OpenAI: AI agents take on bigger share of work at Korean firms
OpenAI Korea General Manager Kim Kyoung-hoon speaks during a press conference for its Korea office's 1-year anniversary in Gangnam District, Seoul, Wednesday. Courtesy of OpenAI
OpenAI said artificial intelligence (AI) agents are taking on more complex work at companies in Korea as use of its workplace and coding tools surges and businesses deploy AI across research, coding and other workplace tasks rather than limiting its use to chatbots.
OpenAI Korea General Manager Kim Kyoung-hoon shared that the number of users at local companies and institutions using its ChatGPT Enterprise has increased about 28-fold over the past year.
“Previously, even when we assigned work to AI, there was a lot of human intervention or human review, but I think we are now entering an era in which we can trust AI more and entrust it with more difficult tasks,” Kim said during a press conference Wednesday to mark the first anniversary of the AI company’s Korean office in Seoul.
He said the weekly active users of ChatGPT Work and the coding agent Codex increased about 14 times in the six months since this March. In August, Work and Codex accounted for 64 percent of output tokens generated by Korean users across ChatGPT’s products, despite having a smaller user base than the standard chatbot service.
Kim added that the use of AI agents was expanding beyond engineers into more diverse functions including legal, finance, human resources, accounting, marketing and sales. “The area where the number of users is increasing the most was actually legal,” he said.
OpenAI cited Samsung Electronics and Seoul National University as examples of organizations extending AI use across large employee and user bases.
Samsung Electronics is using OpenAI products in areas including research and development, marketing, overseas sales and production management, while nearly 50,000 students and faculty members at the university use ChatGPT and Codex for research and administrative work.
A graph shows GPT-6 Astra’s performance and cost efficiency in the BenchCAD evaluation. Courtesy of OpenAI
OpenAI on Friday unveiled the GPT-6 Astra model, which can complete tasks more reliably using the same tools and templates that users already work with. Kim expects the model to push this trend further by combining advanced reasoning with computer-use capabilities, allowing AI to handle more complex, open-ended tasks and produce finished work with less human intervention.
“What we focused on most with Astra was whether the model could really complete the work all the way by itself,” Kim said. “It can solve complex problems in real depth, continuously think about a goal set by a person and remember work it has done before to produce higher-quality results.”
The company is also expanding its local infrastructure and business ecosystem. It is preparing to offer inference residency in Korea, where AI model inference would be processed locally, while also working with local partners to secure the necessary graphics processing units and cloud infrastructure.
“Once the infrastructure is ready, we expect to be able to provide inference locally through a domestic data center,” Kim said.
OpenAI also plans to expand partnerships in cybersecurity and advertising. For its Daybreak cybersecurity initiative, the company plans to select more than five partners in Korea. A day earlier, Samsung SDS said it had become the first Korean company to join Daybreak as a pilot partner.