AI 人力资源全景图 · 一次雇佣决策的重新定价 · 2026-07 · 雇主侧HR × AI · Repricing the Hire · Jul 2026 · The Employer’s Side

把人重新定价的机器 The machine that reprices people

招、升、调、裁,本质都是对一个具体的人做一次重新定价;AI 进入 HR,就是把定价从「人工判断」迁到「算法评分+人类兜底」。图眼是一个自反结构:HR 用 AI 执行筛选与裁员优化(当刽子手),而 HR 自身的事务性职能恰是替代重灾区(自己也在断头台上) Hiring, promoting, moving, firing — each is a repricing of one specific person; AI in HR moves that pricing from human judgment to algorithmic scores with human backstops. The map’s eye is reflexive: HR wields AI to optimise screening and layoffs (the executioner) while HR’s own transactional core is ground zero for replacement (already on the scaffold)

本图相信:冲击是结构性的、沿生命周期不均匀分布——事务密集节点(筛选、初评、入职、核算、政策问答)被深度承包;人性密集节点(辞退、文化、信任、担责)被顽固保留。行政助理类岗位净流失 610 万个、行业预测 HR 部门 2030 年前收缩 30–40%(B)。This map believes: the shock is structural and unevenly spread along the lifecycle — transaction-dense nodes (screening, first-pass interviews, onboarding, payroll, policy Q&A) get contracted out to machines; humanity-dense nodes (dismissal, culture, trust, liability) stubbornly stay. Admin roles face a net loss of 6.1M; HR departments are projected to shrink 30–40% by 2030 (B).
本图不相信:「AI 已全面接管 HR」——最新行业调查显示 54% 的机构 HR 完全没用 AI 且 2026 年无计划;72% 从业者认为「人的温度」会阻止 HR 被完全自动化(B)。厂商叙事与落地现实之间,存在明显的预期—现实鸿沟;宣传远跑在落地前面。It does not believe «AI has taken over HR» — the latest industry survey finds 54% of HR functions using no AI at all with no 2026 plans; 72% of practitioners hold that «human warmth» will block full automation (B). Between vendor narrative and adoption yawns an expectation–reality gap; the marketing runs far ahead of the rollout.

本图与求职图是同一战场的两面:求职图站在劳动者侧(简历柠檬化、投递军备竞赛),本图站在雇主与 HR 职能侧——两图对拼才是完整的算法化雇佣战场(→work)。主脊:雇佣生命周期七环节(品牌与职位定义→筛选→面试→入职与薪酬→培训发展→绩效与监控→留任与裁员),②③标红;三条结构带:平台的 Agent 豪赌、公平性诉讼与合规新品类、中介与背调的两端收窄;判断层:三个诚实的杠杆。姊妹分工:管理者的艰难对话→manage,客服话术→support,责任基建(在建)→liability。 This map and the job-seeking map are two faces of one battlefield: that one stands with labour (résumé lemonisation, the application arms race), this one with the employer and the HR function — read together they complete the algorithmic hiring war (→work). The spine: seven links of the employment lifecycle (brand & job definition → screening → interviews → onboarding & pay → L&D → performance & monitoring → retention & layoffs), ② and ③ flagged; three bands: the platforms’ agent bet, the fairness docket and the compliance industry, the middlemen’s two-end squeeze; the judgment: three honest levers. Siblings: hard conversations → manage, support scripts → support, liability infrastructure (in build) → liability.

11 亿次
Workday 自述在诉讼相关期间拒绝的申请份数;2025-05 法院批准年龄歧视集体诉讼有条件认证,覆盖 2020-09 以来所有 40 岁以上被拒者——可能成史上覆盖最广的雇佣歧视诉讼(A)Applications Workday admits rejecting in the litigation window; in May 2025 the court conditionally certified an age-discrimination collective covering every 40+ applicant rejected since Sep 2020 — potentially the widest hiring-bias suit ever (A)
1.1 万份/分钟
一职场平台每分钟收到的申请数(同比 +45%);超半数求职者用 AI 写简历,雇主用 AI 反制——「两个 AI 在视频两端打代理人战争,人成了各自工具的人形接口」(B)Applications per minute on one careers platform (+45% YoY); over half of seekers draft by AI and employers counter with AI — «two AIs fighting a proxy war across the video call, humans reduced to their tools’ interfaces» (B)
30–40%
行业研究预测 2030 年前 HR 部门的人员收缩幅度:多达 130 个专业化 AI 代理承担事务性工作,战略性工作占比从约 30% 升至 75%(B)——刽子手悖论的数字面Projected shrinkage of HR headcount by 2030: up to 130 specialised agents absorb the transactional load while strategic work rises from ~30% to 75% of the job (B) — the executioner’s paradox, quantified
54%
反向对照:2026 年调查中 HR 职能完全没用 AI 且无计划的机构比例;72% 认为「人的温度」会阻止完全自动化(B)——趋势是真的,现状被夸大The counter-gauge: share of organisations whose HR uses no AI at all, with no plans (2026 survey); 72% say «human warmth» will block full automation (B) — the trend is real, the present overstated
⚠️ 口径裁判(先读):① 本领域市场规模数字高度分裂(常差 2–4 倍):HCM(280 亿 vs 587 亿美元)、AI in HR(60 亿 vs 152 亿)、RPO(80 亿 vs 185 亿)、EAP(53 亿 vs 187 亿)、员工监控(6 亿 vs 68 亿)——定义边界不同,不可跨机构比较或加总,本页只给量级;② 厂商自述效率数字(95% 离职预测准确率、初筛自动化 70%、蓝领效率 30 倍)均未经独立复核,按 D 降权;③ 「经理解释敬业度 70% 方差」与「人们离开的是经理」是两个常被混谈的命题,前者原始数据未公开、后者独立复现仅约 12%(vs 薪酬 11%)——本页把争议本身画进图;④ 「HR 岗位被自动化比例」多为广义行政岗数据外推,存在跳跃;⑤ 法规剧烈变动中:一州全面 AI 法生效前被废止、联邦机构撤下 AI 招聘指导、欧盟高风险义务推迟至 2027-12——早于 2026 年中的合规资料都可能过时;⑥ 两处业内流传信息已勘误(一背调公司的运营主体、一 SaaS 的 AI 产品线名称);⑦ 四份深度研究交叉整理(一份约 350 条来源核验为主力);渗透%为编辑估值。 ⚠️ Basis rulings (read first): ① market sizes here split 2–4×: HCM ($28B vs $58.7B), AI-in-HR ($6B vs $15.2B), RPO ($8B vs $18.5B), EAP ($5.3B vs $18.7B), monitoring software ($0.6B vs $6.8B) — different boundaries, never compare or sum across; this page gives magnitudes only; ② vendor-reported efficiency (95% attrition-prediction accuracy, 70% screening automation, 30× blue-collar gains) is unaudited, graded D; ③ «managers explain 70% of engagement variance» and «people quit bosses» are two routinely conflated claims — the former’s raw data was never released, the latter replicates at ~12% (vs pay at 11%); the dispute itself is drawn into the map; ④ «share of HR automated» mostly extrapolates from broad admin-role data — a leap; ⑤ the law is in flux: one state’s flagship AI act was repealed before effect, a federal agency pulled its AI hiring guidance, the EU’s high-risk duties slipped to Dec 2027 — compliance notes older than mid-2026 may be stale; ⑥ two circulating errors are corrected (one background-check firm’s operator; one vendor’s AI product-line name); ⑦ cross-compiled from four deep-research reports (one ~350-source-verified as backbone); penetration %s are editorial.
◆ 中心装置 · 刽子手悖论The central device · the executioner’s paradox
执行替代的部门,先被替代The department that executes replacement gets replaced first
HR 越擅长用 AI 把「筛人、裁人」做得高效,就越是在证明「事务性执行可以被算法承包」——而 HR 自身最大的岗位存量恰恰是事务性执行。刽子手在磨快的那把刀,图纸上量的是自己的脖子。The better HR gets at running screening and layoffs through AI, the more it proves that transactional execution can be contracted to algorithms — and transactional execution is precisely HR’s largest stock of jobs. The blade the executioner sharpens is measured, on the blueprint, against his own neck.
▸ HR 是刽子手HR as executioner
52% 的招聘经理用 AI 生成的生产力数据支撑「含重组与岗位评估在内」的人力规划决策(B);2025 年 AI 在裁员公告中被明确归因的比例快速上升——某月 1.4 万例裁员归因 AI、占当月 31%(B);筛选端一个算法平台就积累了 11 亿次拒绝(A)。 52% of hiring managers use AI-generated productivity data to underpin workforce planning «including restructuring and role evaluation» (B); layoffs explicitly attributed to AI surged through 2025 — 14k in one month, 31% of that month’s total (B); on the screening side a single platform amassed 1.1 billion rejections (A).
▸ HR 在断头台上HR on the scaffold
行政助理与行政秘书净流失 610 万个岗位、列「消失最快岗位」第二(B);行业预测 HR 部门 2030 年前收缩 30–40%、由多达 130 个 AI 代理接管事务(B);IBM AskHR 已接管大部分基础 HR 职能,Gartner 称 AI 可「增强 100%、直接执行高达 50%」的现有 HR 任务(B)。 Administrative assistants and secretaries face a net 6.1M loss, second on the fastest-disappearing list (B); HR departments are projected to shrink 30–40% by 2030 with up to 130 agents running the transactions (B); IBM’s AskHR already handles most basic HR duties, and analysts rate AI able to «augment 100% and directly execute up to 50%» of current HR tasks (B).
◆ 判词Verdict出路只有一条:从「事务执行者」转型为「AI 治理者+艰难对话的人类兜底」——战略性工作占比从 30% 升到 75% 不是恩赐,是幸存者名单的准入条件。同时记住反向对照:54% 机构尚未用 AI,这是趋势,不是现状。One exit exists: from transaction executor to AI governor plus human backstop for hard conversations — strategic work rising from 30% to 75% of the job is not a gift but the entry bar to the survivors’ list. And hold the counter-gauge: 54% of organisations use none of it yet; this is a trend, not the present.
Reading the Map

从这张图带走的五条规律Five patterns to take away

立场声明:本页是批判性、祛魅的行业结构分析,用 A–D 角标区分法院文件/财报、带源研究、单一转述与厂商自述;市场规模只给量级并标注口径分裂;有争议的「HR 公理」(经理 70% 方差)连同其反证一起呈现。本页提供行业结构信息,不构成雇佣、法律或投资建议;涉及具体合规义务请咨询执业律师。核心判断一句话:AI 沿雇佣生命周期不均匀替代——事务密集节点被承包、人性密集节点(辞退、信任、担责)被保留;刽子手悖论之下,HR 的出路是成为 AI 的治理者而非操作员。 Stance: a critical, demystifying structural analysis; A–D badges separate court filings/financials, sourced research, single-relay figures and vendor claims; market sizes come as magnitudes with their splits flagged; the contested «HR axiom» (the manager-70% line) ships with its rebuttal. Structure only — no employment, legal or investment advice; consult counsel for actual compliance duties. One line: AI replaces unevenly along the lifecycle — transaction-dense nodes get contracted out while humanity-dense ones (dismissal, trust, liability) hold; under the executioner’s paradox, HR survives as AI’s governor, not its operator.