Meta CEO Mark Zuckerberg acknowledged at an internal town hall on the 2nd that the development of AI agents is not moving as fast as the company expected. He said that for at least the past four months the trajectory of development had not accelerated in the way he had anticipated. An AI agent is software that takes a goal and breaks the work into steps and carries it out on its own, without a person giving instructions at each stage. It is also the technology most often named as a candidate to take over office work outright.
That same week, California unveiled an "AI-unemployment tracker" meant to measure directly the damage AI does to jobs. Little has been made public about its detailed design, so it is hard to follow how the indicators are defined. Even so, the fact that such a tool has appeared is itself a signal. An attempt has begun to handle the fear as a measured value rather than leaving it a vague feeling.
The contrast is sharp.
Meta will pour up to $145 billion into AI infrastructure in 2026, about 224 trillion won in Korean money. Even after putting in that much money, the CEO said in public that "it is not going as well as expected." Zuckerberg said the reorganization had not been as smooth as he wanted and that he had also been wrong about when to time it, and he said the investment he put into the newly built organization had not borne fruit.
He also looked back and said he had been optimistic that Meta could build products like Anthropic's coding agent, but that things had actually turned out differently. He also said he had pulled staff from existing teams into the applied AI organization to push development forward.
In the meantime, what jobseekers in Korea feel has moved in the opposite direction. In the "2026 second-half job outlook" survey that Jinhaksa Catch put to 1,025 jobseekers, 71.0% said the second half would be harder than the first. The reason most often given for the deepening difficulty in finding work was the overall economic downturn at 38.0%, followed by intensifying competition among applicants at 29.0%.
"Job losses from the spread of AI" came to 23.0%. In the same survey a year earlier, that item was 8.0%. It jumped nearly threefold.
An interesting mismatch also shows up. While worries that AI is pushing people out have grown, a report late last month said engineer hiring has actually risen. Perception surveys and employment figures measure different things, so they do not collide head-on. The fact alone that the two trends sit in the same period reveals how loose the sentence "AI reduces jobs" is.
The developer says it is four months late, and jobseekers' anxiety about AI and jobs has tripled in a year.
The picture on the international stage also sits on this gap. The independent international scientific panel on AI that the United Nations set up last August has released its first report. The scale of about one billion people using AI each week was presented along with it.
Co-chair Maria Ressa pointed out that the United States alone holds 75% of the world's large AI computing complexes. Use has spread across the globe, while the facilities that run it are concentrated on one side.
Something else that happened inside Meta is not unrelated to this discussion. In April the company installed a monitoring program on U.S. employees' PCs with no option to refuse, then halted its operation in June after an incident in which sensitive information leaked. It said that when it resumes, it will switch to a method that turns on only if the employee consents.
CTO Andrew Bosworth explained that the collected data, such as mouse movements and keystrokes, had not been used to train AI models. The employee monitoring program and AI agent development were under way inside one company at the same time.
Zuckerberg said he expects much bigger results from the AI investment to start appearing within the next 3-6 months. The deadline he named falls between the end of this year and early next year. By around then, California's tracker will also have piled up several months of observations. The first chance comes to place the results the developer promised and measured employment indicators on the same timeline.
Next week, a UN international conference on AI opens in Geneva, Switzerland.
The most useful information for someone revising a résumé right now is not a prophecy that "AI will eliminate something within a few years." It is the fact that a window is being created where it is possible to check, quarter by quarter, which jobs and which tasks have actually been automated and by how much. In the survey, the share who said they wanted a large company was 53.0%, but the places they actually applied to over the past year were mid-sized companies at 61.0%, more than large companies at 58.0%.
They have long been moving with their preferred employers and the employers they apply to kept separate. What is needed is grounds that make it unnecessary to do that adjustment by feel.
