If artificial intelligence (AI) replaces people's work, will the burden on office workers be reduced.

If artificial intelligence (AI) replaces people's work, will the burden on office workers be reduced.

Changes in AI company Antropics were unexpected. While AI was writing code and processing documents, employees' meeting time increased by 20%. This is because there are more and more cases of reviewing AI's outpouring of results and making important decisions. As AI speeds up business processing, human judgment and verification are emerging as a new "bottleneck" that determines overall productivity.

◆ With AI, it's taking a long time to discuss

According to Antropic's own employment and work change report released on the 17th (local time), meeting time per employee increased by 20% from a year ago as AI was introduced to work in earnest. The meeting time of employees working directly on development in the technology sector also increased by 16% year-on-year.

Antropics cited "human trust and responsibility" as the reason. This is because even if AI writes the code, people must discuss it among themselves to solve problems that affect the entire system or to make important decisions. Antropic explained, "When we solve program errors, we eventually find problems at the organizational and system level that several people have to come together to discuss and decide."

In the background is the productivity of AI, which has exploded. According to Antropic, as of August this year, the working hours of AI agents used by employees were six times that of human employees. However, there was a limit to the human handling of the results of AI. Approximately 80% of Claude's entire work session was time to wait for a human response or further instructions. This means that even if AI makes the results quickly, it takes a considerable amount of time for humans to check and give the next instruction. It has been calculated that even non-technical employees need more than half of their weekly working hours to read all the contents generated by AI. Antropic diagnosed, "Human attention and judgment have emerged as a new bottleneck that limits AI productivity."

◆ Human verification is the "bottleneck" of the AI era

The nature of employees' work is also changing. The proportion of tasks that manage AI, such as giving guidelines to AI, reviewing results, and checking operational status, has more than doubled from 10% last fall to 22% this spring. In an in-house survey in April, 20% of employees said their roles had "changed from working in person to managing multiple clods." An engineer said, "Many AIs constantly demand my attention and judgment," comparing it to 'child care'.

The abilities required of humans are also changing. One employee explained that strategic tasks, which used to account for only 20% of all tasks in the past, now account for 90%. Recruiters have also begun to value their ability to judge themselves and lead tasks rather than their ability to implement specific skills. On the other hand, the proportion of beginner manpower in new hiring in the technology sector fell to 9% from 27% at the beginning of last year. As AI replaces simple tasks, the position of less experienced employees is narrowing.

With the proliferation of AI, these changes are occurring everywhere. At the "Research at 2026" event held at Google's headquarters in Mountain View, California on the 14th, Yossi Mathias, Google's vice president and general manager of Google Research, said, "As AI creates hypotheses, researches literature, and builds research models, the core of the research is shifting to verification and confirmation." In Antropics, human judgment is the same as being pointed out as a bottleneck for productivity.

Silicon Valley correspondent Won Ho-seop