More than one in two wage workers uses generative AI at work, but only about one in ten can skillfully write instructions that fit the situation and the goal. This comes from the report "Generative AI and Corporate Productivity: Realities and Tasks," released on the 28th by the Sustainable Growth Initiative (SGI) of the Korea Chamber of Commerce and Industry. A survey of 3,000 wage workers aged 20 and older nationwide found a usage rate of 56.3%, with so-called "advanced users" at 13.6%. It means the tools spread quickly while the proficiency to handle them lags far behind.
Generative AI refers to software that produces documents, code and images when a person gives instructions in writing. The quality of the output varies greatly depending on how the instructions are written, and these instructions are called prompts. The report analyzed that the higher the ability to write prompts, the more statistically meaningful the gain in productivity. Which indicators and models were used to measure that relationship was not included in the report.
The size of the effect is not small. Use of generative AI was analyzed to cut working hours by an average of 17.6%, and users said they would have had to work 8.4 hours more without the tool. The survey ran from September 29 to October 15, 2025. Considering that the performance of the tools has not stayed where it was then, nearly half a year later, it is safer to read these figures as close to a lower bound.
Usage rates split by group. By industry, information and communications was highest at 77.6%, followed by professional services and science at 63.0%. By company size, large companies with 300 or more employees were at 66.5% and small and medium-sized companies with fewer than 300 at 52.7%, a gap of 13.8 percentage points. Usage was relatively high among men by gender, younger workers by age group, high earners by income, and white-collar workers by occupation.
The interesting point is the reasons given by those who do not use it. Workers who do not use it came to 28.5%, and they cited "little use for my work" and "don't know how to use it" as the main reasons. The share citing institutional constraints such as company security policy or internal rules was 25.5% at large companies and 12.3% at small and medium-sized companies. As the wall blocking adoption, rules weigh more heavily at large companies and a lack of skills at small and medium-sized ones.
Looking at work areas, the nature of the gap becomes clearer. The most common use was writing and summarizing documents. But the more frequently a user used it, the larger the share of professional and creative work done with AI. The same tool is a stand-in for a typewriter to some people and a planning partner to others.
Perceptions of whether AI will replace or complement people's work also split by career stage. With complete replacement set at 1 point and complete complementarity at 5, the work of early-career workers scored 2.92, mid-career workers 3.25 and senior workers 3.28. It means the shorter the career, the closer to replacement people see their own work.
These are the results of asking respondents about their perceptions and they do not measure the actual possibility of replacement, but they are reason enough to look again at how entry-level work is designed. Lee Chang-keun, professor at the KDI School of Public Policy and Management, also gave his views in the report.
Companies' next task is shifting from handing out accounts toward managing proficiency. In organizations where usage has passed half, installing more tools leaves little time left to gain. In contrast, the distance between the 13.6% and the rest has plenty of room to narrow through training, redesign of work and the sharing of in-house cases.
Once performance gaps move beyond departments and start to open between colleagues at the next desk, that is not a matter to treat as an individual's problem. It is the result of the organization not building a learning path.
8.4 hours a week is enough to empty out a whole working day. Whether to use that time to cut overtime or to take on harder work is not something the tool decides. If you are now using generative AI only to summarize documents, pick one task you spend the longest time on this week and start by writing down how you would explain that work to AI.
The gap does not open at the moment you log in. It opens in the time spent refining that explanation.
