AI Doppelgangers 🤖: A Terrifying Experiment 😱

THE VIRAL TOOL: COLLEAGUE SKILL
The story begins with Colleague Skill, a GitHub project that rapidly gained traction on Chinese social media. This tool allowed users to “distill” the skills and personality traits of their coworkers into AI agent profiles, sparking immediate interest and, ultimately, considerable debate.

THE WORKER RESPONSE: DOCUMENTATION AND AUTOMATION
A number of tech workers, speaking to MIT Technology Review, revealed that their employers were actively encouraging them to document their workflows using tools like OpenClaw and Claude Code. This shift was driven by the desire to automate tasks and processes using AI agents, mirroring the functionality offered by Colleague Skill. The tool itself was created by Tianyi Zhou, an engineer at the Shanghai Artificial Intelligence Laboratory, as a response to AI-related layoffs and the growing trend of companies asking employees to automate themselves.

A USER’S EXPERIENCE: THE UNCANNY COWORKER
Amber Li, a 27-year-old tech worker in Shanghai, utilized Colleague Skill to recreate a former coworker as a personal AI agent. The tool quickly generated a detailed file outlining the coworker’s job duties and even their unique quirks, such as their punctuation habits. Li found the resulting AI agent surprisingly effective, using it to debug her code and receive instant replies. However, she also described the experience as “uncanny and uncomfortable,” highlighting the potential for a disconcerting shift in the workplace.

THE GROWING TREND: AI AGENTS IN CHINA
The popularity of Colleague Skill has contributed to a broader trend in China, where bosses are pushing tech workers to experiment with AI agents. These agents can perform tasks like reading and summarizing news, replying to emails, and booking restaurant reservations. Despite their potential utility, tech workers on the ground report limited success in business contexts, leading to a reliance on detailed work blueprints like those generated by Colleague Skill.

ANALYSIS AND IMPLICATIONS: STANDARDIZING WORKFLOWS
Hancheng Cao, an assistant professor at Emory University, believes that companies’ push for employee-created work blueprints like Colleague Skill goes beyond simply following a trend. He argues that firms gain valuable internal experience with the tools, alongside richer data on employee know-how, workflows, and decision patterns. This data can then be used to standardize or codify work processes and identify areas that still require human judgment.

WORKER FRUSTRATION AND COUNTERMEASURES: THE “ANTI-DISTILLATION” TOOL
The process of creating agents or blueprints for them has been met with resistance from some workers, who feel it reduces their individuality and expertise. One software engineer, speaking anonymously, described the experience as “reductive,” feeling that their work had been flattened into modules for easier replacement. In response, Koki Xu, an AI product manager in Beijing, created an “anti-distillation” skill on GitHub, designed to sabotage the process. This tool, available in three sabotage modes, rewrites the generated material into generic, non-actionable language.

LEGAL AND ETHICAL CONSIDERATIONS: OWNERSHIP AND IDENTITY
Xu’s tool and the broader trend of Colleague Skill have raised significant legal and ethical questions. While a company might claim ownership of work chat histories and materials created on a work laptop, capturing elements of personality, tone, and judgment becomes far less clear. Xu hopes this prompts discussion about protecting workers’ dignity and identity in the age of AI.

THE FUTURE OF WORK: ADOPTION AND PRECAUTIONS
Several tech workers, including Amber Li, express concerns about the devaluation of their skills and the potential for alienation. While Li’s company hasn’t yet implemented AI agents to replace workers, she acknowledges the feeling that her value is being diminished. The trend is further fueled by the widespread adoption of AI tools like OpenClaw, with seven agents already set up by AI product manager Xu.