Cover Story: How AI Is Mining Worker Data to Reshape the Labor Market
Listen to the full version
When Zhou Tianyi, a 24-year-old researcher at the Shanghai Artificial Intelligence Research Institute, built an open-source AI agent in late March, his aim was straightforward: to preserve his team’s institutional memory.
By importing daily collaboration data from workplace apps such as Feishu and DingTalk, the system — known as colleague.skill — could automate weekly reports, execute workflows and review code. Within days, the project had earned nearly 15,000 stars on GitHub.
Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.
Save an extra $50. Introductory offer for new readers. Subscribe now.
- RELATED
- DIGEST HUB
Digest Hub
Back
- Zhou Tianyi's open-source colleague.skill AI agent preserves team data but misused for "refining colleagues," gaining 15,000 GitHub stars.
- Meta collects employee mouse clicks/keyboard data for AI training amid layoffs; WEF survey: 54% expect AI job replacement.
- China court rules AI efficiency no grounds for 40% salary cut; McKinsey: 57% US working hours automatable.
1. Zhou Tianyi, a 24-year-old researcher at Shanghai Artificial Intelligence Research Institute, created an open-source AI agent called colleague.skill in late March to preserve team knowledge by importing data from Feishu and DingTalk, automating reports, workflows, and code reviews; it quickly gained nearly 15,000 GitHub stars [para. 1][para. 2]. However, users repurposed it for "refining colleagues," causing Zhou discomfort as it veered from knowledge preservation to worker replacement [para. 3][para. 4].
2. This highlights "skill distillation," where firms capture employees' habits, logic, workflows, and decisions to build independent AI digital employees replicating human roles, amid controversy over lost interpersonal chemistry [para. 5][para. 6][para. 7].
3. The divide pits employers compressing costs and AI providers commercializing data against fearful employees excluded from gains [para. 8][para. 9]. Companies collect data like mouse movements to train AI, raising automation fears; Western firms like Meta and Amazon invest heavily amid layoffs, while China faces data ownership issues [para. 10].
4. Evolving from RPA's rule-based bots to contextual AI "brains" [para. 12][para. 13], skill distillation follows four steps: data collection (mouse/keyboard trajectories, comms), cleaning, model training, and deployment shifting humans to oversight [para. 14][para. 15].
5. A SaaS executive noted AI coding advances spurred employee data programs from 2024 testing to 2026 replacements [para. 16]. Meta's April 22, 2026, "Model Capabilities Initiative" captured clicks, inputs, and screens for Muse Spark training without opt-out, amid layoffs [para. 17][para. 18][para. 19].
6. This echoes Grok's aim for scalable digital labor [para. 20]. WEF's January survey of 10,000 executives: 54% see AI replacing jobs, 44.6% boosting profits, only 12.1% raising wages [para. 22][para. 23].
7. China's 2026 job market shows AI role surges with premiums; McKinsey notes sevenfold "AI fluency" demand rise [para. 24][para. 25]. WEF warns of unemployment risks without retraining like cross-generational models [para. 26]. McKinsey: AI could automate 57% US hours, hitting standardized tasks [para. 27].
8. Yet, AI redesigns workflows for human-AI collaboration, retaining 72% skills and unlocking $2.9T US value by 2030 [para. 28][para. 29]. Firms shift workers to oversight; "T-shaped" skills (expertise + empathy) resist replacement [para. 30][para. 31].
9. AI agents disrupt SaaS: Anthropic's February Claude Coworker dropped stocks 40%, shifting to RaaS for outcomes [para. 32][para. 33][para. 34]. Chinese SaaS pivots, but market may shrink two-thirds as AI improves [para. 36][para. 37]. Coding: 90% developers slower than AI; design integrates AI [para. 38][para. 39][para. 40].
10. Even AI sector vulnerable to self-improving systems, reducing engineer roles to monitoring [para. 41][para. 42].
11. China's laws counter: Hangzhou court ruled illegal a 40% pay cut/AI dismissal, awarding 261,000 yuan ($38,300), barring unilateral changes [para. 43][para. 44][para. 45][para. 46]. Experts stress consent for data use, privacy, post-resignation rights [para. 47][para. 48][para. 49][para. 50][para. 51]. Safeguards like local processing urged [para. 52][para. 53][para. 54]. (498 words)
- Meta Platforms Inc.
- Meta Platforms Inc. launched the "Model Capability Initiative" on April 22, 2026, capturing employees' mouse clicks, keyboard inputs, and screen context to train AI agents for its LLM, Muse Spark. Monitoring was deployed pre-announcement without opt-out, coinciding with layoffs, raising fears of AI replacing workers.
- Amazon.com Inc.
- Amazon.com Inc. has invested heavily in "skill distillation" strategies, capturing employee data like mouse movements and workflows to train AI systems for task automation. Amid sector layoffs, Amazon announced plans to cut 16,000 jobs on Jan. 28, 2026, heightening fears of AI-driven displacement.
- 53AI
- 53AI is an open-source AI platform founded by Yang Fangxian. He contrasts traditional RPA (fixed mechanical arms repeating actions) with modern AI systems that have a "brain" for contextual judgments. Yang notes: "Previously, you bought software. Now, you buy a digital employee, which is pure productivity."
- Beijing Langboat Technology Co. Ltd.
- Beijing Langboat Technology Co. Ltd. is a large language model developer. Joint chief executive Li Jingmei states that enterprise AI adoption shifts workers from pure execution to oversight roles, e.g., AI drafts minutes for human review. Workers with standardized tasks are vulnerable, while those with "T-shaped" skills (deep expertise plus communication, empathy, and decision-making) are harder to replace.
- Anthropic
- In February 2026, Anthropic released Claude Cowork, an enterprise legal plug-in, triggering a sharp sell-off in U.S. SaaS stocks, with the sector dropping nearly 40% year-to-date. This reflects the shift toward AI agents disrupting traditional SaaS models.
- Zhejiang Kinding Law Firm
- Zhejiang Kinding Law Firm employs lawyer Shan Qidi, who stated that a company's AI adoption does not qualify as an "objective significant change" under Chinese labor law to justify terminating or altering employment contracts.
- Beijing Chance Bridge Law Firm
- Song Xiaoran, a partner at Beijing Chance Bridge Law Firm, states that using employee data for AI training exceeds normal HR management and requires explicit authorization. Employees retain rights to anonymize data, restrict post-resignation use, and prohibit commercialization of their behavioral traits.
- MOST POPULAR
Sign in with Google
Sign in with Facebook
Sign in with 财新