Introduced at OpenAI’s DevDay conference on September 29, 2026, Dots are persistent, ‘always-on’ AI agents powered by GPT-6 Astra. Unlike a conventional chatbot, which normally waits for a user to ask a question and then produces a response, a Dot can be assigned an ongoing goal.

It can monitor changing information, work across connected applications and continue carrying out tasks even after the user leaves the conversation. OpenAI says Dots can work through ChatGPT, Slack, Microsoft Teams, and other services, and the company eventually envisions users operating multiple specialized Dots at the same time.

OpenAI’s launch of Dots marks another step in the evolution of artificial intelligence from systems that answer questions to systems that can continuously act on a user’s behalf. The main idea behind Dots is delegation. For example, a user might ask a Dot to monitor customer feedback, organize information, prepare reports, investigate unexpected research results, or help manage an ongoing software project. OpenAI has also demonstrated more personal uses, such as an agent recognizing from a user’s calendar that the person will be working through dinner and suggesting food-delivery options.

For businesses, OpenAI is developing specialist Dots for functions such as accounting, marketing, and legal analysis. Because such agents may have access to sensitive accounts and information, OpenAI has also introduced permission controls requiring explicit approval for certain high-risk actions, such as changing passwords or installing software.

Dots arrives only weeks after Meta’s Muse became a major consumer success, creating a direct contest over the emerging personal-agent market. Muse allows users to delegate activities, including shopping, travel booking, and filling out forms and quickly climbed to the top of U.S. and Canadian app-download rankings. Dots competes with Muse by emphasizing persistence, workplace integration, and connections with thousands of applications.

The larger significance is the rapid rise of ‘agentic AI.’ Earlier generations of generative AI primarily created text, images, or computer code. Agentic systems attempt to plan, make decisions, use software, and complete multistep assignments. If they become reliable, businesses could assign large amounts of routine administrative, research, and analytical work to software agents. This could increase productivity while placing pressure on occupations built around repetitive digital tasks.

Those same capabilities could also create significant risks. An AI agent with access to email, financial accounts, company databases, and online services could cause far greater harm through an error, manipulation, or cyberattack than a chatbot that merely generates an incorrect answer. Privacy also becomes increasingly important because useful agents must learn users’ habits, schedules, communications, and preferences. Researchers and regulators will face the challenge of determining how much autonomy such systems should receive and who is responsible when they make mistakes or cause harm.

Dots and Muse suggest that the next phase of the AI competition may not revolve around which chatbot gives the best answers. Instead, it may revolve around which company develops the most trusted digital agent, one that knows what a person wants, has permission to act, and can quietly perform work throughout the day. If that vision succeeds, agentic AI could transform computers from tools people operate into increasingly autonomous assistants that operate alongside them.