A Rush into the Crayfish Business — A Feast of Bandwagoning? Open‑source tools are turning departed colleagues into AI "skills"

What happened

A GitHub project called colleague.skill has gone viral for one simple, unsettling idea: feed a departed employee’s Feishu (飞书) messages, DingTalk (钉钉) documents, emails and screenshots into an AI, and out comes a “skill” that can answer questions and ostensibly “do the job” of that person. The project’s slogan — invoking “赛博永生” (cyber immortality) — is as much a prank as a provocation. It has been reported that the README even advises contributors to prioritise long authored documents and decision‑making replies, because those artifacts are the richest raw material for distilling professional judgement into a model.

Why it matters in Chinese workplaces and beyond

This playbook is eerily similar to practices large employers already label “knowledge management” or “process optimisation.” In some teams, employees were required to document workflows and decisions — material later used to train AI — and then the teams were downsized. It has been reported that Amazon cut more than 57,000 corporate roles over three years, and AI tooling both amplified productivity and produced spectacular failures (an internal AI assistant reportedly triggered a 13‑hour AWS outage last December). The paradox is stark: companies offload risk onto automated systems while human fallbacks are increasingly scarce. Against a backdrop of global competition for AI talent and chip export controls, firms everywhere are under pressure to squeeze more output from software and fewer people.

The labour pipeline and the backlash

Researchers and industry reports are already flagging a deeper problem: junior roles that used to be “level‑up” spaces—writing code, running models, doing analyses—are the first to be automated. It has been reported that Anthropic’s analysis shows a steep drop in employment among 22–25 year‑olds in high‑AI‑exposure jobs since ChatGPT’s launch, and interviews in Nature found early‑career technical tasks repeatedly singled out. Pushback has emerged in the hacker community — “anti‑distill” techniques, intentionally scrubbed handovers and private backups — but these are stopgaps. If the practices behind colleague.skill proliferate, entire apprenticeship pathways will narrow.

What comes next

The colleague.skill trend is both satire and symptom: an open‑source mirror held up to corporate behaviour. Efficiency gains are real. So are long‑term risks to talent development, institutional memory and accountability. And once relationships, bosses and experts are tokenised into callable skills, who manages the managers — and who bears responsibility when a chain of delegated decisions cascades into failure? The answer will shape whether this is a fleeting bandwagon or a structural shift in how work — and careers — are engineered.