Colleague Skill: how a Chinese project digitizes employees before dismissal

The Chinese project Colleague Skill appeared on GitHub and gained thousands of stars within weeks: its idea is both simple and alarming at the same time — to record everything an employee can do before they are laid off and transfer this knowledge to an AI-agent successor.

What is "Employee Distillation"

The term is borrowed from machine learning: model distillation is the transfer of knowledge from a large neural network to a small one. The logic here is the same, except the source of knowledge is a living person. The practice involves creating datasets from a specialist's actual work — their correspondence, documents, decisions made, behavioral patterns — and subsequently fine-tuning a language model that will imitate the style and competencies of that specific employee.

This is not a one-off project. Similar tools are appearing on GitHub under different names — Knowledge Transfer AI, Employee Distillation, Skill Extraction. The trend came from China, where companies actively automate processes, but is quickly spreading to Western startups.

The Colleague Skill project automates this process. A manager connects the employee's corporate chats, email, and work files, runs the pipeline — and gets a "digital imprint" of the specialist, ready for integration into the corporate AI assistant. Moreover, this can be done while the person is still working or immediately after receiving a termination notice.

What Exactly Gets Into the Dataset

A typical set of data that such a system processes:

• Correspondence in Slack, Teams, WeChat over the last 1–3 years
• Code comments, pull requests, and architectural decisions
• Email chains with clients, partners, and colleagues
• Internal documents, spreadsheets, reports with edit history
• Call recordings and meeting transcripts

The result is a model that knows exactly how this specialist answered awkward customer questions, how they formatted code, and how they resolved team conflicts. In essence, a digital copy of a professional personality — created without the person's knowledge.

Legal and Ethical Bomb

This is where the real problems begin. In most countries, an employee has limited but real rights over data about their activities. European GDPR requires explicit consent for the use of employee personal data for new purposes. California's CCPA since 2020 has given workers the right to know what data their employer collects about them. Russian personal data law is formally applicable as well, although its enforcement in labor disputes is still limited.

"This is not just a matter of ethics.

If a company trains a model on employee data without their consent — this is a potential violation of several legislative acts," write participants in a discussion thread on Hacker News.

Most corporate labor contracts are written without accounting for this scenario. They contain neither an explicit prohibition nor an explicit permission to use an employee's digital footprint for training AI. This is a gray area where the outcome of a dispute depends on jurisdiction, contract wording, and the parties' willingness to go to court. A separate issue is what happens to the "distilled" employee after they leave. Their digital copy continues to "work," make decisions in their style, interact with their former colleagues and clients. The person is no longer at the company, but their professional personality is.

What This Means

"Employee distillation" is already a working tool, not a concept of the future. While regulators have not yet formulated rules, companies face a choice: use a powerful knowledge transfer tool with legal risk, or wait while competitors already do it. For employees, this means a new reality: professional data becomes a corporate asset long before laws emerge to regulate it. This is precisely why Colleague Skill is not just a viral GitHub project, but a symptom of a far broader transformation of the labor market.