Chinese Workers Push Back as Bosses Ask Them to Train AI Replacements
Chinese tech workers are being instructed by their employers to train AI agents that could automate or replace their own jobs, sparking debate about worker dignity and the practical limits of AI automation. A viral GitHub project called Colleague Skill, created as satire by an engineer at Shanghai AI Lab, demonstrated how easily AI tools can extract and replicate individual workers' skills, workflows, and personality quirks from workplace chat histories. While some workers find the technology uncanny, companies see value in having employees document their processes to identify which tasks can be standardized versus which require human judgment. The trend reflects broader pressure on Chinese tech workers to embrace AI agent tools like OpenClaw and Claude Code, even as questions mount about job security and what it means to codify human work into replaceable modules.
TL;DR
- →Colleague Skill, a satirical GitHub tool, went viral in China by showing how to extract a coworker's skills and personality from workplace apps like Lark and DingTalk, then replicate them with an AI agent
- →Tech workers report bosses are pushing them to document workflows and create manuals to enable AI automation of their own tasks, creating anxiety about job security and the reductiveness of codifying human work
- →Academics and business leaders argue companies benefit from having employees create detailed work blueprints, which reveal which tasks can be standardized versus which require human judgment
- →Workers express alienation through dark humor on social media, describing the process as flattening their expertise into replaceable modules and turning their professional identity into training data
Why it matters
This reflects a critical tension in AI adoption: as agent tools become more capable, companies are discovering that extracting and standardizing human workflows is essential to making them work at scale. The Chinese tech sector, already shaped by intense competition and rapid AI adoption, is becoming a test case for how workers respond when asked to automate themselves. The backlash suggests that technical feasibility and business logic may collide with worker psychology and labor dynamics in ways that could reshape how companies deploy AI agents.
Business relevance
For operators and founders building AI agent products, this highlights both opportunity and friction. Companies see clear ROI in having employees document their processes, which generates training data and reveals automation opportunities. However, the resistance and anxiety from workers suggests that implementation strategies matter enormously. Framing agent adoption as augmentation rather than replacement, and being transparent about which roles are at risk, will likely determine adoption velocity and employee cooperation.
Key implications
- →AI agent adoption may require companies to explicitly manage worker anxiety and frame automation as augmentation rather than replacement, or face resistance and reduced quality of process documentation
- →The codification of human work into AI-trainable modules creates a new form of labor extraction, where companies gain not just task automation but also detailed intellectual property around employee workflows and decision patterns
- →Worker pushback in China, a market typically seen as early-adopter friendly, suggests that even technically sophisticated workforces may resist being asked to document themselves out of a job, potentially slowing agent deployment
What to watch
Monitor whether Chinese tech companies face measurable resistance or turnover as they push agent adoption, and whether worker concerns influence how other markets approach similar initiatives. Watch for new tools or policies that attempt to address worker anxiety, such as guarantees about redeployment or transparent automation roadmaps. Also track whether the Colleague Skill phenomenon spawns regulatory or labor-related responses in China or elsewhere.
vff Briefing
Weekly signal. No noise. Built for founders, operators, and AI-curious professionals.
No spam. Unsubscribe any time.
