# Digital Twin of an Employee: How Colleague Skill Is Changing the Approach to Knowledge in Companies
TL;DR: The open-source Colleague Skill project lets you create an AI copy of a colleague based on work chats, documents, and audio. The repository gained more than 8,000 stars on GitHub in just a few days. Against the backdrop of layoffs at Chinese IT companies, the project sparked a heated debate about the ethics of “knowledge distillation” and an employee’s rights to intellectual property.
What Is Colleague Skill and Why Is Everyone Talking About It
In April 2026, the Chinese social network RedNote (Xiaohongshu) was set abuzz by a project called Colleague Skill — an open tool for creating an AI agent that imitates a specific employee. Developer @titanwings from Shanghai AI Lab suggested using it before “cold goodbyes” with colleagues: upload chats, documents, spreadsheets, and audio recordings — and get a digital assistant that thinks, writes, and solves tasks in the style of the departed specialist.
The GitHub repository gained more than 8,000 stars in a matter of days and was translated into 7 languages, including Russian.
The project found itself at the epicenter of a debate at the intersection of three trends:
- Mass layoffs in China’s IT sector — according to Chinese media reports, companies asked employees to document work processes in detail, then laid them off and used the materials to train AI agents.
- Standardization of AI agents — the Agent Skills format makes it possible to “plug in” expert knowledge to an LLM without retraining the model.
- Digital twins as a business — the startup Viven, valued at $2.1 billion, is already creating AI copies of employees on a commercial basis.
How Colleague Skill Works
The project is built on a two-layer architecture, described in detail in the accompanying research paper.
Work Skill layer — professional knowledge
Extracts from uploaded materials:
- Technical standards and company policies
- Workflows and pipelines
- Experience and context behind decisions made
- Knowledge base for projects and products
Persona layer — behavioral profile
Forms a 5-level personality model:
- Hard rules — what the colleague never does
- Identity — role, grade, company
- Communication style — tone, length of responses, use of emojis
- Decision-making model — how they assess risks and prioritize
- Interpersonal behavior — how they behave in conflicts, meetings, and correspondence
Supported data sources
- Feishu (Lark) — full automation via API: messages, documents, spreadsheets
- DingTalk — via browser, since the API does not support message history
- Slack — via a bot with admin access (the free plan is limited to 90 days)
- WeChat — via SQLite and third-party tools (WeChatMsg, PyWxDump)
- PDF, screenshots — manual upload
- Email (.eml/.mbox) — manual upload
- Markdown and text — manual input
The “knowledge distillation” phenomenon: why it causes fear
The term knowledge distillation comes from machine learning — it refers to a process in which a compact model is trained on the outputs of a larger one. In the context of Colleague Skill, RedNote users drew a direct parallel: the employee is the large model, and their digital copy is the distilled version.
The situation is worsened by real-world cases. According to Chinese media, a number of companies introduced a procedure requiring detailed documentation of workflows under the pretext of improving efficiency. A few months later, employees were laid off, and their records were allegedly used to train AI agents. Messages from the “digital twins” of dismissed colleagues began appearing in work chats.
This raises two fundamental issues:
The problem of appropriating knowledge. Who owns an employee’s experience — the employee or the company? If knowledge is recorded in chats created during working hours, is its “digitization” lawful? The problem of replaceability. If an AI copy can perform 70–80% of a dismissed employee’s work, what bargaining position do the remaining employees have when discussing salary?Anti Distillation Skill: the community’s response
In the wave of outrage, user @whyyoutouzhele created the project Anti Distillation Skill — a tool that modifies archival documents so they become less useful for training an AI agent.
The methods include adding noise data and pseudo-information, blurring key decisions in a stream of irrelevant details, and systematically obfuscating contextual links between tasks.
This has sparked a new arms race: some create tools for “copying” knowledge, while others create tools to protect it.
Forks and offshoots: digitizing everyone
The success of Colleague Skill spawned a wave of forks for other scenarios:
- Ex-Girlfriend Skill — a digital twin of a former partner
- Boss Skill — an AI copy of a manager
- Digital Necromancy — digitizing deceased relatives
- Yourself Skill — creating your own AI copy for business correspondence
The latter is perhaps the most pragmatic. If knowledge is going to be “digitized” anyway, why not control the process yourself?
Commercial market: Viven and a $2.1 billion valuation
Startup Viven is already creating digital twins of employees on a commercial basis. The company has raised $35 million in funding at a $2.1 billion valuation. Their idea: AI copies of colleagues are used when real employees go on vacation or fall ill.
The market for enterprise digital twin employees is only just taking shape, but the growth rate is impressive. According to analysts' forecasts, the market for AI agents for business processes will reach $50 billion by 2030.
Ethical and legal issues
The project raises several pressing questions.
Consent to “digitization”
Can an AI copy of a colleague be created without their knowledge if the data was collected from shared work chats? In the EU, this may fall under GDPR — personal data includes communication style and behavioral patterns.
The right to knowledge
The developer Colleague Skill described the project as a form of “fighting the loss of institutional memory.” But if the company “owns” the correspondence and the employee owns the experience, who is right?
Quality and accountability
If an AI copy of a colleague gives incorrect advice that causes losses, who is responsible? The project creator, the repository owner, or the company that uploaded the data?
Psychological impact
Working with a “digital twin” of a departed colleague is convenient. But it also normalizes replacing people with algorithms and reduces empathy for those who are being “replaced.”
What employers should do
If your company is considering using Colleague Skill or similar tools:
- Obtain written consent from employees for using their communications to train AI
- Define a data ownership policy — who owns the knowledge recorded in work correspondence
- Do not use this as a substitute for hiring — an AI copy does not replace a live specialist; it speeds up onboarding
- Create a feedback system — employees should have the ability to correct their “digital copy”
What employees should do
- Document your decisions — this is your value, and control over it is your negotiation tool
- Separate public and private channels — not all knowledge should be available for “digitization”
- Consider Yourself Skill — create your own AI copy to control quality
- Know your rights — in some jurisdictions, communication style is personal data
Outlook: where this is heading
The trend toward “digitizing” professional knowledge is irreversible. The question is not whether it will happen, but under what conditions.
Three scenarios are possible:
- “Dark” scenario — companies secretly digitize employees and replace them without warning
- Regulated scenario — legislation defines employees' rights to their digital copies
- Cooperative scenario — employees monetize their own digital twins, creating a subscription-based expertise market
Most likely, reality will end up somewhere in between. But projects like Colleague Skill are already setting the tone for a debate that will shape the labor market for decades to come.
FAQ
What is Colleague Skill?Colleague Skill is an open GitHub project that creates an AI agent simulating a specific employee based on their work correspondence, documents, and communication style. It has gained more than 8,000 stars.
What data is needed to create a digital twin?Messages from work chats (Feishu, DingTalk, Slack), documents, spreadsheets, email, call recordings, and screenshots. The more data there is, the more accurate the copy.
Is it legal to create an AI copy of a colleague without their consent?In most jurisdictions, this is a gray area. In the EU, such data may fall under GDPR. In China, the relevant regulation is still being developed.
Can an AI copy completely replace an employee?At the current level of technology, no. The copy reproduces style and available knowledge, but does not have the ability to learn and adapt in non-standard situations.
What is Anti Distillation Skill?It is a countermeasure tool that obfuscates documents, making them less useful for training AI agents and protecting knowledge from unauthorized copying.
Conclusion
The Colleague Skill project is not just a funny GitHub repository. It is a mirror in which we see the future of the labor market. Digital twins of employees will become a reality — the question is whether they will be a tool for empowerment or a tool for replacement.
If you work in IT, now is the time to think: what would your AI copy do right now?