OpenAI launches Dots: AI agents that work around the clock on your behalf

OpenAI has unveiled Dots, a new class of AI agents designed to operate continuously on users' behalf, rolling them out to ChatGPT Pro and Business Premium subscribers this week. Unlike conventional chatbots that respond only to active prompts, these agents run autonomously on OpenAI's cloud infrastructure, maintaining their own browser sessions and accessing a broad ecosystem of integrated applications.

Autonomous workflow automation#

Dots represent a shift toward persistent, multi-tasking AI assistance. Once activated, they can progress through complex workflows without waiting for human intervention—investigating issues, gathering data, and executing solutions across connected apps. The agents retain context across multiple platforms including ChatGPT, Slack, and Microsoft Teams, allowing them to coordinate activities and maintain project continuity.

OpenAI demonstrated Dots handling software development tasks: scanning customer feedback for recurring issues, designing targeted fixes, testing changes, and generating pull requests complete with video documentation showing what was modified. Beyond coding, the agents can refresh scientific analyses as fresh data arrives, evolve sales proposals based on shifting client needs, and transform interview recordings into highlights, documentation, and shareable social media clips.

Safety through segmented permissions#

Because unsupervised agents pose obvious risks, OpenAI has structured Dots with granular control mechanisms. When users are offline, agents can conduct read-only "proactive research," searching connected apps for relevant information without making changes or sending messages. Actual modifications trigger an automated review layer that evaluates proposed actions against user instructions, OpenAI's safety rules, and any custom policies the user has defined.

Rules can allow certain actions to proceed automatically, require explicit approval, or block operations entirely. Sensitive functions—like password changes—always remain under human control. The system also guards against malicious instructions embedded in external content, monitors agent behavior for anomalies, and allows users to pause or stop agents from a dedicated Activity View. Users can inspect all background work the agent performs and adjust its direction in real time.

Rolling deployment and pricing#

Specialist Dots aimed at enterprise workflows are also in preview, provisioned with organizational identity and access to critical business systems. OpenAI has tested this model internally for procurement, invoice handling, email campaigns, customer service, and contract management, with external enterprise pilots launching soon.