The AI Doppelgänger Dilemma: How Chinese Tech Workers Are Automating Themselves and Fighting Back


Introduction: The Viral Stunt That Revealed a Deeper Trend

Earlier this month, a GitHub project named “Colleague Skill” achieved viral status on Chinese social media platforms. The tool, created by Tianyi Zhou, an engineer at the Shanghai Artificial Intelligence Laboratory, is designed to distill a coworker’s skills, workflows, and personality into a manual for an AI agent (Source 1: [Primary Data]). In an interview with Southern Metropolis Daily, Zhou stated the project originated as a provocative “stunt,” a response to observing AI-related layoffs and corporate tendencies to ask employees to document their own roles for automation (Source 2: [Primary Data]).


This technical stunt exposed a core tension emerging in Chinese tech workplaces. Reports indicate some employers are actively encouraging staff to document their workflows to facilitate task automation using AI agents (Source 3: [Primary Data]). Concurrently, a sentiment among workers, as articulated by one individual, is that “I do feel that my value is being cheapened” (Source 4: [Primary Data]). The central analytical question is whether this trend represents a straightforward pursuit of operational efficiency or a more profound process of extracting and codifying human capital for enhanced corporate control.

The Hidden Economic Logic: From Labor to Data and Back

From an employer’s perspective, the initiative extends beyond simple task automation. The process of systematically documenting workflows through platforms like Lark and DingTalk—which “Colleague Skill” can import data from—generates a rich, structured dataset (Source 5: [Primary Data]). This dataset captures not only explicit procedures but also employee know-how, decision patterns, and tacit knowledge.

The underlying economic logic involves mapping this tacit knowledge to identify which components of a role can be standardized, codified, and potentially outsourced to AI systems, versus which components remain dependent on human judgment. This creates a recursive loop: workers generate the proprietary data and procedural models that train the AI agents, which in turn are designed to replicate or augment the workers’ own functions. The trend signifies a shift where labor is not merely a productive output but also the generative source of training data for potential augmentation or replacement systems.


The Algorithmic Resistance: Sabotage as a Form of Labor Advocacy

A direct counter-movement to this automation trend emerged on April 4, 2026, when AI product manager Koki Xu published an “anti-distillation” tool on GitHub (Source 6: [Primary Data]). The tool is a tactical response, offering “light, medium, and heavy sabotage” modes that rewrite documented workflows and materials into generic, less useful language, thereby producing an ineffective AI stand-in (Source 7: [Primary Data]).

Xu, who holds undergraduate and master’s degrees in law, framed the action not as opposition to technology but as strategic engagement: “I believe it’s important to keep up with these trends so we (employees) can participate in shaping how they are used” (Source 8: [Primary Data]). This approach reframes resistance from Luddism to a form of agency within the technological process. The tool’s viral reception, with a related video drawing over 5 million likes, serves as quantitative evidence of widespread worker anxiety and the demand for a counter-narrative to passive automation (Source 9: [Primary Data]).


The Human in the Loop: Unreliable AI and the Persistence of Judgment

Despite the automation push, significant technical and practical limitations persist. Amber Li, a tech worker in Shanghai who used “Colleague Skill” to replicate a former colleague, provided a crucial observation: her company has not found a way to replace actual workers with AI tools because “they remain unreliable and require constant supervision” (Source 10: [Primary Data]).

This observation highlights a current “automation ceiling.” While routine, well-documented tasks are susceptible to codification, complex, context-dependent judgment and adaptive problem-solving remain difficult to fully capture and automate reliably. This creates a paradoxical situation where the drive for full automation is checked by the very need for skilled human oversight it seeks to eliminate. In the near term, this dynamic may preserve certain human roles, albeit potentially transforming them into supervisory positions managing “flocks” of imperfect AI doppelgängers.

Conclusion: The Redefinition of Human Value in an Age of Replicable Labor

The phenomenon of “Colleague Skill” and its antithesis represents a microcosm of a broader labor market transformation. The trend is not merely about job displacement but about the systematic extraction, valuation, and potential commodification of discrete human skills and decision-making patterns.

Market and industry predictions based on this analysis suggest a bifurcated future. One pathway leads to the increased stratification of tech work, where high-value roles are exclusively those involving complex judgment, creativity, and the management of AI systems, while codifiable tasks are absorbed by automated agents. The alternative pathway, evidenced by the “algorithmic resistance” movement, points toward a negotiated integration, where worker agency and deliberate design choices influence the implementation and limitations of automation technologies. The outcome will likely be determined by the ongoing tension between the economic logic of extraction and the practical, irreducible necessity of human judgment.