AI doubles spark soul-searching in China

Chinese tech workers are teaching AI doubles to replace them.

Tech workers in China are being instructed by their bosses to train AI agents to replace them—and it’s already pushing a wave of soul-searching among even the most enthusiastic early adopters. A spoof project that exploded online this month, Colleague Skill, claims to distill a coworker’s workflows and personality traits into an AI agent. In truth, the initiative started as a stunt, but it laid bare a painful question: what happens when your job is mapped, cloned, and handed off to an algorithm?

The project was created by Tianyi Zhou, an engineer at the Shanghai Artificial Intelligence Laboratory. He described Colleague Skill as something that imports a coworker’s chat history and files from Chinese workplace tools like Lark and DingTalk to generate reusable manuals that describe duties—and even quirks—for an AI agent to emulate. The idea isn’t merely fictional: several Chinese firms are reportedly encouraging employees to document their workflows so AI agents can take over routine tasks and processes. The underlying impulse—turning tacit know-how into codified, repeatable routines—has long been a dream of automation advocates, but the speed and scale of today’s push have transformed it from a thought experiment into something workers actually feel at the desk.

MIT Technology Review notes that the project struck a nerve because it arrives at a familiar junction: the tension between enthusiasm for AI-enabled productivity and the fear of being displaced. When bosses urge staff to automate themselves, workers quickly confront the fragility of the value they bring—how much of it is in hard skills, and how much in lived, day-to-day know-how that can’t be captured in a checklist? The Colleague Skill stunt isn’t just about a clever bot; it’s a litmus test for how firms will measure and monetize tacit knowledge in the AI era.

From a practitioner’s point of view, the episode highlights several tangible realities. First, the automation-promises rely on a fragile bridge between human nuance and machine reproducibility. Replicating decision-making and quirky collaboration styles is far easier in a demo than in the real world where context shifts, conversations drift, and colleagues change roles. Second, data governance and privacy loom large. Accessing chat histories and files from corporate tools to train agents raises questions about consent, ownership, and leakage—especially for sensitive projects or confidential strategies. Third, there’s the classic product-and-process tradeoff: even if the AI can imitate a workflow, keeping it accurate requires continuous human-in-the-loop oversight, versioning, and auditing to prevent drift or misinterpretation.

Analogy time: imagine cloning a colleague’s playbook and assigning an intern to run it. The clone can execute routine passes with speed, but it can’t improvise legitimacy when the room shifts—the human touch, the context, and the occasional gut check are not easily bottled into a manual. The question now isn’t whether AI can imitate work, but whether it can do so reliably at scale without eroding trust or collaboration.

What this means for products shipping this quarter is clear: enterprise AI platforms will need stronger guardrails around automation of human tasks, better tooling to capture provenance and updates to tacit knowledge, and robust privacy controls for data drawn from workplace apps. Vendors should expect increased demand for workflow extraction, explanation of decisions, and audit trails so teams can see when a digital twin deviates and why.

In short, the Colleague Skill moment is less about a single spoof and more about a demand signal. Firms are asking: how do we translate human know-how into AI that can truly augment, not undermine, teams? The answer—at least for now—will require careful design, clear consent, and a steady hand on governance.

Sources

Chinese tech workers are starting to train their AI doubles–and pushing back