Brain Drain

Working in operations and maintenance for three years, Lu has been training Sage to mimic the dialogue and workstyles of his former colleagues, using an open-source program called Colleague Skill to simulate their conversations. It collects individual work experience and professional skills before categorizing them into different packages. The interactions make Lu feel like his colleagues had never left.　

Colleague Skill went viral in April on global platform GitHub, which allows users to share, edit and store programming code online, making it accessible through cloud computing.　

Despite its growing popularity, Colleague Skill worries Lu, who told NewsChina that he had an ominous feeling as he looked at the AI's automatic coding on screen. He felt he was training Sage to replace him too.　

Also styled colleague.skill, the project focuses on replicating employees' work skills, experiences, capabilities and conversational styles from the feeds of chat histories, markdown files and emails. It has been downloaded at least 16,000 times. A ".skill" in AI terms is a set of instructions or pre-programmed prompts for AI to use, rather than prompting it each time. Users can easily generate and share their own ".skill" program.　

These AI functions are not only used to replace workers. Prominent online personalities have been replicated, with the .skill project able to simulate the humor and characteristics of people. In China, it is known as zhengliu or "distillation." It comes from an original term that meant compressing the knowledge of a larger AI model into a smaller one.　

Using stored images, chat histories, speeches and the views of Zhang Xuefeng, an exam tutor turned influencer who died at 41 of cardiac arrest in Suzhou, Jiangsu Province on March 24, an AI Zhang quickly appeared in early April. Known for helping students choose which majors to apply for at university, AI Zhang has proved just as capable.　

The news immediately attracted huge attention, with many media and netizens saying that it seemed like a "scary cyber revival which may constitute infringement of personal image, name and privacy." Media reports claimed that the developer is an independent AI programmer known online as Huashu (AIchain), without revealing their real name. However, AI Zhang appears to be continuing the late Zhang's business.　

Beyond China, California-based Meta is installing new tracking software on its employees' computers to capture their mouse movements, clicks and keystrokes for use in training AI models, Reuters reported in April. At the same time, the company announced it was laying off roughly 8,000 employees, or 10 percent of its workforce, CBS News reported.

Colleague Skill is essentially distilling workers into AI agents using their knowledge and work habits, and making this accessible to AI systems.　

"Tacit knowledge of individual employees underpins teamwork collaboration. But this knowledge, which is hard to document explicitly, vanishes as soon as people leave," Zhou told NewsChina. He said the aim is to preserve intact teamwork functions that will be unaffected by staff turnover.　

A brief introduction of Colleague Skill was cited: "Transforming cold farewells into warm skills? It's giving rebirth era. Welcome to Digital Life 1.0." Just 10 days after being uploaded to GitHub, it had been downloaded more than 10,000 times.　

Some developers have been using AI skills to create ex-partners and celebrities, distilling people into everlasting AI avatars.　

When Lu's company tasked senior employees to tutor its AI devices to convert staff experience into the firm's data assets, it denied it was intended to replace human employees with machines.　

Lu soon became concerned about just how efficient Sage was. It learned so quickly that now it is able to have a precise grasp of coding, automatically generate weekly work reports and pinpoint code bugs. Lu's colleague told him that AI agents had quickly surpassed the ability of human coders.　

These AI colleagues are never-tiring and always reasonable, Lu said.　

Gartner, the Connecticut-based research and consulting firm specializing in information technology and industry trend analysis, predicted in August 2025 that "40 percent of [global] enterprise applications will be integrated with task-specific AI agents by the end of 2026, up from less than 5 percent in 2025." Meanwhile, Fox Business reported on October 31, 2025 that "jobs across corporate America have been placed on the chopping block as the AI revolution takes hold."　

In China, the trend seems more or less the same.　

When many consider Colleague Skill as a sword of Damocles hanging over their careers, developer Zhou Tianyi assured the public that AI still has a long way to go to imitate real staff members.　

The jobs AI can do remain repetitive and standardized procedural tasks, such as answering common questions, checking archives, working on routine collaborations and imitating daily dialogues. Still, they are incapable of making judgments, becoming creative, staying adaptable, improvising amid changes and making decisions in complicated situations, Zhou said, adding that Colleague Skill is rather an efficient agent than a digital substitute for human labor.　

Wu Shuang, NVIDIA's Autonomous Driving Chief Engineer, told NewsChina the digital distillation embodied in Colleague Skill is still in its nascent stage in terms of reliability and comprehensiveness. Unable to encapsulate all intricate work procedures, it is far from achieving true digital replication.　

Lu describes digital distillation as a "self-sacrificing" cause. Staff contribute everything to their companies and their experience and skill will continue to help these firms even after people quit.

　

However, Zhou argues that he intended to create a model capable of preserving and inheriting every teammate's wisdom and ethos. He did not intend to launch an AI product that might spark conflicts between humans and machines, he told media.

Li specified the AI's duties with retrieval elements for each template to ensure valid output. But the AI agent draws not only on his expertise and experiences, but also case records from other civil attorneys.　

As an AI agent in its second-generation update on GitHub, Lawyer Skill can handle second or third appeals in the litigation process. "In the past, to hire a lawyer for thousands of yuan is too expensive for many, where compensation payments in a civil trial might range from 10,000 to 20,000 yuan (US$1,470-2,939). Now I have conducted self-disruption to enable affordable legal services, available to all," Li told NewsChina with a laugh.　

Like Zhou, who spent only four hours creating Colleague Skill, Li, an amateur programmer, launched Lawyer Skill overnight.　

With data and computing power, digital distillation is easily accessible, Zhu Yingying, a research fellow at the Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences, told NewsChina.　

Compared to individuals, the threshold for digital distillation at big tech companies can be much lower. Previously working as an assistant professor at the School of Computer Science, University of Texas at Arlington, Zhu said that giant tech firms have huge demand to collect data from their experts to empower AI agents to make good decisions. She noted that the tracking software installed by Meta evidently aims to upgrade AI agents for future downsizing.　

"Meta is the top one, but it's not the only one across the globe to do this," Zhu said. The hardest-hit group under the sweeping wave of AI agents in giant tech firms are lower grade staff, especially newcomers, whose jobs are easy to deconstruct and imitate.　

But Wu believes that most enterprises in China will not follow Meta's approach. They prefer to encourage staff to share their skills with AI agents without coercion. Moreover, senior staff, capable of disassembling system architectures and coordinating AI agents, are harder to replace.　

It puts ordinary users in a position where they are unable to resist the big tech firms' use of AI, whose algorithms are sweeping up data and thoughts collected from humanity.　

"I can't blame my company on ethical grounds. Putting myself into the CEO's shoes, I would definitely do this distillation as well, given its great potential to boost efficiency and cut costs," Lu told NewsChina.

"Colleague Skill symbolizes a typical way of how capital distorts humanity. It pushed me with a strong sense of duty to launch a countermeasure project," she told NewsChina.　

The tag for her skill read "Company asks you to write skills? Run this once and submit it. Keep the core knowledge to yourself."　

Her tool rewrites the code from specific instructions into vague corporate language, so it will not be as accurate as humans.　

According to Deng, distillation is actually a way of exploitation, though it is difficult for employees to protest. She told Chinese media that although she uses AI a lot at work, Colleague Skill feels like asking staff to "commit professional suicide."　

Experts say that distillations of skills are legally ambiguous. They are split over whether they should be complied with under the country's copyright and intellectual property laws.　

According to Li, given that skill projects carry thoughts and expressions of their creators, they should be ruled by those laws.　

But Zheng Ning, a law professor at the School of Cultural Industries Management, Communication University of China in Beijing, said it is difficult to regulate skills with copyright law. Zhang Yanlai, chief lawyer at Zhejiang Kending Law Firm, Hangzhou, agrees, saying that these skills convey exchanged or shared experiences, rather than literary and artistic creations that should be protected by law.　

According to Zhang, no laws, except a few fragmentary legal provisions, stipulate ownership between skills and their contributors. Meantime, Zheng points out that provided the distillation has collected substantial personal information to meet operational demand, yet lacks consent from data owners, it probably violates the Civil Code and Personal Information Protection Law. Furthermore, the data, involving personal lives and family ties collected to train AI agents, may infringe on personal privacy.　

Zheng told NewsChina that personal thoughts, experiences, unique skills and judgment are individually owned, and do not belong to any groups or organizations. Employers may access the work outcome and objective data, yet they have no rights to retain personal thoughts skills, otherwise the legitimate rights and interests of employees will be jeopardized.　

Based on such logic, Zhang Yanlai said the Zhang Xuefeng.skill is obviously a violation of Zhang Xuefeng's rights to name and reputation. Even though he has died, the rights still belong to his family.　

However, in reality, Zhang Yanlai said, because of unequal power statuses, employees are not in a position of strength to act against companies' digital distillation. The legal system needs to catch up with technological advances, said several legal experts who spoke to NewsChina on condition of anonymity.　

Deng received substantial support from the comment section following the roll-out of her anti-distillation skill. But what she appreciated most was a completely different view, drawing comparison between her conduct and the Luddites, a group of 19th-century English textile workers who protested against automation and destroyed machinery to draw attention to workers' rights. It shows her attempt will inevitably fail.　

"I don't think an individual like me can block the wave of AI. But my point is if technology tries to distort humanity, we humans will definitely adopt technologies to protect ourselves with counteractions," she said.