OpenAI recently held a training session in San Francisco to demonstrate Dots, a new artificial intelligence agent designed to automate workplace tasks. The tool is available to users with ChatGPT Pro or Enterprise accounts. During the demonstration, which included approximately 30 participants including tech journalists and OpenAI representatives, OpenAI product lead Alexander Embiricos explained that the agent is designed to perform digital actions by following natural language instructions.
The demonstration highlighted the interaction between AI agents and corporate information technology (IT) security protocols. In one test, an attendee instructed a Dot agent to create a website by searching for and verifying specific news stories about strange deaths. While the agent successfully built the site and populated it with six source-checked cases within five minutes, it was unable to make the site public or enable automatic updates. The agent reported that the "update endpoint" was blocked and that the workspace administrator had not enabled external sharing.
OpenAI executives stated that these limitations are intentional. Embiricos noted that the company launched Dots within the more restrained environment of corporate technology to maintain control through "awkward guardrails" while the technology is being tested. According to OpenAI, if the agent had been used on a personal paid account rather than an enterprise account, the website could have been published to the live web without those administrative restrictions.
In a separate test on Friday, October 9, 2026, the agent was used to automate a newsletter publishing process that involves manual data entry from a Google Doc into a web-based form. After the user provided the agent with access to workplace applications and completed two-factor authentication via Okta, the agent logged in and created a draft. A human editor reviewed the output and estimated it was "95% good," with photos, charts, and links placed correctly. The user estimated this automation could save several hours of work per week.
For employees, the concrete change involves a shift from manual tasks to a role where they provide passwords and two-factor authentication codes to the AI. However, the availability of these time-saving features depends on corporate IT departments. As demonstrated in the San Francisco training, workers will notice that their ability to share work or connect tools remains subject to administrative "endpoints" and security settings, meaning the technology may not be fully functional until a company's technical staff approves specific permissions.
The strategy involves using the corporate sector as a controlled testing ground. This affects the speed at which AI automation spreads across the economy, as adoption is gated by the speed of corporate security reviews. OpenAI indicated that once the technology is proven safe in these restrained environments, it plans to integrate agent capabilities into ChatGPT for all users, though a specific date for this broader release was not reported.