AI 能力快速变动，使长期人力规划更难：Microsoft 的 Katy George 说，连一年后会有哪些角色都很难看清。对新鲜人与入门职缺尤其不利；根据 Challenger, Gray & Christmas 的资料，美国截至 11 月几乎已宣布裁员约 120 万人，为 2020 年以来同期最高。AI 被列为裁员的主要原因之一，且去年被归因于 AI 的裁员「多出数万人」于破产所致的裁员。Brex 表示仍招募部分初阶员工，但把训练加速，让他们在 3 个月内能胜任更进阶角色。
一些公司把更严格的审核视为避免日后大裁员的方式：Envoy 要求主管提交清楚的商业理由与成功指标备忘录，估计把招募时程拉长约 3 周；Duolingo 则在需要时偏向以承包商替代全职；HubSpot 某些部门选择暂停招募，先「买时间」观察，再于 3 个月后重新评估。同时，快速演进也催生新需求：企业需要内部专家持续评估层出不穷的 AI 新创；Box 的 CEO 表示几乎每周都有可能改写策略的突破，若没人专注追踪，将处于明显劣势。
In the age of fast-evolving AI, companies are raising the bar for adding headcount: leaders say managers must prove a role can’t be done with AI before hiring a human. Cathie Wood framed this as needing a strong case to bring “another human” in; Shopify has required such justification for months. JPMorgan’s CFO described a strong bias against hiring when new needs arise, citing AI-driven productivity “tail winds.” The result isn’t always a hiring freeze, but more friction and fewer approvals.
Rapidly shifting capabilities make workforce planning harder. Microsoft’s Katy George says it’s difficult to predict what roles will exist even a year from now. Entry-level workers face the sharpest squeeze: Challenger, Gray & Christmas reports the US announced almost 1.2 million job cuts through November, the highest for that period since 2020. AI is among the top cited drivers, blamed for tens of thousands more job cuts last year than bankruptcies were. Brex still hires some junior staff but compresses training so they can move into advanced roles within three months.
Companies are tightening processes to avoid future layoffs: Envoy now requires a memo with a clear business case and success metrics, extending hiring timelines by about three weeks. Duolingo sometimes opts for contractors instead of full-time hires. Some HubSpot teams pause hiring to “buy time,” revisiting decisions three months later. At the same time, AI’s pace creates at least one new human need: in-house experts to continuously evaluate AI startups; Box’s CEO warns weekly breakthroughs can reshape strategy, and not tracking them is a steep disadvantage.
