在达拉斯以北的普莱诺一家必胜客，AI用于语音下单、后厨优先排序、合成的评论摘要，以及一个告诉新员工配料用量的聊天机器人，表明职场生成式AI的采用往往是渐进而非颠覆性的。《经济学人》仅限订户的播客“Boss Class”新一季于1月29日发布，用这些案例来追问企业看到了哪些收益、进展被什么拖慢、以及哪些管理选择有帮助。
影响分布不均：Indeed发现，典型软件开发招聘中提到的技能中有很大多数可能被AI深刻影响，而典型护理岗位中列出的多数技能目前仍在其能力之外。尽管OpenAI在12月报告称ChatGPT Enterprise在用户使用的每一天平均节省约40–60分钟，并且其9月的GDPval评估称前沿模型在许多现实任务上接近与行业专家持平，但一项涵盖四国的高管调查发现，近四分之三的企业以某种方式使用AI，然而86%表示过去三年劳动生产率没有受到影响。
障碍是行为、技术与组织层面的，而通用技术往往需要多年才会显现广泛收益；Bret Taylor认为这感觉就像Excel上周末才被发明，并预计在五年内会出现更成熟的“智能体”供应商生态。具有狭窄范围与护栏的部署例子包括带有人类兜底的监督模型、Garfield为低于£10,000（$13,800）的小额索赔提供的受监管AI法律服务，以及强生从大量试验转向聚焦于产生了85%价值的15%应用，同时基于结果的定价只在智能体真正解决问题时收费，而且调查受访者仍预计未来三年会出现岗位减少与生产率提升。
At a Pizza Hut in Plano, north of Dallas, AI is used for voice ordering, kitchen prioritisation, synthesized review summaries, and a chatbot that tells new staff ingredient quantities, illustrating that workplace generative-AI adoption is often incremental rather than transformative. The new season of The Economist’s subscriber-only “Boss Class” podcast, released on January 29, uses such cases to ask what benefits firms are seeing, what slows progress, and what management choices help.
Impact is uneven: Indeed finds a large majority of skills in typical software-development postings could be profoundly affected by AI, while most skills in typical nursing jobs remain beyond it. Even as OpenAI reported in December that ChatGPT Enterprise saved users about 40–60 minutes on each day they used it and its September GDPval evaluation said frontier models approach parity with industry experts on many real-world tasks, a four-country executive survey finds nearly three-quarters of businesses use AI yet 86% report no labour-productivity impact over the past three years.
Barriers are behavioural, technical, and organisational, and general-purpose technologies can take years to show broad gains; Bret Taylor argues it feels like Excel was invented last weekend and expects a more mature “agents” vendor landscape within five years. Examples of narrow, guardrailed deployment include supervisor models with human fallback, Garfield’s regulated AI legal service for small claims under £10,000 ($13,800), and Johnson & Johnson’s shift from many experiments to focusing on the 15% of applications that generated 85% of value, while outcomes-based pricing charges only when an agent resolves a case and survey respondents still expect job losses and productivity gains in the next three years.
Source: How to avoid common AI pitfalls in the workplace
Subtitle: Advice from our latest season of “Boss Class”
Dateline: 1月 29, 2026 06:22 上午 | Plano and San Francisco