A humanoid robot finished the 100 meters in 9.39 seconds on the first day of the 2026 World Humanoid Robot Games in Beijing. For more than fifteen years, Usain Bolt held the record, which was set in Berlin in 2009 and was regarded by many as unbeatable.
On a Saturday morning, in front of hundreds of people who had mostly gathered to watch robots fall over, the machine erased it in a speed skating venue that had been converted. A few did fall. One caught fire. However, one also ran faster than any human has ever done. Before discussing spreadsheets, it is probably worth taking a moment to consider that.
The job discussion quickly becomes abstract, full of decade-long projections and percentages that seem far off. According to PwC, by the middle of the 2030s, up to 30% of jobs might be automatable. According to Goldman Sachs, there are about 300 million impacted full-time job equivalents worldwide.
AI may replace up to two million manufacturing workers by 2026 alone, according to MIT and Boston University. Because the timeline still seems negotiable, these numbers are mentioned, discussed, and then mostly ignored. People may have been underestimating the amount of time remaining for negotiations.
The simultaneous occurrence of both physical and cognitive displacement distinguishes the current era from previous waves of automation. In earlier industrial revolutions, machines took the place of manual labor while increasing the need for knowledge workers.
The hand weaver was replaced by the loom, but to keep the mill operating, engineers, managers, and accountants were required. It is now more difficult to rely on that offset. While AI models are performing tasks that previously required years of training, such as software development, financial analysis, and legal research, Figure AI’s robots are producing cars in South Carolina at a rate of one unit per hour. Both ends of the offset that traditionally cushioned displacement are being compressed at the same time.
According to Stanford University’s analysis of employment data, employment among 22- to 25-year-olds has decreased by 2.7% since the widespread use of large language models, with the most vulnerable industries—finance, software, and the creative industries—seeing an increase of 12.8%. These are neither assembly line workers nor coal miners.
The safe side of the automation divide was supposed to be represented by these individuals. As this data accumulates, there’s a sense that the categories we used, such as routine versus non-routine and manual versus cognitive, are no longer holding up the way economists thought they would.
Research from Harvard Business School adds a perspective that is rarely discussed in policy discussions. According to the study, local economies eventually adapt when manufacturing jobs are lost to Chinese competition; new industries emerge, and the population stabilizes.
Something different occurs when jobs are taken over by robots. In fact, the population in the impacted commuter zones decreases. Researchers discovered that for each robot added, three fewer people enter an area. The gap is not filled by the emergence of new industries. The implication is that local labor markets may find it more difficult to adjust to robot-driven displacement on their own schedule.
None of this implies that simple pessimism is the solution. Customer service and data entry seem to be less durable than jobs requiring real human connection, such as nursing, teaching, counseling, and surgery. The World Economic Forum, which is not known for being overly optimistic, predicted that although AI would eliminate about 85 million jobs by 2026, it would also create 97 million new ones in a variety of unexplored industries.
AI could boost global economic output by $13 trillion by 2030, according to McKinsey’s global modeling. If the math is correct, it suggests that expansion and disruption are occurring simultaneously in various locations, which is precisely what makes the transition challenging to handle without seriously harming people in the interim.
Whether policy is changing quickly enough to matter is the more difficult question. The changes that the data describes are already happening. Since ChatGPT’s widespread adoption, online job postings in industries heavily exposed to AI have decreased. The UK is especially vulnerable on this measure due to its concentration of employment in the service sector.
The majority of governments are still debating frameworks. Due to cost pressures, businesses are implementing robots and AI models without waiting for regulatory approval. It’s still unclear if the organizations intended to oversee this shift—education systems, social safety nets, and retraining programs—were ever built for displacement at this rate.
The physical reality of the situation was made visible by the robot games in Beijing, something that industry reports seldom do. Two thousand machines competed in kickboxing, playing table tennis, running obstacle courses, and using their hands to sort objects. Some of them experienced severe failures. Enough of them were successful to cause the crowd to pause and reflect. The sprint record was merely a figure. However, when numbers are set by a machine, they tend to land differently.
