Chinese tech companies are asking employees to document their workflows so AI agents can automate them. The workers have noticed.
MIT Technology Review reported this week on a pattern emerging across Chinese tech employers. Workers are being instructed to capture their workflows, personality traits, communication habits, and decision patterns in structured formats that AI agents can learn from and replicate.
What's Actually Happening
A GitHub project called Colleague Skill, built originally as satire by engineer Tianyi Zhou, went viral by doing exactly this. Connect it to workplace apps Lark and DingTalk, and it extracts how you communicate, what decisions you typically make, and what you know, then packages that information for AI replication.
The tool spread because it was built as a joke. It stuck because it revealed something real: this process is already happening, formally and informally, in companies large enough to think about AI workforce planning. What Zhou built as satire is what many HR and operations teams are trying to build in earnest.
Amber Li, a Shanghai tech worker who tested the tool on her own profile, found it "surprisingly good" at capturing how she writes and thinks. She described the experience as "uncanny and uncomfortable." That combination is worth noting. Not "this is wrong" or "this is clearly broken." Uncanny and uncomfortable means it works, and she knows what that means.
The Worker Response Spectrum
On the other end of the response spectrum: Koki Xu, a Beijing AI product manager who built an "anti-distillation" tool that converts workflow documentation into generic, unusable language before it reaches AI systems. Her video explaining it got five million likes. That's not a niche reaction. Five million likes on a video about resisting AI workflow capture represents a significant slice of the workforce with both the technical capability and the motivation to push back.
The resistance isn't about AI in general. It's about the specific experience of being reduced to a replicable module. The value workers have historically assigned themselves isn't fully captured by documented workflows. The judgment calls, the relationship knowledge, the contextual adaptation that doesn't fit into a structured format, that's what workers instinctively protect. They're not wrong that it's real. The question is whether the organizations asking for documentation understand this or are ignoring it.
What This Means for Organizations Building AI Automation
At difrnt.ai, we build AI workflow automation for clients. The technical question of whether AI can replicate a given workflow is usually answerable within a week. The harder question is almost always organizational.
Workers who feel ownership over their workflows either actively support the AI transition or actively undermine it. The difference between an implementation that succeeds at rollout and one that fails isn't usually the technology. It's the trust that was or wasn't built in the design phase.
Companies that approach this as "give us your workflows so we can automate you" get the Koki Xu response. Companies that approach it as "help us build the AI infrastructure that handles the repetitive parts so you can focus on the judgment-intensive work" get something very different. Both approaches produce documentation. Only one produces cooperation.
The technology works either way. The organization only works in the second case.
The Chinese tech worker story isn't primarily about China. It's about what happens when you implement AI automation without resolving the human question first. That story will play out in every market over the next three years.
FAQ
What is AI workflow documentation and why are companies pursuing it?
It's the process of capturing how specific employees do their jobs in structured formats that AI systems can learn to replicate. Companies pursue this to automate routine work and reduce headcount for standardizable roles. It also creates institutional knowledge that persists beyond individual employment, which has independent value even without automation goals.
How should employees think about AI workflow documentation requests?
With both clarity and strategy. Contributing to workflow documentation changes your role over time. The question worth asking before participating: what work will remain for me after automation, and is that work more or less valuable than what I do now? The answer should inform how you engage with the process, not whether you engage with it.
What should business leaders know before rolling out AI automation to existing teams?
The technical implementation is usually the easy part. The organizational dynamic is harder. Workers who feel the process is transparent and focused on productivity improvement are far more cooperative than workers who feel they're being replaced without acknowledgment. Communication about what happens after automation matters as much as the automation itself.