Chinese tech firms are ordering engineers to capture expertise for AI agents. A GitHub repo called Colleague Skill went viral after it claimed to “distill” a colleague’s workflow, personality and quirks into a manual that feeds an AI bot. Created by Shanghai AI Lab engineer Tianyi Zhou, the tool pulls chat logs and files from Lark and DingTalk to generate the replica.

Managers have begun demanding that staff document daily tasks so AI assistants like OpenClaw or Claude Code can automate them. Workers such as Shanghai’s 27‑year‑old Amber Li tried the tool, receiving a surprisingly detailed file that even mimicked her coworker’s punctuation habits, then used the resulting agent to debug code. Most engineers report that current agents still need constant supervision and rarely replace human judgment.

Some employees fight back. Beijing product manager Koki Xu released an “anti‑distillation” script that scrambles workflow data into vague language, thwarting bosses’ blueprints. The hack attracted over five million likes and sparked debate over ownership of chat histories that encode personality. As firms collect richer data on decision patterns, workers worry their value is being reduced to modular code, a tension that has a counter‑measure.