"I want my own site that handles trip planning, monitors exclusive stories from other outlets and tracks major events around the clock."
That was the request typed into Dots, the artificial intelligence agent from OpenAI. The aim was to hand over to AI the daily work of checking news on dozens of companies and keeping up with press briefings. The ultimate goal was to build a work site that would gather the needed information without further instructions while the reporter was busy with other tasks, and display the results at a glance. Dots replied that it was "certainly possible," asked two or three follow-up questions about the desired features, then built them one by one.
Dots is an always-on AI agent that OpenAI unveiled on Sept. 29. While the standard ChatGPT is built to answer user questions, Dots is designed to carry out assigned work continuously. It runs on the GPT-6 Astra model and comes with its own cloud computer and browser. It keeps working after the user closes the chat window, checking for information on a set schedule and delivering the results.
The reporter named the agent "TechDots" and assigned three tasks. The first was to search hourly for exclusive stories from major domestic business and IT outlets. When a new story appeared, it was to analyze the full text, separate newly disclosed facts from previously reported material and provide links and dates for earlier Seoul Economic Daily coverage. The second was to check daily at 8 a.m. for press briefings in deep-tech fields such as robotics, AI chips and startups, along with National Assembly policy forums and conferences. The last was to compare flights and lodging for overseas trips, track price changes and recommend reasonable options.
The most striking result was the work dashboard. After the three features were built, the reporter asked for a single screen showing the status and latest results of each task. Within minutes, Dots created a private website. Exclusive-story search results appeared as a list, reporting events as a calendar, and the trip-planning screen included search fields for destination, dates, preferred departure times and a lodging budget cap. In effect, a personal assistant was built through natural-language instructions alone, with no web development knowledge.
Dots' ability to design its own workflow stood out most in managing the reporting calendar. After a single log-in to the company email account, Dots read press briefing and event invitation emails without further intervention and extracted event names, dates and times, venues and application deadlines on its own. It then combined those with events found through web searches, including National Assembly policy forums, corporate product launches and technology seminars, to assemble a single reporting calendar.
There was also no need to spell out every feature from the start. Asked during use whether it could distinguish events found in email from those found through web searches, Dots proposed color-coding by source and adding reporting-priority badges graded A through C. It implemented the feature without a separate development instruction and verified that it worked on the actual screen.
That autonomy also showed in how it set reporting priorities. Asked how it had assessed the importance of each event, Dots said it had used "reporting value rather than the source of the information" as its standard. "Events with scheduled announcements of new products, investments or policies, or presentations by chief executives or chief technology officers, were graded A; events useful for expert interviews or contact with industry officials were graded B; and general industry trends or routine events were graded C," it said.
Rather than simply repeating the initial instructions, Dots improved its work tools to match additional requests. By linking event collection, source identification, importance assessment and calendar display, it set itself apart from a simple scheduling tool.
It was less adept, however, at presenting its results in terms a person could easily grasp. In its analysis of rival outlets' exclusives, for instance, it used unnecessary technical terms such as "canonical URL/AMP failure" and "JSON-LD and Open Graph check," prompting repeated requests for revision.
The links between features were not seamless either. The dashboard allowed input and saving of trip destinations, dates and preferred flight times, but buttons alone could not launch flight and lodging searches or price tracking, requiring separate instructions in the chat window with Dots. Verifying the accuracy and reporting value of the results also remained the reporter's job. Still, once work criteria were set, the agent collected and reported information on a fixed schedule, easing the burden of routine tasks.
Dots is currently being rolled out to ChatGPT Pro and Business Premium users. Enterprise users can access a beta version once an administrator enables the feature. Alongside Dots, OpenAI has introduced Spaces, a shared workspace, and Pages, a document writing and editing tool. It plans to add Slides and Sheets functions to expand AI-based collaboration features.