OpenAI introduces Dots, always-on AI agents

OpenAI has introduced Dots, AI agents that can work continuously, connect to your applications and move tasks forward without you directing every step. The service is beginning to reach some ChatGPT users and aims to make AI a permanent assistant, not just a tool you consult when a question comes up.

Each Dot runs on GPT-6 Astra and has its own cloud computer, with a browser and access to the applications you authorize. It can research, prepare documents, track projects and alert you when it needs a decision. It can also work on multiple tasks at once.

The idea is simple: give it a goal, explain how you like to work and review its progress. Over time, the Dot learns your preferences and criteria. For example, it could detect that an invoice is missing, prepare it and send it only after asking for your approval.

An AI connected to your tools

Dots can connect to more than 4,000 applications through OpenAI's plugin ecosystem. You can contact them from ChatGPT on desktop, the web and mobile, as well as from Slack and Teams. Support for text messages will come later.

Context is preserved across channels. You can start a project in ChatGPT, share it with your team in Slack and continue the work without having to explain all the details again.

OpenAI also describes examples of how Dots are used inside the company:

• Investigating an error that appears in Slack.
• Turning a new design into a working application.
• Keeping a team coordinated during a planning cycle.
• Tracking administrative tasks, such as invoices or hiring processes.

These are early internal testing examples, not a guarantee that any Dot can automatically handle all of these tasks.

You decide how far it can go

A Dot can operate autonomously, but it includes limits so you remain in control. You can choose which applications it can use, review its activity and set custom rules. Those rules can allow specific actions, require approval or block them entirely.

Sensitive tasks still require your involvement. Changing a password, for example, always remains in your hands. In addition, an automatic review system compares actions that could affect your accounts or share information against your instructions, rules and security requirements.

When it works in the background, the “proactive research” feature uses read-only tools. That means it can view information in connected applications, but it cannot send messages, modify content or control your browser or computer without additional permissions.

Your team and its files remain separate from the Dot's cloud computer unless you choose to connect them. OpenAI says Dots can also use passwords saved on supported sites without exposing them to the model. Even so, it warns that they can make mistakes and recommends reviewing any work with important consequences.

Availability and pricing

Dots are beginning to roll out in eligible markets to ChatGPT Pro and Business Premium users. The first Dot is included in those plans at no additional cost and can be used to hold conversations, organize ideas and manage tasks continuously.

Enterprise users, including Edu and Healthcare workspaces, can try a beta if their organization's administrator enables it. OpenAI is also beginning pilots with specialized Dots for specific tasks, such as purchasing, invoicing, email marketing, customer support and contracts.

These specialized Dots will have their own identity, credentials and permissions within a company. OpenAI is also working with Microsoft to integrate them with Agent 365 and the security and governance controls companies already use.

Conversations with your Dot do not count toward the usual ChatGPT usage limits. However, tasks it launches or manages in Codex or ChatGPT Work do consume the corresponding limits.

The rollout is gradual, and OpenAI plans to let each user add additional Dots later and adjust their speed or monthly workload. To get started, you need to create the first Dot from the desktop app or desktop browser, connect the necessary applications and complete the initial setup.

The importance of Dots lies in the change of model: you move from requesting one-off answers to delegating complete objectives. The challenge will be finding out whether that autonomy really saves time without creating new problems with permissions, errors or oversight. That is why the key question will not only be what a Dot can do, but how easy it is to know what it is doing and stop it when necessary.