At the DevDay conference in San Francisco, OpenAI introduced dots, AI agents capable of independently performing long-running tasks and continuing work without constant user involvement.

A proprietary computer and over 4,000 applications

Each Dot receives its own cloud computer and browser, and can connect to more than 4,000 applications and corporate services via plugins.

The agent is capable of working with connected applications, using a browser, and performing actions within software interfaces. The user can check its progress at any time, and a single Dot can manage multiple projects in parallel.

Background operation

Dots are powered by GPT-6 Astra and can continue executing tasks in the cloud even after the browser is closed or the computer is turned off.

Round-the-clock availability does not mean unlimited computing: complex tasks consume subscription resources, and the use of Codex or ChatGPT Work is counted against the limits of these services.

OpenAI cites software development as an example. Dot can analyze user feedback, identify recurring issues, fix bugs, test changes, and submit a ready-to-use pull request to the developer.

Over time, the agent learns the user’s preferences and takes their requirements for the result into account.

ChatGPT, Slack, and Teams

You can communicate with Dot via ChatGPT, including voice calls, as well as through Slack and Microsoft Teams. Context is preserved across services, so you can start a discussion in one application and continue it in another.

The agent can also independently report on progress, ask questions, and request user decisions. In the future, OpenAI plans to add support for standard text messages.

Control and security

The user determines which applications Dot can access and which actions it can perform independently. For critical operations, mandatory confirmation can be enabled, and some sensitive actions, including password changes, are unavailable to the agent.

In the background, Dot can independently analyze data in connected applications, but only in read-only mode. Without user intervention, it cannot send messages, modify application content, or control the computer.

If suspicious instructions or potentially dangerous actions are detected, the task execution may be stopped.

Release dates

The rollout of dots has already begun for ChatGPT Pro and Business Premium users in supported regions. The first Dot is included in the subscription.

Enterprise, Edu, and Healthcare customers gain access to the beta version once it is enabled by an administrator. Moving forward, OpenAI plans to expand access, add the ability to create multiple agents, and increase available computing resources.

Follow MobiDevices
Telegram