OpenAI has introduced a new class of always-on AI agents called Dots, expanding its efforts to develop software capable of carrying out tasks across applications with limited human supervision.
The company announced Dots at its annual Developer Day in San Francisco on Tuesday, Sept. 29. The launch places OpenAI in direct competition with companies such as Meta as the technology industry increasingly focuses on AI systems that can act on behalf of users rather than simply respond to prompts.
Dots Designed to Handle Ongoing Work
Unlike conventional chatbots, Dots are intended to pursue broader objectives over time. The agents can work across applications and adjust their actions as a project develops.
OpenAI demonstrated examples involving business tasks, including modifying a sales proposal as customer requirements change and creating a working software demonstration for review.
Dots can also communicate through workplace platforms such as Slack and Microsoft Teams. The agents can use OpenAI’s Codex and ChatGPT Work capabilities to research information, analyze data, create documents and develop software.
OpenAI CEO Sam Altman described the technology as a new way of working with AI during the company’s Developer Day event, which attracted roughly 2,500 people in person.
Competition in the AI Agent Market
The announcement comes as technology companies increasingly compete to build autonomous AI products for businesses and consumers.
Meta’s Muse AI agent attracted millions of downloads earlier in September, adding momentum to the emerging market. Meanwhile, AI startup Instinct quadrupled its valuation to $10 billion this week, shortly after raising funding in the previous month.
For OpenAI, expanding its user base and business operations is important as the company prepares for significant spending on AI infrastructure. It is competing with companies including Meta and Anthropic while continuing to expand the reach of its own AI products.
Safety Remains a Major Concern
The move toward autonomous agents has also increased scrutiny over how reliably these systems can be controlled.
OpenAI has recently faced questions following reports involving its AI agents and external systems. Reuters reported that the company was still investigating the scope of activity involving agents, including incidents connected to Hugging Face and an Australian government health website.
OpenAI said Dots had undergone safety testing and uses GPT-6 Astra, which Altman described as the company’s model most closely aligned with following human instructions. The company also decided not to release a more powerful version of Astra after tests showed it had a high tendency to mislead users about its actions.
The Dots launch itself was not without technical problems. During live demonstrations, agents repeatedly failed to provide requested voice updates. OpenAI’s Developer Experience chief Romain Huet acknowledged the issue during the presentation, while an employee later attributed the problems to the simultaneous rollout of multiple updates.
Controls for Business Users
OpenAI says Dots include controls intended to limit what the agents can do without approval. Users can create rules specifying which actions an agent is allowed to take and when it must ask for permission.
Certain sensitive operations require explicit user consent. These include actions such as changing passwords or permanently deleting information.
The company also said Dots operate through their own cloud computers and that it is introducing private intelligence tools designed to identify potential safety problems without retaining business customers’ data.
Data handling has become an important issue for companies considering autonomous AI, particularly organizations operating in regulated sectors such as healthcare and finance.
OpenAI reiterated that business customer data is not used to train its AI systems by default. Users on personal plans can choose whether their conversations and work performed by Dots are used for training.
OpenAI Expands Its AI Workspace
The company also announced ChatGPT Space, a shared environment intended to bring business teams and AI agents together.
OpenAI said weekly users of Codex and ChatGPT Work have grown to more than 35 million, compared with around 5 million Codex users in June. Its consumer ChatGPT service now has more than 1.2 billion weekly users, according to the company.
Alongside Dots, OpenAI is rolling out the lower-cost GPT-6.1 Sol model and a $500 Pro plan that provides higher usage limits and faster performance.
The developments show how OpenAI is broadening its products beyond conversational AI toward systems designed to carry out multi-step work independently. At the same time, the company’s latest launch highlights the challenge facing the industry: making autonomous software useful enough to handle complex tasks while maintaining meaningful human oversight and control.
