AI 零工平台全景图 · AI 劳动形态的预演场 · 2026-07 · 制度与平台层Gig Platforms × AI · The Rehearsal Stage · Jul 2026 · Institutions & Platforms
人类零工的今天,AI 劳工的明天
The gig worker’s today, the AI labourer’s tomorrow
在 AI 把 work 拆成 task 之前,平台经济已经先把 job 拆成 gig——按任务计价、算法分配、评分信用、无长期雇主。这套制度模板不是被 AI 劳动市场「参考」,而是被直接继承为操作系统;人类零工的今天,大概率是 AI 劳工的明天——而 2025–2026,两者已开始在同一个平台上同场竞价
Before AI split work into tasks, platforms had already split jobs into gigs — per-task pricing, algorithmic dispatch, reputation scores, no standing employer. That institutional template is not being «referenced» by the AI labour market; it is inherited outright as its operating system. The human gig worker’s today is the AI labourer’s tomorrow — and by 2025–26 the two have begun bidding on the same platforms
本图相信:替代已在最灵敏的探测器上实时发生——技能零工平台的接单量:写作 −33%、翻译 −19%、客服 −16%;一中国技能众包老平台被生成式 AI「降维打击」三年半累亏超 10 亿元;AI agent 已在真实平台发帖雇人(平台 CEO 一手证实,B/A)。This map believes: replacement is already live on the most sensitive detector — skilled-gig order volumes: writing −33%, translation −19%, support −16%; Zhubajie, the veteran Chinese skill marketplace blindsided by generative AI, lost ¥1B+ over 3.5 years; AI agents already post jobs to hire humans on a real platform (CEO-confirmed, first-hand, B/A).
本图不相信:「AI 正在消灭零工」——丹麦全域行政数据对雇员收入与工时给出「精确零效应」;agent 单干完成率极低、配一名人类完成率提升 70%+;当下的真相是价值曲线两极化+人机配对,不是纯竞价替代。平台即时下滑与宏观零效应之间的张力,恰恰说明零工市场是最前沿的探测器。It does not believe «AI is abolishing gig work» — Danish full-registry data finds a «precise zero effect» on employee earnings and hours; agents alone barely complete tasks while pairing with one human lifts completion 70%+; the present truth is a polarising value curve plus human-machine pairing, not pure substitution. The tension between platform-level decline and macro zero is exactly what makes the gig market the frontier detector.
最小单元:一单被拆散的工作——与首页的 task 单元直接对接;司机骑手的劳动现场见交通图与体力图,本图专做制度与平台切片。主脊:一单的生命周期七环节(发布拆单→算法定价→接单执行→评分信用→结算抽成→觉醒监管→人机同台),②④⑦标红;三条结构带:三大法域立法战、抽成爬坡与净时薪的诚实层、AI 合流的实时行情表;判断层:多平台+直客 ＞ 单平台冲单;信用可携带权=最高杠杆的制度变量。姊妹:就业→work,任务化写作→novel,客服→support。
The atom: one job, disassembled into an order — plugging straight into the master map’s task unit; for drivers’ and riders’ shop floors see transport and labor; this map cuts the institutional and platform layer. The spine: one order’s lifecycle in seven links (posting → algorithmic pricing → execution → reputation → settlement → awakening & regulation → humans-and-agents), ②④⑦ flagged; three bands: the three-jurisdiction legislative war, the honesty layer of take-rate climbs and net wages, the real-time AI ticker; the judgment: multi-platform plus direct clients ＞ single-platform grinding; reputation portability is the highest-leverage institutional variable. Siblings: employment → work, taskified writing → novel, support → support.
≈2 亿人
中国灵活就业的官方锚点口径(其中「新就业形态劳动者」约 8400 万,A);更高的 2.4 亿/3.2 亿口径宽窄不一慎用——就业缓冲垫本身正在被自动化抽薄China’s official anchor for flexible employment (with ~84M «new-form workers», A); wider 240M/320M counts vary by definition — and the employment cushion itself is being thinned by automation
15.4→18.0%
一技能零工平台市场抽成率一年的爬坡(2023→2024,净利 0.47→2.16 亿美元);同业最高约 27.6%;网约车鼻祖从 15–20% 爬到平均 40%、单笔可超 50%(A/B)——成熟即涨价One skill platform’s take rate climbing in a single year (2023→2024; net income $47M→$216M); the peer peaks near 27.6%; the ride-hail pioneer climbed from 15–20% to a 40% average with 50%+ on some trips (A/B) — maturity means a raise
−33%
生成式 AI 后写作类接单量的下滑(翻译 −19%、客服 −16%;视频剪辑 +39%、AI 聊天机器人开发 +2000%)——技能零工的价格曲线,就是 AI 替代的实时行情表(B)The fall in writing orders after generative AI (translation −19%, support −16%; video editing +39%, chatbot building +2,000%) — the skilled-gig price curve is the real-time ticker of AI substitution (B)
40%
Upwork 2025Q4 完成任务中含「显著 AI agent 协助」的比例(2023 年 <5%);agent 单干完成率极低、配一名人类 +70%——人机同台的当下形态是配对,不是替代(B/D 平台自报)Share of Upwork’s Q4-2025 completed tasks with «significant AI-agent assistance» (under 5% in 2023); agents alone barely finish while one human partner lifts completion 70%+ — today’s co-stage is pairing, not replacement (B/D self-reported)
⚠️ 口径裁判(先读):① 中国灵活就业总量三口径(2 亿/2.4 亿/3.2 亿)宽窄不一,本页取官方锚点 2 亿;② 数据标注市场规模窄口径约 77 亿元 vs 含采集宽口径 130–180 亿元,勿混用;③ 一网约车平台抽成两口径(2020 披露 20.9% vs 2024 称 14%,补贴计入方式不同);④ 「时薪 $3.37、74% 低于最低工资」为方法论有争议的初版测算,后续修订更高,引用需注明;⑤ 愿景/营销/仿真(「全面自动化经济」「Stop Hiring Humans」、agent 竞价模型)与已发生事实(按 ACU/按 action 计价、agent 发帖雇人)在本页严格分开;⑥ 平台自报数据(抽成、agent 完成率)有自利披露倾向,标 D 降权;⑦ 三处常见误传已勘误:某州公投合宪判决在 7 月非 8 月;中国职伤试点首批第 7 省是四川非山东;首批 7 家平台不含头部网约车平台(其为 2025 扩围新增);⑧ 四份深度研究交叉整理(一份 130+ 次检索核验为主力);渗透%为编辑估值。
⚠️ Basis rulings (read first): ① China’s flexible-employment total spans three definitions (200M/240M/320M) — this page anchors on the official 200M; ② data-labelling market sizes split ~¥7.7B narrow vs ¥13–18B broad — never mix; ③ one ride-hail platform’s take rate carries two bases (20.9% disclosed 2020 vs 14% claimed 2024, differing subsidy treatment); ④ the «$3.37/hr, 74% under minimum wage» figure is a methodologically contested first estimate, later revised upward; ⑤ visions and marketing («fully automated economy», «Stop Hiring Humans», agent-bidding simulations) are strictly separated from realised facts (per-ACU / per-action pricing, agents posting jobs); ⑥ platform self-reports (take rates, agent completion) lean self-serving — graded D; ⑦ three common errors corrected: the ballot-measure ruling came in July not August; the seventh pilot province was Sichuan not Shandong; the first seven pilot platforms exclude the top ride-hail firm (added in the 2025 expansion); ⑧ cross-compiled from four deep-research reports (one 130+-call-verified as backbone); penetration %s are editorial.