OpenAI launches Dots, always-on AI agents in ChatGPT with their own cloud computers

OpenAI has introduced Dots, a new set of constantly active AI agents in ChatGPT, at its DevDay developer conference. These agents enable users to assign tasks and let the agents handle them independently, even while managing multiple projects concurrently. Each agent adapts to the user's preferences through feedback. The technology behind Dots is powered by GPT-6 Astra, and each agent operates from a dedicated cloud computer that users can access at any time to monitor its activities.

With access to over 4,000 apps via OpenAI plugins, agents can also operate directly on users' laptops with permission. Communication with agents is possible through ChatGPT on desktop, web, or mobile devices, as well as in Slack and Teams, where context follows the agent, allowing seamless project transitions. Agents provide progress updates and ask questions, and they actively seek ways to assist when not assigned tasks, a feature known as proactive research.

Although agents have read-only access to user apps during this time, any actions requiring account access or information sharing must undergo a safety check called auto-review, with monitoring systems capable of pausing or stopping an agent if safety concerns arise. OpenAI advises users to review significant work, as agents may still make errors.

Additionally, OpenAI showcased specialized agents tailored to specific tasks, provisioned with individual identities and credentials, such as procurement and invoice processing. These tests are currently being conducted within OpenAI, with plans to roll out the service to ChatGPT Pro and Business Premium subscribers, with limited access for some enterprise customers.

Dots is set to launch to ChatGPT Pro and Business Premium subscribers, with some accounts experiencing a delay in access. Enterprise customers, including those on Edu and Healthcare plans, can access a beta version upon activation by a workspace administrator.