Has *Shapez* Become Reality? Indian Workers Wear Cameras to Feed AI

They made a very direct trade-off here: sacrificing some of workers’ sense of boundaries first in exchange for a trainable, reusable, and quantifiable action database. On April 13, Fast Technology mentioned that a video from a garment factory in southern India sparked heated discussion on social platforms. In the footage, workers were uniformly wearing specially designed camera rings, whose core purpose was to record hand movements. From an industry perspective, this is not simply an “upgrade in surveillance,” but rather an upfront step in data labeling for future automated sewing, intelligent quality inspection, and robotic applications.

From a design perspective, this system is very much like breaking down a master craftsperson’s “muscle memory” into individual operational instructions that can be learned. According to the report’s analysis, the factory hopes to let artificial intelligence learn standardized work motions, process details, and skilled sewing techniques, then turn that experience into models to reduce reliance on skilled technicians. This approach is itself highly professional, because what textile manufacturing fears most is not low efficiency at a single point, but excessive quality fluctuations between different workers, which in the end drag down defect rates, training costs, and delivery stability all at once.

Technically speaking, motion capture is only the first layer. What becomes truly difficult afterward is data cleaning, action segmentation, label definition, and the removal of abnormal samples. Put simply, recording something does not mean it can be used for training. This is exactly where many factories are most likely to cut corners, ending up with a half-finished product of “lots of video, very little usable data.” More realistically, this kind of equipment can also record workers’ movements in real time, which objectively means it can be extended to work-status monitoring and workload calculation. That turns what was originally a production optimization system into a more fine-grained management system as well.

And this is exactly where the controversy erupted. Because when a device can both train AI and supervise pace and calculate output, what workers face is no longer just a “new tool,” but a behavioral data collection mechanism that is effectively always on by default. For companies, the math is very clear: unify standards, ease hiring difficulties, cope with uneven skill levels, and reduce dependence on skilled workers. But for frontline laborers, privacy, dignity, and bargaining power may all be compressed again. If such a system lacks clear boundaries, it can very easily leave traces of a rush-job style of management.

From a business-logic perspective, this also reflects the pressure facing India’s manufacturing sector. World Bank data show that manufacturing’s share of India’s GDP has fallen from about 17% 20 years ago to 13% in 2022, while the total number of manufacturing jobs nationwide is about 65 million. In January this year, The Wall Street Journal also noted that although the Indian government has continued to promote manufacturing development, tens of millions of workers have still flowed back to rural areas, and many factories are struggling to recruit. In the textile industry specifically, as India’s second-largest employer by workforce size, it carries both export responsibilities and the dual pressure of unstable labor and the need to improve efficiency.

The export data already make the problem clear. From April 2025 to February this year, India’s textile and apparel exports totaled $29.5 billion, lower than the $29.8 billion of the previous year. Yet India had previously set itself a goal of raising annual textile exports to $100 billion by 2030. The target is ambitious, but reality is unforgiving, which is why companies are treating AI, automation, and data-driven management as a kind of remedial solution. The problem is that if this solution focuses only on efficiency, without strengthening labor protection and institutional design, it may ultimately resemble a system overhaul that only improves metrics without caring about user experience—boosting KPIs in the short term while eroding morale in the long term.

My judgment is that similar practices will not remain confined to garment factories in the future, but will spread to more manufacturing links with strong repetitiveness and high dependence on skill. The issue is not whether data should be collected, but whether the scope of collection, data ownership, usage periods, and workers’ right to know can be put in place at the same time. Technology itself has no moral tendency, but how the system is designed and how power is distributed determine whether it becomes “a thoughtful production upgrade” or “a supervisory tool wrapped in AI.”