OpenAI Introduces Dots, Always-On AI Agents That Can Work Across Your Apps

OpenAI Dots are new personal AI agents designed to work on tasks with limited human supervision.

Announced at OpenAI’s DevDay event in San Francisco on September 29, 2026, Dots are being presented as proactive assistants that can continue working in the background rather than waiting for users to send a new prompt.

The idea is to give AI a more active role in everyday work, allowing an agent to monitor tasks, interact with connected applications, and ask for approval when an important decision or action requires human involvement.

What Are OpenAI Dots?

Dots are AI agents that can be configured to perform ongoing tasks on a user’s behalf.

Rather than functioning only as a chatbot, a Dot can work through a task over time and communicate with the user when it needs information, reaches a decision, or has completed part of its work.

OpenAI describes the system as an “always-on” assistant. In one example presented by the company, a Dot noticed that a user had forgotten to bill a publication for an article. The agent prepared an invoice and sent it after receiving the user’s approval.

This type of workflow represents a significant change from traditional AI assistants. Instead of simply generating an answer, the agent can potentially identify a task, prepare an action, and wait for permission to complete it.

OpenAI describes Dots as proactive AI agents designed to work on tasks over time and interact with connected tools and applications. The company says users can customize their Dots, communicate with them across ChatGPT, and control when they are allowed to act independently. You can read more about the technology and its capabilities in OpenAI’s official announcement about Dots.

Dots Can Connect to Thousands of Apps

One of the most important parts of Dots is their ability to interact with external applications.

OpenAI says Dots can work with more than 4,000 apps through its ecosystem of integrations. The company is also introducing integrations with platforms including Slack and Microsoft Teams, allowing users to communicate with their agents outside the main ChatGPT interface.

According to OpenAI, Dots can use their own cloud computer, browser, and connected applications to carry out tasks.

Users can also open the Dot’s computer environment to see what the agent is doing. OpenAI says Dots can additionally be connected to a user’s own PC, potentially allowing an agent to work directly alongside the user.

That capability could make Dots useful for tasks that involve multiple applications instead of a single conversation.

From AI Assistant to AI Coworker

OpenAI’s examples suggest that Dots are intended to handle longer-running workflows.

A scientist, for example, could use an agent to incorporate new information into a research paper as laboratory results become available.

A software developer could ask a Dot to collect user feedback and then use that information to help develop bug fixes or new features.

These examples point toward a broader direction for AI development: agents that continuously handle portions of a person’s workflow instead of responding only when explicitly asked.

The distinction is important.

A conventional chatbot might help write a report when prompted. An autonomous agent could potentially monitor information, determine when the report needs updating, gather relevant material, and return to the user when a decision is required.

OpenAI Is Building Approval Controls Into Dots

Giving an AI agent access to applications and computers also introduces additional security concerns.

OpenAI says Dots have built-in rules that determine when they can act independently and when they should request user approval. Users can also create custom rules specifying which actions require confirmation.

For example, an agent might be allowed to gather information independently but require approval before sending a message or changing information inside an application.

OpenAI says users can manage these permissions through ChatGPT’s existing app controls.

The company is also emphasizing protections around passwords. According to OpenAI, Dots can use saved passwords when signing into websites without exposing those credentials directly to the underlying AI model.

Proactive Research Has Additional Restrictions

OpenAI says its proactive research functionality has been designed with additional safeguards.

The company says connected applications can initially be restricted to read-only access during certain background research activities. Under those restrictions, the agent cannot send messages, modify application content, or control the user’s browser or computer.

These limitations are particularly important because an AI system that can independently browse websites, access applications, and operate a computer has substantially more opportunities to make mistakes than a system that only generates text.

Restricting what an agent can do without approval can therefore provide an additional layer between an AI’s decision and a potentially consequential action.

OpenAI Plans to Let Users Control Multiple Dots

At launch, eligible users in selected markets will receive access to one primary Dot that they can name and customize.

But OpenAI’s longer-term vision goes beyond a single agent.

The company says it eventually wants users to be able to control teams of Dots, with multiple agents potentially handling different responsibilities.

That could mean one agent monitoring communications, another working with documents, and another handling research or software-related tasks.

However, coordinating several autonomous agents could also make permission management and oversight considerably more complicated.

The more systems an agent can access—and the more agents a user controls—the more important it becomes to understand exactly what each system is allowed to do.

Who Can Use Dots?

OpenAI is rolling out Dots to Pro, Business Premium, and Enterprise users in selected markets, with availability beginning September 29.

For eligible Pro and Business Premium users, OpenAI says the first Dot is included. Users can interact with their Dot through ChatGPT on desktop, the web, and mobile.

The company says usage inside ChatGPT does not have the same limits as Codex and Work, although those products continue to have their own usage restrictions.

Why Dots Could Change How People Use AI

The introduction of Dots reflects a broader shift from generative AI toward agentic AI.

Early AI assistants primarily answered questions and generated content. More recent systems are increasingly being designed to perform multi-step tasks, use external tools, and operate with less direct supervision.

Dots take that concept further by making continuous background work part of the product’s central idea.

If these systems work reliably, users could delegate repetitive workflows that currently require constant attention.

But autonomy also creates a different class of risks. An AI that can take action has more opportunities to make an incorrect decision, misunderstand instructions, access information it should not use, or perform an action that the user did not intend.

That makes permission controls and meaningful human oversight just as important as the agent’s ability to complete tasks.

The Challenge of Trusting Always-On AI

OpenAI’s Dots arrive at a time when the security of increasingly autonomous AI systems is receiving significant attention.

The central challenge is no longer simply whether an AI can produce a useful answer. It is whether the system can reliably determine when it should act, what it is authorized to access, and when it should stop and ask a human.

OpenAI’s approval controls and restrictions are designed to address some of those concerns, but real-world use will ultimately reveal how reliably the safeguards work across different tasks and applications.

For users, Dots could represent a major step toward AI that functions more like a digital coworker than a traditional chatbot. For OpenAI, the challenge will be making that autonomy useful without allowing convenience to come at the expense of control.

As OpenAI moves toward multiple cooperating agents, that balance is likely to become even more important.