Chinese tech workers are starting to train their AI doubles–and pushing back

A satirical GitHub project called Colleague Skill, created by Shanghai AI Lab engineer Tianyi Zhou, has gone viral on Chinese social media. The tool lets users name a coworker, automatically imports their chat history and files from workplace apps like Lark and DingTalk, and generates a detailed manual of that person's skills, duties, and even unique quirks—such as punctuation habits—for an AI agent to replicate. Though intended as a stunt commenting on AI-driven layoffs, it resonated because many tech workers report their bosses are actively pushing them to use agent tools like OpenClaw or Claude Code to document and automate their own tasks.

This push for creating AI 'doubles' is prompting a wave of soul-searching and resistance. While companies see value in codifying employee know-how to standardize workflows, workers find the process alienating and reductive, feeling their roles are being flattened into replaceable modules. In response, some have created countermeasures, like an 'anti-distillation' tool published on GitHub by AI product manager Koki Xu, designed to sabotage the workflow creation process with light to heavy interference modes. The phenomenon highlights the tension between corporate efficiency gains and the preservation of worker dignity and job security in the AI age.