SK Telecom has put 1,500 AI agents into practical use. HD Hyundai Robotics said it brought in Palantir AI and cut the time required for a specific task by about 63%. An AI agent is software that breaks work assigned by a person into multiple steps and handles it on its own. Both are large figures.
Such figures do not in themselves prove that work has changed. The 1,500 figure shows only the scale of adoption and does not tell how often employees call on the agents or what results they get.
One wall SK Telecom ran into was its internal data.
The company is pursuing an organization-level AI transition under the name 'AX Innovation 2.0', and 4,000 employees have built agents themselves. Yet column names in its internal data were written only as English abbreviations or numbers, making it hard for the agents to work out their meaning. A column is a vertical field that separates items in a data table. After the company used AI to attach meanings and descriptions to about 310,000 columns across some 9,000 network center databases, the accuracy rate reportedly doubled.
It is too early to give data cleanup all the credit. If models were replaced or features improved during the same period, the doubling would have to be divided among those factors, and the cleanup scope presented also covers only one place, the network center.
HD Hyundai Robotics adopted the agents in August. The company said the time taken to resolve problems shortened by 13.6 days, from an average of 21.5 days to 7.9 days, and added that engineers' repetitive work also decreased. Calendar time and actual hours worked are different metrics, so depending on which one the 63% refers to, it may mean an easier time for those waiting or fewer working hours for engineers.
At SK Telecom the preparation process is visible but there are no work performance metrics, while at HD Hyundai Robotics the results are clear but how it reworked its internal data is not visible.
HD Hyundai Robotics projects that it can cut robot programming costs by more than 50% and plans to grow its physical AI solutions business as well. Physical AI refers to AI applied to machines that move in the real world, such as robots. The 50% is not an achieved result.
Open a data table used at your company and start by reading the column names. If they are nothing but rows of abbreviations, agents will have a hard time working out what they mean.
