memory of user preferences, and deep app connectivity distinguishes Dots from typical single-session copilots or narrow task bots.

OpenAI is also introducing a second tier called specialist Dots, aimed at organizations. These come with their own identity for access management, IT-provisioned hardware, and support for deep integration with a company’s systems of record, allowing them to take on well-defined responsibilities inside a business rather than acting as a general personal assistant.

Availability and what it means for users

OpenAI says it is beginning to roll out Dots across Pro, Business Premium, and Enterprise plans in eligible markets, with plans to expand access to more users over time. Initially, users start with a single “primary” Dot they can name and customize; OpenAI describes a longer-term vision of teams of Dots collaborating on a person’s behalf, though that capability is framed as a future direction rather than something available today.

For developers, the practical appeal is offloading routine but time-consuming work — bug triage, small fixes, documentation updates — to an agent that returns reviewable output rather than requiring constant supervision. For product and business teams, the pitch is similar: Dots absorb the busywork of keeping materials, plans, and assets updated as circumstances shift.

Caveats are worth noting. OpenAI’s own examples are curated success stories, and details on error handling, cost, data privacy boundaries, and how much oversight is genuinely required in practice are not fully specified in the announcement. Since Dots can act with real permissions across connected apps and devices, questions about access control, auditability, and mistakes at scale are likely to be central as the rollout expands.

Frequently asked questions

Do Dots require constant supervision? OpenAI says Dots are designed to work independently and bring back completed work, though users can check in via chat, voice, or by inspecting the Dot’s cloud computer directly.

What is GPT-6 Astra? It is the underlying model OpenAI says powers Dots, though the announcement does not provide detailed specifications about the model itself.

Can businesses use Dots for specific internal roles? Yes. OpenAI is previewing specialist Dots for organizations, which include IT-managed access and integration with internal systems of record for well-defined responsibilities.

Source: Dots: Always-on agents